Methods and systems for prediction of immune checkpoint inhibition therapeutic response and toxicity from an immune signature

Abstract

Methods and systems for sorting a candidate immune checkpoint inhibition (ICI) recipient as a responder or non-responder, by inferring cell classifications generated from immunophenotyping and inputting the inferred cell classification into a machine learning model trained to predict an outcome with training data comprising at least: (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints, (b) one or more indications of the outcome for the plurality of ICI recipients at respective timepoints; outputting an outcomes candidate ICI recipient; and sorting the candidate ICI recipient as a responder or a non-responder based at least on the predicted probability of outcome. Also provided herein are methods and systems for monitoring response to ICI therapy.

Description

Attorney Docket No.: 279362000640METHODS AND SYSTEMS FOR PREDICTION OF IMMUNE CHECKPOINT INHIBITION THERAPEUTIC RESPONSE AND TOXICITY FROM AN IMMUNE SIGNATURECROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the priority benefit of U.S. Provisional Patent Application number 63 / 761,725, filed February 21, 2025, and U.S. Provisional Patent Application number 63 / 883,390, filed September 17, 2025, the entire contents of each of which are incorporated herein by reference.FIELD

[0002] This disclosure relates to predicting a response to an immune checkpoint inhibition (ICI) therapeutic (ICI therapy) in an individual using an immune signature generated from a blood sample before or after the individual receives the ICI therapeutic. The disclosure further relates to predicting and preventing adverse events following administration of an ICI therapeutic to an individual using the immune signature.BACKGROUND

[0003] Immune checkpoint inhibition (ICI; alternatively immune checkpoint blockade (ICB)) therapies have revolutionized cancer treatment by driving remarkable therapeutic responses (including remission) in a subset of cancer patients. The therapies may be especially effective for treating non- small cell lung cancer, melanoma and renal cell carcinoma.

[0004] ICI therapies are antibody drugs that target immune checkpoint molecules. At the time of filing the Food and Drug Administration (FDA) have approved multiple ICI therapies, including therapies targeting CTLA4, PD1 and PDL1. ICI therapies for new targets such as LAG3, TIM3, and TIGIT are in pre-clinical and clinical validation studies.

[0005] Although the therapies are successful for some cancer patients, less than 20% of patients respond to ICI therapy. The variation in response is only partially explained by cancer type. Even within cancer types that are known to be susceptible to ICI therapy, patient responses vary dramatically. Some patients also acquire resistance to ICI therapy. If a patient does not respond to ICI therapy, administration of the ICI therapy delays administration of alternative treatment that could treat the cancer, extend a patient’s life or quality of life. Further, ICI treatment can be more expensive than other treatments and thus administration of ICI1MF-367064958Attorney Docket No.: 279362000640therapies to individuals who will not response places a substantial financial burden on healthcare systems, insurance providers and patients.

[0006] Even if a patient is responsive to ICI therapy, they may be at risk for immune-related adverse events (irAEs) ranging from mild to severe (permanent organ damage, death). The emergence of an irAE in a patient often necessitates termination of the treatment and / or treatment with immunosuppressants.

[0007] There remains a need for advanced but repeatable and clinically practical methods for predicting and monitoring response to ICI therapies and irAE development in cancer patients. Sharma et al. Immune Checkpoint therapy- current perspectives and future directions. 186(8) Cell (2023). Such methods would aid in stratifying patients for ICI therapy who are most likely to benefit from it, aid doctors in making rapid clinical decisions during ICI therapy to maximize patient benefit, and pre-empt and / or mitigate the onset and severity of irAEs. The methods and systems described herein can be used to achieve these goals and more.SUMMARY

[0008] Provided herein are methods and systems for sorting a candidate ICI recipient as a responder or non-responder, monitoring a response to, and predicting therapeutic outcomes following administration of an ICI therapeutic to a patient in need thereof. The methods and systems rely on machine learning models trained to predict response to and clinical outcomes following administration of an ICI therapeutic to a patient in need thereof from high resolution immune profiling of a patient’s blood before, during and / or after administration of the ICI therapeutic. The methods and systems comprise machine learning models trained with samples collected from individuals who have received an ICI therapeutic at multiple timepoints, before, during, and / or after administration of the ICI therapeutic. Also provided herein are methods of training a machine learning models for use in the methods and systems.

[0009] Provided herein are methods for sorting a candidate immune checkpoint inhibition (ICI) recipient as a responder or non-responder, comprising: fluorescently labeling cells contained within a sample from the candidate ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer; inferring cell classifications based on the fluorescent intensity data; providing at least a subset of the cell classifications as input to an adverse event machine learning model, wherein the adverse event machine learning model has been trained to predict a probability of2MF-367064958Attorney Docket No.: 279362000640an immune-related adverse event (irAE) with training data comprising at least: (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints, (b) one or more indications of the irAE for the plurality of ICI recipients at respective timepoints; outputting from the adverse event machine learning model a predicted probability of the irAE for the candidate ICI recipient; and sorting the candidate ICI recipient as a responder or a non-responder based at least on the predicted probability of the irAE.

[0010] In some aspects, the one or more indication of the irAE comprise one or more severity indications of the irAE. In some aspects, the severity indication of the one or more severity indications of the irAE is based on Common Terminology Criteria for Adverse Events (CTCAE). In some aspects, the severity indication of the irAE is severe if the CTCAE greater than or equal to 3 and the severity indication of the irAE is non- severe if the CTCAE is less than 3. In some aspects, the respective timepoints for the one or more indications of the irAE are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0011] In some aspects, the sample is a first sample from a first timepoint. In some aspects, the method comprises inferring cells classification for a second sample at a second timepoint by: fluorescently labeling cells contained within a second sample from the candidate ICI recipient, by contacting at least an aliquot of the second sample with at least one immunophenotyping fluorescent labeling panel, generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer; inferring cell classifications based on the fluorescent intensity data. In some aspects, the methods comprise providing at least a subset of the cell classifications from the second sample as input to the adverse event machine learning model.

[0012] In some aspects, the adverse event machine learning model has been trained to average over the hidden representations of the first sample and the second sample or applies an attention mechanism over the hidden representations of the first sample and the second sample to predict a trajectory-based prediction.

[0013] In some aspects, the candidate ICI recipient is sorted as a responder if the predicted probability of the irAE is less than a predetermined adverse event threshold and sorted as a non-responder if the predicted probability of the irAE is greater than the predetermined adverse event threshold. In some aspects, the predetermined adverse event threshold is about 25%, about 50%, about 75%, about 95%, or about 99%.3MF-367064958Attorney Docket No.: 279362000640

[0014] In some aspects, the adverse event machine learning model is a time sensitive machine learning model.

[0015] Also provided herein are methods for sorting a candidate immune checkpoint inhibition (ICI) recipient as a responder or non-responder, comprising: fluorescently labeling cells contained within a sample from the candidate ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer; inferring cell classifications based on the fluorescent intensity data providing at least a subset of the cell classifications as input to a response machine learning model, wherein the response machine learning model has been trained to predict the probability of a positive outcome with training data comprising at least: (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints (b) one or more indications of the positive outcome for the plurality of ICI recipients at respective timepoints; outputting from the response machine learning model, a predicted probability of the positive outcome for the candidate ICI recipient; and sorting the candidate ICI recipient as a responder or a non-responder based at least on the predicted probability of the of positive outcome.

[0016] In some aspects, the positive outcome relates to cancer stage, cancer progression, cancer severity, cancer remission, health-related quality of life, symptom management, a RECIST 1.1 response criteria or survival. In some aspects, the respective timepoints for the one or more indications of the positive outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0017] In some aspects, sample is a first sample from a first timepoint. In some aspects, the method comprises inferring cells classification for a second sample at a second timepoint by: fluorescently labeling cells contained within a second sample from the candidate ICI recipient, by contacting at least an aliquot of the second sample with at least one immunophenotyping fluorescent labeling panel, generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer; inferring cell classifications based on the fluorescent intensity data. In some aspects, the methods comprise providing at least a subset of the cell classifications from the second sample as input to the response machine learning model. In some aspects, the response machine learning model has been trained to average over the hidden representations of the first sample and the second4MF-367064958Attorney Docket No.: 279362000640sample or applies an attention mechanism over the hidden representations of the first sample and the second sample to predict a trajectory-based prediction.

[0018] In some aspects, the candidate ICI recipient is sorted as a responder if the predicted probability of the positive outcome is greater than a predetermined positive outcome threshold and sorted as a non-responder if the predicted probability of the positive outcome is less than the predetermined positive outcome threshold. In some aspects, the predetermined positive outcome threshold is about 25%, about 50%, about 75%, about 95%, or about 99%.

[0019] In some aspects, the response machine learning model is a time sensitive machine learning model.

[0020] Also provided herein are methods for sorting a candidate immune checkpoint inhibition (ICI) recipient as a responder or non-responder, comprising: fluorescently labeling cells contained within a sample from the candidate ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel, generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer; inferring cell classifications based on the fluorescent intensity data; providing at least a subset of the cell classifications as input to an adverse event machine learning model and a response machine learning model; wherein the adverse event machine learning model has been trained to predict the probability of an immune-related adverse event (irAE) with a first set of training data comprising at least: (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints, (b) one or more indications of the irAE for the plurality of ICI recipients at respective timepoints; wherein the response machine learning model has been trained to predict the probability of a positive outcome with a second set of training data comprising at least: (a) the plurality of cell classifications, (b) one or more indications of the positive outcome for the plurality of ICI recipients at respective timepoints; outputting from the response machine learning model, a predicted probability of the positive outcome and from the adverse event machine learning model, a predicted probability of the irAE for the candidate ICI recipient; and sorting the candidate ICI recipient as a responder or a non-responder based at least on the predicted probability of the of positive outcome and the predicted probability of the irAE.

[0021] In some aspects, sorting the candidate ICI recipient comprising combining the predicted probability of the positive outcome and the predicted probability of the irAE using voting, stacking, averaging, or Logit blending methods. In some aspects, the adverse event machine learning model and the response machine learning model are time sensitive machine learning 5MF-367064958Attorney Docket No.: 279362000640models. In some aspects, the respective timepoints for the one or more indications of the irAE are the different respective timepoints from the plurality of cell classifications or are alternative timepoints. In some aspects, the respective timepoints for the one or more indications of the positive outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0022] Also provided herein are methods for sorting a candidate immune checkpoint inhibition (ICI) recipient as a responder or non-responder, comprising: fluorescently labeling cells contained within a sample from the candidate ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel, generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer; inferring cell classifications based on the fluorescent intensity data; providing at least a subset of the cell classifications as input to a multioutput machine learning model, wherein the multioutput machine learning model has been trained to predict the probability of a positive outcome and / or a probability of an immune-related adverse event (irAE) with training data comprising at least: (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints (b) one or more indications of the positive outcome for the plurality of ICI recipients at respective timepoints; (c) one or more indications of the irAE for the plurality of ICI recipients at respective timepoints; outputting from the multioutput machine learning model, a predicted probability of the positive outcome and / or a predicted probability of the irAE for the candidate ICI recipient; and sorting the candidate ICI recipient as a responder or a non-responder based at least on the predicted probability of the of positive outcome and / or the predicted probability of the irAE.

[0023] In some aspects, the one or more indication of the irAE comprise one or more severity indications of the irAE. In some aspects, the severity indication of the one or more severity indications of the irAE is based on Common Terminology Criteria for Adverse Events (CTCAE). In some aspects, the severity indication of the irAE is severe if the CTCAE greater than or equal to 3 and the severity indication of the irAE is non- severe if the CTCAE is less than 3. In some aspects, the positive outcome relates to cancer stage, cancer progression, cancer severity, cancer remission, health-related quality of life, symptom management, a RECIST 1.1 response criteria or survival. In some aspects, the respective timepoints for the one or more indications of the irAE are the different respective timepoints from the plurality of cell classifications or are alternative timepoints. In some aspects, the respective timepoints for the6MF-367064958Attorney Docket No.: 279362000640one or more indications of the positive outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0024] In some aspects, the sample is a first sample from a first timepoint. In some aspects, the method comprises inferring cells classification for a second sample at a second timepoint by: fluorescently labeling cells contained within a second sample from the candidate ICI recipient, by contacting at least an aliquot of the second sample with at least one immunophenotyping fluorescent labeling panel, generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer; inferring cell classifications based on the fluorescent intensity data. In some aspects, the methods comprise providing at least a subset of the cell classifications from the second sample as input to the multioutput machine learning model, the adverse event machine learning model and / or the response machine learning model. In some aspects, the multioutput machine learning model has been trained to average over the hidden representations of the first sample and the second sample or applies an attention mechanism over the hidden representations of the first sample and the second sample to predict a trajectory-based prediction.

[0025] In some aspects, the candidate ICI recipient is sorted as a responder if the predicted probability of the irAE is less than a predetermined adverse event threshold and / or the predicted probability of the positive outcome is greater than a predetermined positive outcome threshold and sorted as a non-responder if the predicted probability of the irAE is greater than the predetermined adverse event threshold and the predicted probability of the positive outcome is less than the predetermined positive outcome threshold. In some aspects, the predetermined adverse event threshold is about 25%, about 50%, about 75%, about 95%, or about 99%. In some aspects, the predetermined positive outcome threshold is about 25%, about 50%, about 75%, about 95%, or about 99%.

[0026] In some aspects, the multioutput machine learning model is a time sensitive machine learning model.

[0027] Also provided herein are methods for sorting a candidate immune checkpoint inhibition (ICI) recipient as a responder or non-responder, comprising: fluorescently labeling cells contained within a sample from the candidate ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel, generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer; inferring cell classifications based on the fluorescent intensity data; providing at least a subset of the cell classifications as input to a numerical quantification machine learning model, wherein the numerical quantification machine learning model has 7MF-367064958Attorney Docket No.: 279362000640been trained to predict a numerical quantification of a clinical outcome with training data comprising at least: (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints (b) one or more numerical quantifications of a clinical outcome for the plurality of ICI recipients at respective timepoints; outputting from the numerical quantification machine learning model, a predicted numerical quantification of the clinical outcome for the candidate ICI recipient; and sorting the candidate ICI recipient as a responder or a non-responder based at least on the predicted numerical quantification of the clinical outcome.

[0028] In some aspects, the one or more numerical quantifications of a clinical outcome comprise progression-free survival time (PFS), overall survival (OS) time, or time to a change in a biomarker. In some aspects, the respective timepoints for the one or more numerical quantification are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0029] In some aspects, sample is a first sample from a first timepoint. In some aspects, the method comprises inferring cells classification for a second sample at a second timepoint by: fluorescently labeling cells contained within a second sample from the candidate ICI recipient, by contacting at least an aliquot of the second sample with at least one immunophenotyping fluorescent labeling panel, generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer; inferring cell classifications based on the fluorescent intensity data. In some aspects, the methods comprise providing at least a subset of the cell classifications from the second sample as input to the numerical quantification machine learning model. In some aspects, wherein the numerical quantification machine learning model has been trained to average over the hidden representations of the first sample and the second sample or applies an attention mechanism over the hidden representations of the first sample and the second sample to predict a trajectorybased prediction.

[0030] In some aspects, candidate ICI recipient is sorted as a responder if the predicted numerical quantification of the clinical outcome is longer than a predefined time threshold and sorted as a non-responder of the predicted numerical quantification of the clinical outcome is shorter than the predefined time threshold. In some aspects, the predefined time threshold is about 1 month, 6 month, 1 year, 5 years or 10 years.

[0031] In some aspects, the numerical quantification machine learning model is a time sensitive machine learning model.8MF-367064958Attorney Docket No.: 279362000640

[0032] In some aspects, the candidate ICI recipient has received ICI therapy. In some aspects, the ICI recipients in the plurality of ICI recipients have received the same ICI therapy as the candidate ICI recipient. In some aspects, the methods comprise terminating ICI therapy if the candidate ICI recipient is sorted as a non-responder. In some aspects, the methods comprise administering the ICI therapy to the candidate ICI recipient if they are sorted as a responder.

[0033] In some aspects, the ICI therapy comprises administration of an anti-PDl therapy, administration of an anti-PDLl therapy, administration of an anti-CTLA4 therapy or administration of an anti-LAG3 therapy.

[0034] In some aspects, the sample comprises peripheral blood cells. In some aspects, the sample comprises isolated peripheral blood mononuclear cells (PBMCs).

[0035] Also provided herein are methods of monitoring response to an ICI therapy in candidate immune checkpoint inhibition (ICI) recipient, comprising: administering an ICI therapy to the candidate ICI recipient; at a first timepoint, sorting the candidate ICI recipient as a responder according to any one of methods described herein; at a second timepoint, sorting the candidate ICI recipient as a responder according to any one of according to any one of methods described herein; monitoring the response based on the candidate ICI recipient being sorted as a responder or a non-responder at the first timepoint and the second timepoint.

[0036] In some aspects, monitoring the response comprises maintaining a treatment plan if the candidate ICI recipient is sorted as a responder at the first timepoint and the second timepoint and modifying the treatment plan if the candidate ICI recipient is sorted as a non-responder at either the first timepoint or the second timepoint.

[0037] Also provided herein are methods of training an adverse event machine learning model to predict a probability of an immune-related adverse event, comprising: obtaining for each ICI recipient of a plurality of ICI recipients: (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, at least two cell classifications corresponding to different respective timepoints, (b) one or more indications of the irAE at respective timepoints; training an adverse event to predict the probability of the immune-related adverse event with training data comprising at least: (a) the plurality of cell classifications and (b) one or more indications of the irAE. In some aspects, the adverse event machine learning model is a time sensitive machine learning model. In some aspects, the respective timepoints for the one or more indications of the irAE are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0038] Also provided herein are methods of training a response machine learning model to predict a probability of a positive outcome, comprising: obtaining for each ICI recipient of a 9MF-367064958Attorney Docket No.: 279362000640plurality of ICI recipients: (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, at least two cell classifications corresponding to different respective timepoints, (b) one or more indications of the positive outcome at respective timepoints; training a response machine learning model to predict the probability of a positive outcome with training data comprising at least: (a) the plurality of cell classifications and (b) one or more indications of the positive outcome. In some aspects, response machine learning model is a time sensitive machine learning model. In some aspects, the respective timepoints for the one or more indications of the positive outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0039] Also provided herein are methods of training a multioutput machine learning model to predict a probability of a positive outcome and / or a probability of an immune-related adverse event (irAE), comprising: obtaining for each ICI recipient of a plurality of ICI recipients: (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, at least two cell classifications corresponding to different respective timepoints, (b) one or more indications of the positive outcome at respective timepoints; (c) one or more indications of the irAE at respective timepoints; training a response machine learning model to predict the probability of a positive outcome with training data comprising at least: (a) the plurality of cell classifications and (b) one or more indications of the positive outcome and one or more indications of the irAE. In some aspects, the multioutput machine learning model is a time sensitive machine learning model. In some aspects, the respective timepoints for the one or more indications of the irAE are the different respective timepoints from the plurality of cell classifications or are alternative timepoints. In some aspects, the respective timepoints for the one or more indications of the positive outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0040] Also provided herein are methods of training numerical quantification machine learning model to predict a numerical quantification of a clinical outcome, comprising: obtaining for each ICI recipient of a plurality of ICI recipients: (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, at least two cell classifications corresponding to different respective timepoints, (b) one or more numerical quantifications of a clinical outcome at respective timepoints; training a response machine learning model to predict the probability of a positive outcome with training data comprising at least: (a) the plurality of cell classifications and (b) one or more numerical quantifications of a clinical outcome. In some aspects, the numerical quantification machine learning model is a time sensitive machine learning model. In some aspects, the respective timepoints for the one or numerical 10MF-367064958Attorney Docket No.: 279362000640quantifications of a clinical outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0041] In some aspects, the cell classifications of the plurality of cell classification have been obtained by obtaining fluorescent intensity data, generated from a plurality of fluorescently labeled cells from the ICI recipient; inferring cell classifications based on the fluorescent intensity data.

[0042] In some aspects, the different respective timepoints comprise timepoints before administration of an ICI therapy, during administration of an ICI therapy and / or after administration of an ICI therapy. In some aspects, different respective timepoints are encoded into the time sensitive machine learning model using transformer-based architecture.

[0043] In some aspects, the transformer-based architecture comprises a temporal transformer. In some aspects, the time sensitive machine learning model comprises a recurrent neural network (RNN), a long short-term memory (LSTM) model, or a gated recurrent units (GRUs) model In some aspects, the time sensitive machine learning model comprises a feature extraction model and a longitudinal data model In some aspects, the feature extraction model comprises a convolutional neural network (CNN) or a graph neural network (GNN). In some aspects, the longitudinal data model comprises a recurrent neural network (RNN) a long shortterm memory (LSTM) model, a transformer, or a gated recurrent units (GRUs) model.

[0044] In some aspects, the fluorescent intensity data, generated from a plurality of fluorescently labeled cells is generated by a method comprising; fluorescently labeling cells contained within a sample from the ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel; generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer.

[0045] In some aspects, the sample was collected before administration of the ICI therapy, during administration of the ICI therapy and / or after administration of the ICI therapy.

[0046] In some aspects, one of the at least one immunophenotyping fluorescent labeling panel comprises a panel of fluorescent-labeled antibodies directed to cell surface proteins associated with antigen-presenting cells (APCs) and / or a panel of fluorescent-labeled antibodies directed to intracellular proteins. In some aspects, the panel of fluorescently-labeled antibodies comprises fluorescently-labeled antibodies directed to CD3, CD4, CD8, CD25, CD45, CD19, CD27, IgD, IgM, CD56, CD16, CD14, HLA-DR, CDllc, CD56, TCRgd, TCR Va7.2. TCR V61, TCR V62, TCR Va24-Jal8, CCR10, CD103 / ITGAE, CD122 / IL2RB, CD161 / KLRB1, CD223 / LAG-3, CD274 / PD-L1, CD335 / NKp46, CD43, CD10, CD138, CD141,11MF-367064958Attorney Docket No.: 279362000640CD183 / CXCR3, CD185 / CXCR5, CD194 / CCR4, CD197 / CCR7, CD279 / PD-1, CD28, CD294 / CRTH2, CD337 / NKp30, CD38, CD39, CD5, CD62L, CD86, CD95, ICOS, TIGIT, TIM-3, CD40, KLRG1, CD69, CD196 / CCR6, CDlc, CD24, CD267 / TACI, CD303 / BDCA-2 / CLEC4C, CD31, CD319, CD57, CD127, CD45RO, CD45RA, CCR2, CCR6, CXCR4, CX3CR1, Ki67, Granzyme B, TBET, GATA3, EOMES, BLIMP1, CTLA4, TCF1, TOX, BATE, IRF4, LEF1, ZEB2 or any combination thereof.

[0047] In some aspects, the panel of fluorescently labeled antibodies comprise fluorescently-labeled antibodies directed to cell surface markers that are indicative of live cells, dead cells, or both. In some aspects, the fluorescent intensity data comprises mean fluorescent intensity (MFI) data. In some aspects, inferring the cell classifications comprises inputting the fluorescent intensity data into a cell classification machine learning model and outputting cell classifications.

[0048] In some aspects, the cell classifications comprise cell ownership into at least 10, at least 20, at least 30, at least 40, at least 50, at least 60, at least 70, at least 80, at least 90, or at least 100 cell populations. In some aspects, the cell populations comprise distinct immune cell subpopulations. In some aspects, the distinct immune cell subpopulations comprise white blood cells (WBC), Eosinophils, Eosinophil / CD5+, Neutrophils, Neutrophils / big, Neutrophils / CD5+, Neutrophils / small. B-cells, B-cells / CD5-CD27-, Monocytes / CD56+, Monocytes / CD56-, NK-cells, Dendritic cells (DC), T-cells, iNKT cells, gamma delta T-cells (total GD), Vdl cells, Vd2 cells, Vdx cells, Mucosal-associated invariant T (MAIT) cells, TEMRA cells, CD4 naive cells, T helper cells, CD4 effector memory cells, Treg cells, Leukocytes, Helper T cells, Non-T / Non-NK B cells, Naive T cells, Memory T cells, Naive Cytotoxic T cells, Memory Cytotoxic T cells, Granulocytes, Activated T cells, Non-T / Non-B / Non-NK activated cells, CD3+CD4+FOXP3+ (regulatory T cells), or any combination thereof.

[0049] In some aspects, the flow cytometer is configured for at least about 5, at least about 10, at least about 15, at least about 20, at least about 30, at least about 40, at least about 50, at least about 60, at least about 70, at least about 80, at least about 90, or at least about 100 fluorescence detection channels. In some aspects, the flow cytometer is a full spectrum flow cytometer. In some aspects, the flow cytometer outputs mean fluorescent intensity (MFI) data.

[0050] Also provided herein are systems comprising: one or more processors; a memory communicative couples to the one or more processors and configures to store instructions that, when executed by the one or more processors cause the system to perform any of the methods described herein.12MF-367064958Attorney Docket No.: 279362000640

[0051] Also provided herein are non-transitory computer readable storage medium storing instructions which, when executed by one or more processors of a system, cause the system to perform the method of any one of the methods described herein.BRIEF DESCRIPTION OF THE FIGURES

[0052] Various aspects of the disclosed methods, devices, and systems are set forth with particularity in the appended claims. A better understanding of the features and advantages of the disclosed methods, devices, and systems will be obtained by reference to the following detailed description of illustrative embodiments and the accompanying drawing.

[0053] FIG. 1 illustrates an exemplary method for sorting a candidate ICI recipient as a responder or non-responder based on the output of an adverse event machine learning model.

[0054] FIG. 2 illustrates an exemplary method for sorting a candidate ICI recipient as a responder or non-responder based on the output of a response machine learning model.

[0055] FIG. 3 illustrates an exemplary method for sorting a multioutput ICI recipient as a responder or non-responder based on the output of an adverse event machine learning model.

[0056] FIG. 4 illustrates an exemplary method for sorting a candidate ICI recipient as a responder or non-responder based on the output of numerical quantification machine learning model.

[0057] FIG.5 shows an exemplary process for logistic regression modelling according to some of the embodiments described herein. Dotted outlines represent optional steps.

[0058] FIG.6 shows an exemplary process for logistic regression modelling according to some of the embodiments described herein. Dotted outlines represent optional steps.

[0059] FIG 7 shows an exemplary process for Adaptive Best Subset ensemble (ABSS) modeling according to some of the embodiments described herein.

[0060] FIG. 8 illustrates an exemplary computing system, in accordance with some of the embodiments and systems described herein.

[0061] FIG.9 illustrates an association between cell classifications and ICI treatment outcome.FIG. 9A illustrates an association between cell classifications inferred from samples collected before treatment and ICI treatment outcomes. FIG. 9B illustrates an association between cell classifications inferred from samples collected during ICI treatment and ICI treatment outcomes. FIG.9C illustrates an association between cell classifications inferred from samples collected before ICI treatment and during ICI treatment and ICI treatment outcomes.13MF-367064958Attorney Docket No.: 279362000640

[0062] FIG. 10A-10C illustrates an association between cell classifications and irAE severity.FIG. 10A illustrates an association between cell classifications inferred from samples collected before treatment and irAE severity. FIG. 10B illustrates an association between cell classifications inferred from samples collected during ICI treatment and irAE severity. FIG.10C illustrates an association between cell classifications inferred from samples collected before ICI treatment and during ICI treatment and irAE severity.

[0063] FIG. 11A-11B shows results of a logistic regression approach to ICI cohort - predicting treatment benefit from baseline and first on treatment (C2) samples. ROC-AUC performance on hold-out set of mean test and permutation models from 100 iterations of 80:20 CV. FIG.11A show baseline samples. FIG. 11B shows C2 samples.

[0064] FIG. 12A- 12B show results of an application of a logistic regression approach to ICI cohort - predicting max irAE severity from baseline, first on treatment (C2) and AE Dev samples. ROC-AUC performance on hold-out set of mean test and permutation models from 100 iterations of 80:20 cross validation (CV). FIG. 12A shows baseline samples. FIG. 12B shows C2 samples.

[0065] FIGs. 13A- 13B show results of an application of a logistic regression approach to an ICI cohort - predicting progression free survival (PFS) events with baseline (BL). FIG. 13A shows AUC. FIG. 13B shows balanced accuracy.

[0066] FIGs. 14A- 14B show results of an application of a logistic regression approach to an ICI cohort - predicting progression free survival (PFS) events from first on treatment samples (C2). FIG. 14A shows AUC. FIG. 14B shows balanced accuracy.

[0067] FIGs. 15A- 15B show results of an application of a logistic regression approach to an ICI cohort - predicting irAE severity with baseline (BL). FIG. 15A shows AUC. FIG. 13B shows balanced accuracy.

[0068] FIGs. 16A- 16B show results of an application of a logistic regression approach to an ICI cohort - predicting irAE severity from first on treatment samples (C2). FIG. 16A shows AUC. FIG. 16B shows balanced accuracy.

[0069] FIGs. 17A- 17B show results of an application of an ABSS approach to an ICI cohort for predicting irAE severity from baseline samples. FIG. 17A shows a ROC-AUC for a holdout cohort. FIG. 17B shows balanced accuracy.

[0070] FIGs. 18A- 18B show results of an application of an ABSS approach to an ICI cohort for predicting irAE severity from first on treatment (C2) samples. FIG. 18A shows a ROC-AUC for a hold-out cohort. FIG. 18B shows balanced accuracy.14MF-367064958Attorney Docket No.: 279362000640

[0071] FIGs. 19A-19J show results for an application of 5 models trained with an ABSS approach comprising logistic regression (LR) classifiers to predict progression free survival (PFS) from baseline samples for a first (FIGs. 19A, 19B, 19C, 19D, 19E) and a second (FIGs.19F, 19G, 19H, 191, 19J) hold out set.

[0072] FIGs. 20A -20B show results for an application of a model trained with an ABSS approach comprising LR classification and a multi-layer perceptron (MLP) for the final model to predict progression free survival from baseline samples for a first holdout (FIG. 20A) and a second holdout (FIG. 20B) set.DETAILED DESCRIPTION

[0073] Provided herein are methods and systems that can be used to classify a candidate ICI recipient as a responder or a non-responder to an ICI therapy. The methods and systems can be used for predicting response to, monitoring response following, and predicting immune related adverse events (irAEs) related to administration of an ICI therapy in a patient in need thereof. The methods can be performed with samples collected using non-invasive methods before, during, or after treatment to predict / monitor response and aid in treatment decisions. The methods and systems rely at least on machine learning models trained using samples collected from patients who received an ICI therapy at multiple timepoints, before, during, and / or after administering of the ICI therapy. Accordingly, the methods and systems described herein are likely to be more accurate at monitoring responses and adverse events in real time than previously described methods. The methods and systems described herein are dynamic, rapid, low cost, high throughput, and standardized for easy clinical application.

[0074] The FDA has approved two methods for predicting response to ICI therapy before administration. In contrast, to the methods and systems described herein, the known methods rely on information from a tumor sample. The first comprises a measurement of tumor mutational burden (TMB) and the second relies on a measurement of PDL1 expression in the tumor. These methods show marginal improvement in patient stratification, have variable effect across different tumor types, and rely on experimental / diagnostic techniques that can vary widely between different treatment centers, resulting in inconsistent data. By relying on tumor biopsies, the methods are inherently invasive, not compatible to continual sampling for treatment monitoring, and are susceptible to special heterogeneity.

[0075] The methods and systems described herein improve on other methods in that they can be performed with a blood sample rather than a tumor sample. As blood samples can be easily15MF-367064958Attorney Docket No.: 279362000640collected before, during, and after treatment of a patient, the methods and systems can provide real-time feedback and inform treatment decisions in real time. Some methods are being developed that rely capturing tumor cells from liquid biopsies. The accuracy of these methods may depend on tumor fraction in the sample and may not work for samples with low tumor fraction. As tumor fraction changes during treatment, the accuracy of the methods may change.

[0076] Response to ICI therapeutics is traditionally measured by imaging a solid tumor and comparing tumor size over time. Tumor shrinkage is downstream of effective immune system activation by ICI; thus, this methodology is slow and can only detect responses after substantial intervals, e.g., months. The methods described herein rely on immune signatures that can be seen in a blood sample as soon as a day after administration of treatment or between cycles of a treatment program. The methods and systems described herein can be used to predict response in real time by measuring the primary target of ICI therapeutics, the immune system of the patient.

[0077] Research based, unapproved methods using machine learning to predict response to ICI therapeutics without a tumor sample, such as in Yoo et al, Prediction of checkpoint inhibitor immunotherapy efficacy for cancer using routine blood tests and clinical data, Nature Medicine (2025), rely on models trained on demographic data about individuals and routine blood biochemistry data. In comparison to the methods described herein, the previously described methods rely on a limited view of the immune system which likely decreases their accuracy. Methods that are using immune phenotyping to predict ICI response, see Dyikanov et al, Comprehensive peripheral blood immunoprofiling reveals five immunotypes with immunotherapy response characteristics in patients with cancer Cancer Cell (2024), use immune phenotyping methods that may be too laborious to perform at multiple timepoints before, during, and / or after treatment and do not provide rich enough data to build individual prediction models. For example, Dyikanov et al., simplifies classifications using five immunophenotypes. In contrast, the methods described herein use immune profiling methods comprising standardization and scalability to permit large-scale profiling that generate huge datasets and sample profiling at multiple timepoints. The large and standard datasets allow for the construction and use of more accurate models. The immune profiling as described herein allows for complex predictions that match induvial treatment plans for candidate ICI recipient.

[0078] Currently, irAEs are monitored at the time of clinical presentations through patient reporting, routine blood marker tests, and physical examination. The methods and systems described herein can be used to predict and monitor irAE before clinical presentation because the methods rely on information directly from the target of the ICI therapeutics and where the 16MF-367064958Attorney Docket No.: 279362000640irAE’s stem from, the immune system. This is important because if irAEs are caught early, immunosuppressant drugs can be given to the patient to prevent severe adverse events.

[0079] irAE development is thought to reflect off-target impacts of the intended immune activation by ICI therapy. However, irAEs can be associated with improved patient outcomes, suggesting that ICI drug efficacy in activating an anti-cancer immune response is linked to an auto-inflammatory / auto-immune reaction. Das & Johnson, Immune-related adverse events and anti-tumor efficacy of immune checkpoint inhibitors, Journal for ImmunoTherapy of Cancer (2019). By combining response monitoring and irAE predictions, the methods and systems can also be used to determine if the treatment with the ICI therapeutic should be terminated. For example, if an individual is having a positive response yet is showing signs of a potential irAE, the severity of the irAE can be precited. The physician may recommend an individual maintains treatment with the ICI therapeutic if the irAE is predicted to be minor but may recommend termination of the treatment if a severe irAE is predicted.

[0080] The invention relies in part on improved methods for immunophenotyping and inference of cell classifications from detailed immune profiles. Conventionally methods for generating an immune profile by immunophenotyping include, e.g., enzyme-linked immunosorbent assays (ELISAs), immunoblotting techniques, and flow cytometry-based techniques include the use of panels of fluorescently-labeled antibodies directed to a variety of cell surface receptors and manual gating of the flow cytometry data. These techniques are often laborious and time consuming and are not easily scalable to a level that enables the processing of hundreds or thousands of samples. Recently, high throughput manifestations of deep phenotyping methods such as full spectrum flow cytometry have been developed as cost-effective techniques for immune cell profiling. The use of immune profiling needs only a non-invasive blood test that requires only a small volume of blood, enhancing patient comfort and enabling frequent testing with reduced burden on clinical infrastructure and costs.

[0081] The invention also relies in part on the inventors findings of associations between circulating immune cell phenotypes and patient outcomes with ICI therapy. By utilizing immunophenotyping and flow cytometry, the methods identify detailed immune cell signatures that can then be used in training and using machine learning algorithms for predicting and / or monitoring outcomes following administration of an ICI therapeutic to a patient in need thereof. The methods described herein can thus be used to monitor response to treatment to inform treatment decisions and to predict response to the treatment in order to stratify patients as those likely to respond to the treatment and / or have positive clinical outcomes following ICI therapy in comparison to those who would benefit from alternative treatment options. The 17MF-367064958Attorney Docket No.: 279362000640methods described herein can also be used to predict and monitor irAE events and severity to inform treatment with immunosuppressants or the need to stop treatment with the ICI therapy.

[0082] A candidate ICI recipient may be a patient in need of ICI therapy. A candidate ICI recipient may be a patient who has received ICI therapy. The patient may have been diagnosed with a cancer, wherein the cancer is known or suspected to be responsive to an ICI therapy. The models described herein are trained with data from a plurality of ICI recipients. The plurality of ICI recipients comprises patients who have received ICI therapy or are in need of ICI therapy. The patients may have been diagnosed with a cancer, wherein the cancer is known or suspected to be responsive to an ICI therapy. In some embodiments, the candidate ICI recipient patients and / or the patients in the plurality of ICI recipients are participants in the Understanding ImmunE-related toXicities by multifACeTed profiling (EXACT) trial and have been treated according to the trial protocols.I. Definitions

[0083] Unless otherwise defined, all technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the field to which this disclosure belongs.

[0084] As used in this specification and the appended claims, the singular forms “a”, “an”, and “the” include plural references unless the context clearly indicates otherwise. Any reference to “or” herein is intended to encompass “and / or” unless otherwise stated, and encompasses any and all possible combinations of one or more of the associated listed items.

[0085] As used herein, the terms “includes”, “including,” “comprises,” and / or “comprising” specify the presence of stated features, integers, steps, operations, elements, components, and / or units but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, units, and / or groups thereof.

[0086] Throughout this application, various parameter values may be presented in a range format. It should be understood that the description in range format is merely for convenience and brevity, and should not be construed as an inflexible limitation on the scope of the disclosure. Accordingly, the description of a range should be considered to have specifically disclosed all possible subranges as well as individual numerical values within that range, irrespective of whether a specific numerical value or specific sub-range is expressly stated. For example, description of a range such as from 1 to 6 should be considered to have specifically disclosed subranges such as from 1 to 3, from 1 to 4, from 1 to 5, from 2 to 4, from 2 to 6, from 3 to 6, etc., as well as individual numbers within that range, for example, 1, 1.4, 2, 3, 3.6, 4, 5, 5.8, and 6. This applies regardless of the breadth of the range.18MF-367064958Attorney Docket No.: 279362000640

[0087] Numbers may be expressed herein as being “about” a particular value. Similarly, ranges may be expressed herein as from “about” one particular value and / or to “about” another particular value. The terms “about” and “approximately” shall generally mean an acceptable degree of error or variation for a given value or range of values, such as, for example, a degree of error or variation that is within 20 percent (%), within 15%, within 10%, or within 5% of a given value or range of values.

[0088] It should be recognized that use of ordinal terms such as “first” and “second” in the description of methods and systems disclosed herein does not by itself connote any priority, order of importance of one system component over another, or the temporal order in which acts of a method are performed, but are used merely as labels to distinguish, for example, one system component having a certain name from another system component having the same name but for the use of the ordinal term to distinguish the two system components.

[0089] Additionally, various implementations of the methods and systems set forth herein may be described in terms of exemplary block diagrams, process flow charts, and other illustrations. As will become apparent to one of ordinary skill in the art after reading this document, the various implementations set forth herein can be implemented without confinement to the illustrated examples. For example, block diagrams and their accompanying description should not be construed as mandating a particular architecture or configuration. Similarly, in exemplary process flow charts, some blocks are optionally combined, the order of some blocks is optionally changed, and some blocks are optionally omitted. In some implementations, additional steps may be performed in combination with the exemplary processes. Accordingly, the methods and systems as described and illustrated in greater detail below are exemplary by nature and, as such, should not be viewed as limiting.

[0090] As used herein, the terms “flow cytometry” and “flow cytometer” refer to a technique and instrument, respectively, for performing flow cytometry where the instrument is configured to capture emission of fluorescent molecules using arrays of highly sensitive light detectors, thereby enabling the capture of highly multiplexed fluorescence intensity data sets. This includes all variants of flow cytometry and mass cytometry technology, including but not limited to conventional flow cytometry and full spectrum flow cytometry.

[0091] As used herein, the term “immunophenotyping panel” refers to a panel of binding agents, for example antibodies, abdurins, affibodies, affimers, affitins, anticalins, bicyclic peptides, darpins, fynomers, kunitz domains, and monobodies, (e.g., fluorescently-labeled antibodies) that bind to specific antigens or markers present on the surface of the cells, or in some cases, within the cell. These binding agents are labeled, such as fluorescently labeled,19MF-367064958Attorney Docket No.: 279362000640such that flow cytometry may be used to identify cells that have the antigen or marker for which the binding agent is specific. In one embodiment, fluorescently antibodies are used in the immunophenotyping panel.

[0092] The section headings used herein are for organizational purposes only and are not to be construed as limiting the subject matter described. The description is presented to enable one of ordinary skill in the art to make and use the invention and is provided in the context of a patent application and its requirements.

[0093] The disclosures of all publications, patents, and patent applications referred to herein are each hereby incorporated by reference in their entireties. To the extent that any reference incorporated by reference conflicts with the instant disclosure, the instant disclosure shall control.II. Methods for sorting a candidate ICI recipient as a responder or a nonresponder

[0094] Provided herein are methods that can be used to sort a candidate ICI recipient as a responder or a non-responder. The methods may be performed before the candidate receives ICI therapy and may be used to inform treatment decisions. The methods may be performed during or after the candidate received ICI therapy to inform follow up treatment or in the case of severe irAEs, potentially to inform termination of treatment with the ICI therapy. The methods for sorting a candidate ICI recipient as a responder or non-responder may comprise the methods in FIG. 1, FIG. 2, FIG. 3 or FIG. 4.

[0095] The methods comprise fluorescently labeling cells contained within a sample from the candidate ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel, generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer; inferring cell classifications based on the fluorescent intensity data; providing at least a subset of the cell classifications as input to a machine learning model (e.g. adverse event machine learning model, response machine learning model, multioutput machine learning model, numerical quantification machine learning model), wherein the machine learning model has been trained with training data comprising at least: (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints, (b) one or more response variables (e.g. indication of an irAE, indication of a positive outcome, numerical quantification of a clinical outcome), outputting from the machine learning model (e.g. adverse event machine learning model, response machine learning model, multioutput machine learning model,20MF-367064958Attorney Docket No.: 279362000640numerical quantification machine learning model) a prediction from the machine learning model, and sorting the candidate ICI recipient as a responder or a non-responder based at least on the prediction.

[0096] In some embodiments, the sample is a second sample from a second timepoint. Additional samples for additional timepoints may be collected and used as input into the methods described in FIGs. 1-4. In some embodiments, the methods comprise labeling cells from samples from two or more timepoints (e.g. before, during, or after treatment), generating fluorescent intensity data corresponding to the two or more timepoints, inferring cell classifications corresponding to the two or more timepoints, using one or more of the machine learning models (e.g. adverse event machine learning model, response machine learning model, multioutput machine learning model, numerical quantification machine learning model) to output predictions for each timepoint and sorting the candidate ICI recipient as a responder or non-responder based on the multiple predictions. In some embodiments, the sample is a first sample from a first timepoint.A. Generating fluorescent intensity data for a sample from a candidate ICI recipient

[0097] FIGs.1-4 provide and exemplary embodiments of a method for sorting a candidate ICI recipient as a responder, or a non-responder as described herein. At block 100, 200, 300, and 400, cells contained within a sample from the candidate ICI recipient are fluorescently labeled by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel as described herein. In some embodiments, the sample is collected before administration of ICI therapy. In some embodiments, the sample is collected after administration of ICI therapy. In some embodiments, the sample is collected during ICI therapy (e.g. between ICI therapy cycles).

[0098] In some embodiments, the sample is a fresh blood sample collected from the candidate ICI recipient. In some embodiments the sample is a cryopreserved blood sample that was collected preserved according to methods known in the art. In some embodiments, the sample comprises peripheral blood cells. In some embodiments, the sample comprises isolated peripheral blood mononuclear cells (PBMCs). In some embodiments, the methods comprise isolating PBMCs from a sample comprising peripheral blood samples.

[0099] In some embodiments, fluorescent labeling cells contained within the sample comprises contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel as described herein. In some embodiments, the sample is separated into one or more aliquots and each aliquot is contacted with an immunophenotyping labeling panel. The one or more immunophenotyping fluorescent labeling panel may be any of the labeling panels described herein. In some embodiments, one of the at least one immunophenotyping21MF-367064958Attorney Docket No.: 279362000640fluorescent labeling panel comprises a panel of fluorescent-labeled antibodies directed to cell surface proteins associated with antigen-presenting cells (APCs), as described herein.

[0100] At block 102, 202, 302, and 404 fluorescent intensity data is generated by processing the fluorescently labeled cells from block 100, 200, 300, and 400 respectively using a flow cytometer. Any of the methods described herein for generating fluorescent intensity data may be used at block 102, 202, 302, or 403. In some embodiments, the fluorescent intensity data may comprise mean fluorescent intensity (MFI) data, as described herein. In some embodiments, the flow cytometer outputs the MFI data.B. Inferring cell classifications

[0101] At block 104, 204, 304, and 404, cell classifications are inferred based on the fluorescent intensity data generated in block 102, 202, 302, and 402 respectively. In some embodiments, inferring the cell classifications comprises inferring cell classifications as part of an immune profile, as described herein. In some embodiments, inferring the cell classifications comprises a manual gating. In some embodiments, inferring the cell classifications comprises inputting the fluorescent intensity data into a cell classification machine learning model and outputting cell classifications (e.g. cell counts, cell ratios, or cell frequencies). In some embodiments, the machine learning models is a pretrained model trained using the method described in US18 / 353,022, and corresponding U.S. Patent publication US2024-0192210-Al, hereby incorporated by reference in its entirety.

[0102] In some embodiments, the cell classifications comprise predictions of cell type or subtype (e.g., immune cell sub-population) for individual cell detection events. In some embodiments, the cell classifications comprise cell counts (or frequencies) for each of a plurality of distinct cell types or subtypes. In some embodiments, the cell classifications comprise summary ratio data of cell classifications, such as ratios of immune cell subpopulations. In some embodiments, the cell classifications comprise summary ratio data of cell classifications, such as pre vaccine and post vaccine cell classifications. In some embodiments, the cell classifications comprise immune cell counts or classification for any of the immune cells or immune cell sub-populations described herein.C. Predicting the probability of an outcome based on one or more machine learning models

[0103] The cell classifications from blocks 104, 204, 304, and 404, are used as input to blocks 106, 206, 306, and 406 respectively. Blocks 106, 206, 306, and 406 comprise inputting the cell22MF-367064958Attorney Docket No.: 279362000640classification into a machine learning model (e.g. adverse event machine learning model, response machine learning model, multioutput machine learning model, numerical quantification machine learning model) to output a prediction (e.g. one or more predicted probability of an irAE, one or more predicted probability of a positive outcome, or one or more predicted numerical quantification of a clinical outcome and / or one or more combined predictions). In some embodiments, the adverse event machine learning model, response machine learning model, multioutput machine learning model and / or numerical quantification machine learning model are time sensitive machine learning models as described herein.

[0104] At block 106, at least a subset of the cell classifications generated at block 104, is provided into an adverse event machine learning model. The adverse event machine learning model has been trained to predict a probability of an immune-related adverse event (irAE) with at least (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints, (b) one or more indications of the irAE for the plurality of ICI recipients at respective timepoints.

[0105] In some embodiments, the adverse event machine learning model comprises a logistic regression machine learning model. In some embodiments, training the adverse event machine learning model comprises feature selection with logistic regression. In some embodiments, training the adverse event machine learning model comprises the processes of FIG. 5, as described herein.

[0106] In some embodiments, the adverse event machine learning model comprises a Adaptive Best Subset Ensemble (ABSS, also referred to an ABSSE) machine learning model. In some embodiments, training the adverse event machine learning model comprises the processes of FIG. 6, as described herein.

[0107] In some embodiments, the plurality of cell classifications comprises cell classifications for each ICI recipient of a plurality of ICI recipients from different timepoints. In some embodiments, the adverse event machine learning model has been trained with cell classification for at least 2, at least 3, at least 4, or 5 or more different respective timepoints for each ICI recipient. The cell classification may comprise cell classification generated according to any of the methods described herein and in US18 / 353,022, and corresponding U.S. Patent publication US2024-0192210-Al, hereby incorporated by reference in its entirety.

[0108] In some embodiments, the indications of an irAE are any of the indications of an irAE described herein. In some embodiments, the adverse event machine learning model has been trained with one or more, two or more, three or more, or four or more indications of the irAE 23MF-367064958Attorney Docket No.: 279362000640for each ICI recipient at the respective timepoints. In some embodiments, the different timepoints associated with the cell classifications may be the same or different from the respective timepoints associated with the indications of the irAE.

[0109] At block 108, the adverse event machine learning model from block 106 is used to output a predicted probability of the irAE for the candidate ICI recipient.

[0110] At block 206, at least a subset of the cell classifications generated at block 204, is provided into a response machine learning model. The response machine learning model has been trained to predict the probability of a positive outcome with training data comprising at least: (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints (b) one or more indications of the positive outcome for the plurality of ICI recipients at respective timepoints.

[0111] In some embodiments, the response machine learning model comprises a logistic regression machine learning model. In some embodiments, training the response machine learning model comprises feature selection with logistic regression. In some embodiments, training the response machine learning model comprises the processes of FIG. 5, as described herein.

[0112] In some embodiments, the response machine learning model comprises a Adaptive Best Subset Ensemble (ABSS) machine learning model.

[0113] In some embodiments, the plurality of cell classifications comprises cell classifications for each ICI recipient of a plurality of ICI recipients from different timepoints. In some embodiments, the response learning model has been trained with cell classification for at least 2, at least 3, at least 4, or 5 or more different respective timepoints for each ICI recipient. The cell classification may comprise cell classification generated according to any of the methods described herein and in US18 / 353,022, and corresponding U.S. Patent publication US2024-0192210-Al, hereby incorporated by reference in its entirety.

[0114] In some embodiments, the indications of a positive outcome are any of the indications of a positive outcome described herein. In some embodiments, the response machine learning model has been trained with one or more, two or more, three or more, or four or more indications of the positive outcome for each ICI recipient at the respective timepoints. In some embodiments, the different timepoints associated with the cell classifications may be the same or different from the respective timepoints associated with the indications of the positive outcome.24MF-367064958Attorney Docket No.: 279362000640

[0115] At block 208, the response machine learning model from block 206 is used to output a predicted probability of the positive outcome for the candidate ICI recipient.

[0116] In some embodiments, at least a subset of the cell classification generated at block 104 or block 204, is provided into a response machine learning model at block 206 and an adverse event machine learning model at block 106. Both the response machine learning model and the adverse event machine learning model can be used to output a predicted indication of a positive response and to output a predicted indication of an irAE.

[0117] At block 306, at least a subset of the cell classifications generated at block 304, is provided into a multioutput machine learning model. The multioutput machine learning model has been trained to predict the probability of a positive outcome and / or a probability of an irAE with training data comprising at least: (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints (b) one or more indications of the positive outcome for the plurality of ICI recipients at respective timepoints; (c) one or more indications of the irAE for the plurality of ICI recipients at respective timepoints.

[0118] In some embodiments, the plurality of cell classifications comprises cell classifications for each ICI recipient of a plurality of ICI recipients from different timepoints. In some embodiments, the multioutput machine learning model has been trained with cell classification for at least 2, at least 3, at least 4, or 5 or more different respective timepoints for each ICI recipient. The cell classification may comprise cell classification generated according to any of the methods described herein and in US18 / 353,022, and corresponding U.S. Patent publication US2024-0192210-Al, hereby incorporated by reference in its entirety.

[0119] In some embodiments, the indications of an irAE are any of the indications of an irAE described herein. In some embodiments, the multioutput machine learning model has been trained with one or more, two or more, three or more, or four or more indications of the irAE for each ICI recipient at the respective timepoints.

[0120] In some embodiments, the indications if a positive outcome are any of the indications of a positive outcome described herein. In some embodiments, the multioutput learning model has been trained with one or more, two or more, three or more, or four or more indications of the positive outcome for each ICI recipient at the respective timepoints.

[0121] In some embodiments, the different timepoints associated with the cell classifications may be the same or different from the respective timepoints associated with the indications of25MF-367064958Attorney Docket No.: 279362000640the irAE and may be the same or different from the respective timepoints associated with the indications of the positive outcomes.

[0122] At block 308, the multioutput machine learning model from block 306 is used to output a predicted probability of the irAE and / or the predicted probability of the positive outcome for the candidate ICI recipient.

[0123] At block 406, at least a subset of the cell classifications generated at block 404 is provided into a numerical quantification machine learning model. The numerical quantification machine learning model has been trained to predict a numerical quantification a clinical outcome with training data comprising at least: (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints (b) one or more numerical quantifications of a clinical outcome for the plurality of ICI recipients at respective timepoints.

[0124] In some embodiments, the plurality of cell classifications comprises cell classifications for each ICI recipient of a plurality of ICI recipients from different timepoints. In some embodiments, the numerical quantification machine learning model has been trained with cell classification for at least 2, at least 3, at least 4, or 5 or more different respective timepoints for each ICI recipient. The cell classification may comprise cell classification generated according to any of the methods described herein and in US18 / 353,022, and corresponding U.S. Patent publication US2024-0192210-Al, hereby incorporated by reference in its entirety.

[0125] In some embodiments, the numerical quantifications of a clinical outcome are any of the numerical quantifications of a clinical outcome described herein. In some embodiments, the numerical quantification machine learning model has been trained with one or more, two or more, three or more, or four or more indications of the numerical quantification of a clinical outcome for each ICI recipient at the respective timepoints. In some embodiments, the different timepoints associated with the cell classifications may be the same or different from the respective timepoints associated with numerical quantification of the clinical outcome.

[0126] At block 408, the numerical quantification machine learning model from block 406 is used to output a predicted numerical quantification of a clinical outcome for the candidate ICI recipient.

[0127] In some embodiments, at least a subset of the cell classification generated at block 104, block 204, block 304, or block 404, is provided into any combination of one or more response machine learning models, one or more adverse event machine learning models, one or more multioutput machine learning models, and one or more numerical quantification machine 26MF-367064958Attorney Docket No.: 279362000640learning models. The machine learning models may be used to predict one or more indication of a positive outcome, one or more indication of an irAE and / or one or more predicted numerical quantification. The at least a subset of the cell classifications input into any combination of the machine learning models may comprise data generated from samples collected at one or more timepoint (e.g. before, during, or after the candidate ICI recipient receives ICI therapy).

[0128] In some embodiments, the methods comprise combining the one or more predicted probability of an irAE, one or more predicted probability of a positive outcome, or one or more predicted numerical quantification of a clinical outcome. In some embodiments, combing comprises a voting algorithm. The voting algorithm may be based on a majority vote or weighted voting. In some embodiments, the combining comprises a stacking algorithm. In some embodiments, the stacking algorithm comprises feeing each of the prediction into a multilearning model. In some embodiments, the combining comprises average and / or Logit blending. In some embodiments, Logit blending comprises use of a mixed logit model.D. Sorting a candidate ICI recipient according to the results of one or more machine learning models

[0129] The methods described herein comprise sorting a candidate ICI recipient as a response or non-responder based on prediction. The predictions may be one or more predicted probability of an irAE, one or more predicted probability of a positive outcome, or one or more predicted numerical quantification of a clinical outcome and / or one or more combined predictions. The predictions may be outputs of blocks 108, 208, 308, and / or 408. The predictions may be based on samples collected at one or more timepoints from the candidate ICI recipient (e.g. before, during, or after the candidate ICI recipient receives ICI therapy).

[0130] At block 110, at least the predicted probability of an irAE from block 108 is used to sort the candidate ICI recipient as a responder or non-responder. In some embodiments, the candidate ICI recipient is sorted as a responder if the predicted probability of the irAE is less than a predetermined adverse event threshold and sorted as a non-responder if the predicted probability of the irAE is greater than the predetermined adverse event threshold.

[0131] In some embodiments, the predetermined adverse event threshold is about 25%, about 50%, about 75%, about 90% or about 99%. In some embodiments, the predetermined adverse event threshold is between about 0% and 25%, 0% and 50%, 0% and 75%, 0% and 90%, 0% and 99%. In some embodiments, the predetermined adverse event threshold is between about 25% and 50%, 25% and 75%, 25% and 90%, 25% and 99%, 25% and 99%, 25% and 99%,27MF-367064958Attorney Docket No.: 27936200064050% and 99%, 75% and 90%, or 90% and 99%. It is understood that the predetermined adverse event threshold may depend on the severity of the irAE, the severity of the cancer, and / or additional clinically relevant parameters. In some embodiments, the predetermined adverse event threshold is user defined.

[0132] At block 210, at least the predicted probability of a positive outcome from block 208 is used to sort the candidate ICI recipient as a responder or non-responder. In some embodiments, the candidate ICI recipient is sorted as a responder if the predicted probability of the positive outcome is greater than a positive outcome and sorted as a non-responder if the predicted probability of the positive outcome is less than the predetermined positive outcome threshold.

[0133] In some embodiments, the positive outcome event threshold is about 25%, about 50%, about 75%, about 90% or about 99%. In some embodiments, the predetermined positive outcome threshold is between about 0% and 25%, 0% and 50%, 0% and 75%, 0% and 90%, 0% and 99%. In some embodiments, the predetermined positive outcome threshold is between about 25% and 50%, 25% and 75%, 25% and 90%, 25% and 99%, 25% and 99%, 25% and 99%, 50% and 99%, 75% and 90%, or 90% and 99%. It is understood that the predetermined positive outcome threshold may depend on the severity of an irAE, the severity of the cancer, and / or additional clinically relevant parameters. In some embodiments, the predetermined positive outcome threshold is user defined.

[0134] In some embodiments, the predicted probability of an irAE from block 108 and the predicted probability of a positive outcome from block 208 are used to sort the candidate ICI recipient as a responder or non-responder. In some embodiments, the candidate ICI recipient is sorted as a responder if the predicted probability of the irAE is less than a predetermined adverse event threshold and / or the predicted probability of the positive outcome is greater than a predetermined positive outcome threshold and sorted as a non-responder if the predicted probability of the irAE is greater than the predetermined adverse event threshold and the predicted probability of the positive outcome is less than the predetermined positive outcome threshold.

[0135] At block 310, at least the predicted probability of the positive outcome and / or the predicted probability of the irAE from block 308 are used to sort the candidate ICI recipient as a responder or non-responder. In some embodiments, the candidate ICI recipient is sorted as a responder if the predicted probability of the irAE is less than a predetermined adverse event threshold and / or the predicted probability of the positive outcome is greater than a predetermined positive outcome threshold and sorted as a non-responder if the predicted probability of the irAE is greater than the predetermined adverse event threshold and the 28MF-367064958Attorney Docket No.: 279362000640predicted probability of the positive outcome is less than the predetermined positive outcome threshold.

[0136] At block 410, at least the predicted numerical quantification of the clinical outcome from block 408 is used to sort the candidate ICI recipient as a responder or non-responder. In some embodiments, the candidate ICI recipient is sorted as a responder if the predicted numerical quantification of the clinical outcome is longer than a predefined time threshold and sorted as a non-responder of the predicted numerical quantification of the clinical outcome is shorter than the predefined time threshold.

[0137] In some embodiments, the predetermined time threshold is about 1 month, 6 months, 1 year, 5 years, or 10 years. In some embodiments, the predestined time threshold is between 1 month and 6 months, 1 month and 1 year, 1 month and 5 years, 1 month and 10 years, 6 months and 1 year, 6 months and 5 years, 6 months and 10 years, 1 year and 5 years, 1 year and 10 year, or 5 years and 10 years. It is understood that the predetermined time threshold may depend on the severity of any irAE, the severity of the cancer, and / or additional clinically relevant parameters. In some embodiments, the predetermined time threshold is user defined.

[0138] In some embodiments, one or more predictions (e.g. one or more predicted probability of an irAE, one or more predicted probability of a positive outcome, or one or more predicted numerical quantification of a clinical outcome and / or one or more combined predictions) are used to sort the candidate ICI recipient as a responder or non-responder. In some embodiments, the one or more predictions are based on two or more timepoints. In some embodiments, demographic and / or clinical considerations may also be used to also be used to sort the candidate ICI recipient as a responder or non-responder. In some embodiments, ensemble or meta learning models may be used to sort a candidate ICI recipient as a responder or non-responder.E. Applications

[0139] The methods for sorting a candidate ICI recipient as a responder or a non-responder according to the methods described herein may be used to inform treatment of the candidate ICI recipient. In some embodiments, the candidate ICI recipient is administered the ICI therapy if they are sorted as responder. In some embodiments, the ICI therapy is terminated if the candidate ICI recipient is sorted as a non-responder. In some embodiments, additional anticancer therapies are administered, ICI therapy dose is changed, or immunosuppressants are administered to the ICI recipient if the candidate ICI recipient is sorted as a non-responder. Sorting of the ICI candidate ICI recipient can be used to inform clinical decision making for 29MF-367064958Attorney Docket No.: 279362000640the candidate ICI recipient and may improve the prognosis or quality of life for the candidate ICI recipient.

[0140] In some embodiments, the methods for sorting a candidate ICI recipient as a responder or a non-responder according to the methods described herein can be used to inform biomarker discovery for response to ICI therapy or development of an irAE. In some embodiments, the biomarkers may be targetable inflammatory signatures that can be used for informing discovery of new methods to improve ICI therapy and prevent irAEs.III. Methods monitoring response to an ICI therapy

[0141] Also provided herein are methods of monitoring a response to an ICI therapy in a candidate ICI recipient. The methods comprise administering and ICI therapy to the candidate ICI recipient and using the methods described herein to sort the candidate ICI receipt as a responder or non-responder at multiple time points (e.g. first timepoint and a second timepoint). In some embodiments, the methods comprise sorting the candidate ICI recipient as a responder at 2 or more, 3 or more, 4 or more, or 5 or more timepoints. Monitoring the response is based on the soring of the candidate ICI recipient as a responder or non-responder at two or more timepoints.

[0142] Sorting the candidate ICI receipt as a responder and / or a non-responder at two or more timepoints can inform treatment. In some embodiments, monitoring the response comprises maintaining a treatment plan if the candidate ICI recipient is sorted as a responder at the first timepoint and the second timepoint and modifying the treatment plan if the candidate ICI recipient is sorted as a non-responder at either the first timepoint or the second timepoint.

[0143] In some embodiments, monitoring a response comprises using the methods described herein that can be used to sort a candidate ICI recipient as a responder or a non-responder. The methods for sorting a candidate ICI recipient as a responder or non-responder may comprise the methods in FIG. 1, FIG. 2, FIG. 3 or FIG. 4 and in the accompanying embodiments described herein.

[0144] In some embodiments, the first timepoint is before administration of the ICI therapy, immediately after administration of the ICI therapy, during administration of an ICI therapy cycle, between cycles of the ICI therapy, or after completion of the therapy cycles. In some embodiments, the second timepoint is any timepoint after the first timepoint. In some embodiments, the second timepoint is before administration of the ICI therapy, immediately after administration of the ICI therapy, during administration of an ICI therapy cycle, between cycles of the ICI therapy, or after completion of the therapy cycles. In some embodiments, a30MF-367064958Attorney Docket No.: 279362000640third timepoint, a fourth timepoint, and any additional timepoints are sequential and take place during administration of an ICI therapy cycle, between cycles of the ICI therapy, or after completion of the therapy cycles. In some embodiments, including timepoints before, during, and after administration of the ICI therapy allow for monitoring of the change in the immune profile of the candidate ICI recipient in response to the therapy.IV. Methods for training a machine learning model

[0145] Also provided herein are methods for training a machine learning model (e.g. adverse event machine learning model, response machine learning model, multioutput machine learning model, numerical quantification machine learning model). As described herein, the machine learning models (e.g. adverse event machine learning model, response machine learning model, multioutput machine learning model, numerical quantification machine learning model) may comprise time sensitive machine learning models. The training methods disclosed herein can be used to train the machine learning models described herein (e.g. adverse event machine learning model, response machine learning model, multioutput machine learning model, numerical quantification machine learning model) for use in sorting a candidate ICI recipient as a responder or a non-responder. In some embodiments, the training methods can be used to train machine learning models for use in monitoring a response to ICI therapy as described herein.

[0146] The machine learning models (e.g. adverse event machine learning model, response machine learning model, multioutput machine learning model, numerical quantification machine learning model) are trained with training data from a plurality of ICI recipients collected a two or more respective timepoints. In some embodiments, the training data comprises cell classification generated using the methods described herein and one or more indication of an irAE as described herein, one or more indications of a positive outcome as described herein, and / or one or more numerical quantifications of a clinical outcome as described herein.

[0147] In some embodiments, the machine learning models comprise logistic regression machine learning models as described herein. In some embodiments, training the logistic regression machine learning models comprises selecting features from the training data using logistic regression. In some embodiments, training a logistic regression model as described herein comprises the exemplary process in FIG. 5. In some embodiments, training a logistic regression model as described herein comprises the exemplary process in FIG. 6.31MF-367064958Attorney Docket No.: 279362000640

[0148] In some embodiments, the machine learning models comprise Adaptive Best Subset Ensemble (ABSS) machine learning models as described herein. In some embodiments, training the Adaptive Best Subset Ensemble (ABSS) machine learning model In some embodiments, training the ABSS comprises the processes of FIG. 7, as described herein.

[0149] The plurality of ICI recipients comprises patients who have received ICI therapy or are in need of ICI therapy. The patients may have been diagnosed with a cancer, wherein the cancer is known or suspected to be responsive to an ICI therapy. In some embodiments, the patients in the plurality of ICI recipients are participants in the Understanding ImmunE-related toXicities by multifACeTed profiling (EXACT) trial and have been treated according to the trial protocols.A. Obtaining training data

[0150] Training a machine learning model (e.g. adverse event machine learning model, response machine learning model, multioutput machine learning model, numerical quantification machine learning model) comprises obtaining for each ICI recipient of a plurality of ICI recipients: (a) a plurality of cell classifications, wherein the plurality of cell classifications includes, at least two cell classifications corresponding to different respective timepoints. The methods for sorting a candidate ICI recipient as a responder or non-responder and for monitoring response to an ICI therapy rely at least on the time richness in the training data as described herein. Immune profiles comprising cell classifications from multiple time points for each of the ICI recipients in the plurality of ICI recipients. As described herein, the time-sensitive models can work with and learn from data collected along a continuous timeframe.

[0151] In some embodiments, the cell classifications of the plurality of cell classifications as described herein and as described in described in US 18 / 353,022, and corresponding U.S. Patent publication US2024-0192210-Al, hereby incorporated by reference in its entirety. In some embodiments, the cell classifications have been obtained by: obtaining fluorescent intensity data, generated from a plurality of fluorescently labeled cells from the ICI recipient; inferring cell classifications based on the fluorescent intensity data. In some embodiments, the fluorescent intensity data, generated from a plurality of fluorescently labeled cells is generated by a method comprising; fluorescently labeling cells contained within a sample from the ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel; generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer.32MF-367064958Attorney Docket No.: 279362000640

[0152] In some embodiments, the plurality of ICI recipients comprises at least 10, at least 20, at least 30, at least 40, at least 50, at least 60, at least 70, at least 80, at least 90 at least 100, at least 200, at least 300, at least 400, at least 500, at least 600, at least 700, at least 800, at least 900, at least 1000, at least 1500, at least 2000, at least 2100, at least 2200, or at least 2500 ICI recipients. In some embodiments, the plurality of ICI recipients comprises between about 10 and about 100 individuals, about 20 and 100 individuals, about 40 and 100 individuals, 50 and 100 individuals 60 and 100 individuals, or 100 more individuals. In some embodiments, the plurality of plurality of ICI recipients comprises between 1000 and 2500, between 1000 and 2000, or between 1000 and 1500 individuals.

[0153] In some embodiments, training the adverse event machine learning model and the multioutput machine learning model comprises obtaining for each ICI recipient one or more indications of an irAE at respective timepoints. The one or more indication of the irAE may be any of the indication of an irAE as described herein. In some embodiments, the different timepoints associated with the cell classifications may be the same or different from the respective timepoints associated with the indications of the irAE.

[0154] In some embodiments, training the response machine learning model and the multioutput machine learning model comprises obtaining one or more indication of a positive outcome at respective timepoints. The one or more indications of a positive outcome may be any of the indication of positive outcomes as described herein. In some embodiments, the different timepoints associated with the cell classifications may be the same or different from the respective timepoints associated with the indications of the positive outcomes.

[0155] In some embodiments, training the numerical quantification machine learning model comprises obtaining one or more numerical quantifications of a clinical outcome at respective timepoints. The one or more numerical quantification of a clinical outcome may be any of the numerical quantifications of a clinical outcome as described herein. In some embodiments, the different timepoints associated with the cell classifications may be the same or different from the respective timepoints associated with the one or more numerical quantifications of the clinical outcome.

[0156] It is understood that training models may comprise obtaining cell classification, indications of irAEs, indications of a positive outcomes and / or numerical quantification of a clinical outcome at multiple timepoints. The multiple timepoints may be any timepoint before, during, or after administration of an ICI therapy. The multiple timepoints may be during or between cycles of an ICI therapy.33MF-367064958Attorney Docket No.: 279362000640

[0157] In some embodiments, additional details related to the ICI recipient may be used for training the machine learning models. The details may include but are not limited to cancer stage, cancer type, results from additional clinical test and / or demographic data. In some embodiments, the ICI therapy administered to the ICI recipients may be used for training the machine learning models.

[0158] In some embodiments, the different timepoints associated with the cell classifications may be the same or different from the respective timepoints associated with the indications of the irAE, may be the same or different from the respective timepoints associated with the indications of the positive outcomes, and may be the same or different from the respective timepoints associated with the one or more numerical quantifications of the clinical outcome.B. Training the machine learning model

[0159] Once training data is obtained, the methods comprise training an adverse event to predict the probability of the immune-related adverse event with training data comprising at least: (a) the plurality of cell classifications and (b) one or more indications of the irAE. Training the model may comprise any of the methods described herein for training a machine learning model (e.g. a time sensitive machine learning model).

[0160] Once training data is obtained, the methods comprise training a response machine learning model to predict the probability of a positive outcome with training data comprising at least: (a) the plurality of cell classifications and (b) one or more numerical quantifications of a clinical outcome. Training the model may comprise any of the methods described herein for training a machine learning model (e.g. a time sensitive machine learning model). In some embodiments, training the machine learning model comprises using logistic regression to select representative features as described herein. In some embodiments, training the machine learning model comprises multi-layer feature selection.

[0161] Once training data is obtained, the methods comprise training a multioutput machine learning model to predict the probability of a positive outcome with training data comprising at least: (a) the plurality of cell classifications and (b) one or more indications of the positive outcome and one or more indications of the irAE. Training the model may comprise any of the methods described herein for training a machine learning model (e.g. a time sensitive machine learning model). In some embodiments, training the machine learning model comprises using logistic regression to select representative features as described herein. In some embodiments, training the machine learning model comprises multi-layer feature selection.34MF-367064958Attorney Docket No.: 279362000640

[0162] Once training data is obtained, the method comprise training a numerical quantifications machine learning model to predict the probability of a positive outcome with training data comprising at least: (a) the plurality of cell classifications and (b) one or more numerical quantifications of a clinical outcome. Training the model may comprise any of the methods described herein for training a machine learning model (e.g. a time sensitive machine learning model).

[0163] In some embodiments, the training may be based on the results of the training of other machine learning models trained using the methods described herein. In some embodiments, the methods may comprise training multiple machine learning models (e.g. adverse event machine learning models, response machine learning models, multioutput machine learning models, and / or numerical quantification machine learning models) with data from different subsets of ICI recipients from the plurality of ICI recipients or with data from different subsets of timepoints. The subsets may be selected based on characteristics of the ICI recipients or the ICI therapy the trained model will be used for inference for. As described herein, training machine learning models with data from different subsets of individuals may improve the accuracy of a prediction generated with the machine learning model.V. Fluorescent intensity data and Cell classifications

[0164] Provided herein are methods comprising generating and using an immune profile for an individual or plurality of individuals. The immune profile may comprise cell classification inferred from fluorescent intensity data based on samples collected before and / or after administration of an ICI therapy. In some embodiments, the cell classifications comprise cell counts, cell ratios, and / or cell frequencies. The methods comprise fluorescently labeling cells contained within a sample (e.g. before, during, and / or after ICI treatment), by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel and generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer. In some embodiments, the methods comprise inferring cell classification from the fluorescent intensity data. In some embodiments, a machine learning algorithm may be used to infer cell classifications from fluorescent intensity data. In some embodiments, matched samples from an individual receiving an ICI therapy are used in the methods described herein for real time monitoring before, during, and after treatment.

[0165] In some embodiments, the fluorescent intensity data is obtained using flow cytometry. In some embodiments, the fluorescent intensity data are processed into flow cell classifications.35MF-367064958Attorney Docket No.: 279362000640In some embodiments, the fluorescent intensity data is obtained using flow cytometry followed by machine learning models to analyze cells and classify them using a standardized set of immune system status antibody panels. In some embodiments, the flow cytometer outputs cells classifications. In some embodiments, the flow cytometer outputs mean fluorescent intensity (MFI) data.

[0166] In some embodiments, the fluorescent intensity data is obtained using flow cytometry. In some embodiments, the fluorescent intensity data is generated using a flow cytometry to process fluorescently labeled cells from the sample. In some embodiments, the flow cytometer is configured for at least about 5, at least about 10, at least about 15, at least about 20, at least about 30, at least about 40, at least about 50, at least about 60, at least about 70, at least about 80, at least about 90, or at least about 100 fluorescent detection channels. In some embodiments, the flow cytometer is configured for between about 5 and about 100, between about 10 and about 90, between about 20 and about 80, between about 30 and about 70, or between about 40 and about 60 fluorescent detection channels. In some embodiments, the flow cytometry is a full spectrum flow cytometer.

[0167] In some embodiments, flow cytometry is performed on sample. In some embodiments, the samples are received at the laboratory facility, the sample is prepared and analyzed with flow cytometry. In some embodiments, preparing the sample comprises performing one or more of a dilution step, a centrifugation step, a staining step (using one or more fluorescently-labeled antibody panels) and / or a wash step.

[0168] In some embodiments, the staining step comprises contacting cells contained within a sample with at least one immunophenotyping fluorescent labeling panel (i.e., an immunophenotyping panel or flow cytometry panel). In some embodiments, the one immunophenotyping fluorescent labeling panel comprises fluorescently-labeled antibodies directed to a set of specific cell surface antigens (e.g., cell surface proteins) that collectively enable discrimination between the cell types or cell subtypes of interest. Sample processing may also include immunophenotyping panel design. A sample processing platform may comprise contacting each of one or more sample aliquots (e.g., 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, or more than 10 sample aliquots) with one or more flow cytometry panels (e.g., 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, or more than 10 flow cytometry panels).

[0169] In some embodiments, the sample is collected from an individual (e.g. patient) receiving administration of an ICI therapy in an individual in need thereof. In some embodiments, the sample comprises isolated peripheral blood mononuclear cells (PBMCs).36MF-367064958Attorney Docket No.: 279362000640

[0170] In some embodiments, a flow cytometry panel or immunophenotyping panel may comprise at least about 5, at least about 10, at least about 15, at least about 20, at least about 30, at least about 40, at least about 50, at least about 60, at least about 70, at least about 80, at least about 90 or at least about 100 fluorescently-labeled antibodies directed to a set of cell surface antigens. In some embodiments, a flow cytometry panel or immunophenotyping panel may comprise between about 5 and about 100, between about 10 and about 90, between about 20 and about 80, between about 30 and about 70, or between about 40 and about 60 fluorescently-labeled antibodies directed to a set of cell surface antigens.

[0171] In some embodiments, cells from each of the sample may be divided into aliquots and the aliquots may be stained with a different flow cytometry panel, one focusing on the antigen-presenting cell (APC) arm of the immune system (A panel), which comprises antibodies directed to a plurality different cell surface proteins, e.g. 36, and the other focusing on the adaptive arm of the immune systems (T panel), which comprises antibodies directed to a second plurality of cell surface markers, e.g. 41 cell surface proteins, and one focusing on T intracellular proteins (T intracellular panel, TIC), e.g. 33 intracellular proteins. In some instances, the panels may also include cell viability staining to distinguish between live cells and dead cells. In some instances, the panels may also comprise an autofluorescence measurement as a “marker”. Non-limiting examples of the cell surface proteins and additional markers that may be included in these panels are listed in Table 1.Table 1. Non-limiting examples of cell surface receptor proteins and other markers for distinguishing between immune cell sub-populations.37MF-367064958Attorney Docket No.: 279362000640

[0172] In some embodiments, the panels include markers for determining immune cell type, immune system activation, lineage (e.g., the main marker(s) that are commonly used to define a certain cell population prior to further subsetting the cell type; examples include, but are not limited to, CD3 to define total T cells, and CD56 and CD16 to define natural killer cells), and exhaustion (cells that express markers associated with “cell exhaustion” (e.g., PD-1, TIGIT) can no longer proliferate and lose their functionalities as a result of chronic stimulation / prolonged activation of immune response).

[0173] In some embodiments, the panels include markers such as lineage markers for aPT cells, invariant T cells, y6T cells, B cells, NK cells, monocytes, macrophages, dendritic cells, neutrophils, eosinophils, and basophils. In some embodiments, the lineage markers include CD3, CD4, CD8, CD25, CD45, CD19, CD27, IgD, IgM, CD56, CD16, CD14, HLA-DR, CDllc, CD56, FOXP3, TCRgd, TCR Va7.2. TCR V61, TCR V62, TCR Va24-Jal8.

[0174] In some embodiments, the panels include markers such as functional markers relating to but not limited to activation, migration, exhaustion, senescence, or memory status of cells. In some embodiments, the functional markers include CCR10, CD103 / ITGAE, CD122 / IL2RB, CD161 / KLRB1, CD223 / LAG-3, CD274 / PD-L1, CD335 / NKp46, CD43, CD10, CD138, CD141, CD183 / CXCR3, CD185 / CXCR5, CD194 / CCR4, CD197 / CCR7, CD279 / PD-1, CD28, CD294 / CRTH2, CD337 / NKp30, CD38, CD39, CD5, CD62L, CD86, CD95, ICOS, TIGIT, TIM-3, CD40, KLRG1, CD69, CD196 / CCR6, CDlc, CD24, CD267 / TACI, CD303 / BDCA-2 / CLEC4C, CD31, CD319, CD57, CD127, CD45RO, CD45RA, CCR2, CCR6, CXCR4, CX3CR1, Ki67, Granzyme B, TBET, GATA3, EOMES, BLIMP1, CTLA4, TCF1, TOX, BATE, IRF4, LEF1, and ZEB2.

[0175] In some embodiments, the panels include markers able to detect the binding of administered ICI therapy to cells. In some embodiments, the markers include an anti-human IgGl to detect an anti-PDl ICI therapeutics bound to immune cells. In some embodiments, the markers include an anti-human IgGl to detect an anti-CTLA4 ICI therapeutics bound to immune cells. In some embodiments, the markers include an anti-human IgGl to detect an anti-LAG3 ICI therapeutics bound to immune cells.

[0176] In some embodiments, the fluorescent intensity data is obtained using flow cytometry and is in the form of a flow cytometry standard FCS file. In some embodiments the flow cytometry data comprises mean fluorescent intensity (MFI) data. In some embodiments, the FCS file may comprise, for example, fluorescence intensity data for one or more fluorescence detection channels (e.g., 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 20, 25, 30, 35, 40, 50, or more than 50 fluorescence detection channels), as well as data derived therefrom (e.g., forward 38MF-367064958Attorney Docket No.: 279362000640scatter height data, forward scatter area data, side scatter height data, side scatter area data, autofluorescence data, or any combination thereof). In some instances, the number of fluorescence detection channels available may be determined by, for example, a combination of the detection hardware available as part of the flow cytometry instrument (e.g., comprising 5, 10, 20, 25, 50, 75, 100, 125, 150, 175, 200, or more than 200 detectors) and the number of spectrally-distinct fluorophores (e.g., 5, 10, 20, 25, 30, 35, 40, 45, 50, 60, or more than 60 spectrally-distinct fluorophores).

[0177] In some embodiments, manual gating may be used to determine flow cell classification for the plurality of cells. In some embodiments, manual gating results in about 10-2500, about 200-2500, about 1000-2500 cell classifications, or 2000-2300 cell classifications. In some embodiments, manual gating may be performed by an expert, e.g. an immunologist. In some embodiments, the cell classification may relate to cell types, cell subtypes, or cell states.

[0178] In some embodiments, manual gating results in about 1000, about 2000, about 3000, about 4000, about 5000, about 6000, about 7000, about 8000, about 9000, about 10000, about 11000, about 12000, about 13000, about 14000, about 15000, about 16000, about 17000, about 18000, about 19000 or about 20000 cell classifications. In some embodiments, manual gating results in about 1000-20000, about 2000-20000, about 3000-20000, about 4000-20000, about 5000-20000, about 6000-20000, about 7000-20000, about 8000-20000, about 9000-20000, about 10000-20000, about 11000-20000, about 12000-20000, about 13000-20000, about 14000-20000, about 15000-20000, about 16000-20000, about 17000-20000, about 18000-20000, about 19000-20000 cell classifications. In some embodiments, manual gating results in about 1000-19000, about 1000-18000, about 1000-17000, about 1000-16000, about 1000-15000, about 1000-14000, about 1000-13000, about 1000-12000, about 1000-11000, about 1000-10000, about 1000-9000, about 1000-8000, about 1000-7000, about 1000-6000, about 1000-6000, about 1000-5000, about 1000-4000, about 1000-3000, about 1000-2000 cell classifications. In some embodiments, manual gating may be performed by an expert, e.g. an immunologist. In some embodiments, the cell classification may relate to cell types, cell subtypes, or cell states.

[0179] In some embodiments, trained machine learning models may be used to infer cell classification. In some embodiments, the models trained using the method described in US18 / 353,022, and corresponding U.S. Patent publication US2024-0192210-A1. In some embodiments, the models may be ensemble machine learning models. In some embodiments, the models are configured to process fluorescent intensity data and classify individual cells as belonging to one or a plurality of distinct immune cell sub-profiles.39MF-367064958Attorney Docket No.: 279362000640

[0180] In some embodiments, trained machine learning models are used to produce predictions of cell type or subtype (e.g., immune cell sub-population) for individual cell detection events and to determine cell counts (or frequencies) for each of a plurality of distinct cell types or subtypes. In some embodiments, the trained machine learning model uses a common hierarchy (a.k.a., a gating tree) to process the fluorescence profile data for each detected event and determine which and how many events belong to each measured populations (e.g., immune cell sub-population) in the hierarchy. In some embodiments, this may comprise over about 200 gates for the APC panel and over about 2000 gates for the T cell panel. The advantages of using this approach can be found in US18 / 353,022, and corresponding U.S. Patent publication US2024-0192210-A incorporated by reference in its entirety.

[0181] In some embodiments, a gating tree is constructed. In some embodiments, the gating tree comprises cell subsets. In some embodiments, the cell subsets are CD45+ (Leukocytes), CD3+ (T cells), CD4+ (Helper T cells), CD8+ (Cytotoxic T cells), CD19+ (B cells), CD14+ (Monocytes), CD56+ (Natural Killer cells), CD 16+ (Neutrophils), HLA-DR+ (Activated T cells), CD3-CD56+ (NK cells), CD3-CD19+ (Non-T / Non-NK B cells), CD3+CD16+ (NKT cells), CD4+CD45RA+ (Naive T cells), CD4+CD45RO+ (Memory T cells), CD8+CD45RA+ (Naive Cytotoxic T cells), CD8+CD45RO+ (Memory Cytotoxic T cells), CD14+HLA-DR+ (Activated Monocytes), CD16+CD45+ (Granulocytes), CD3-CD19-HLA-DR+ (Non-T / Non-B / Non-NK activated cells), CD3+HLA-DR+ (Activated T cells), CD3+CD4+FOXP3+ (regulatory T cells).

[0182] In some embodiments, the comprises cell classifications comprise summary ratio data for the cell classifications identified using the methods described herein. In some embodiments, the cell classifications may comprise immune cell populations. In some embodiments, the immune cell populations are distinct immune cell populations. In some embodiments, the immune cell populations are 20 distinct immune cell populations. In some embodiments, the immune cell populations are 20, 40, 60, 80, 100, 200, 300, 400, 500, 600, 700, 800, 900, 1000, 2000, 3000, 4000, 5000, 6000, 7000, 8000, 9000, 10000, 11000, 12000, 13000, 14000, 15000, 16000, 17000, 18000, 19000, or 20000 distinct immune cell populations.A. Cancers

[0183] The methods described herein relate to ICI treatments for cancer and sorting patients with cancer based on predicted responses to the ICI therapeutic. ICI therapeutics can be used to treat a wide range of cancers. The individuals revieing an ICI therapeutic, individuals in need40MF-367064958Attorney Docket No.: 279362000640thereof, candidate ICI recipients and / or ICI recipients as described herein have been diagnosed with cancer.

[0184] Cancer as described herein refers to any clinical diagnosis of cancer regardless of the cancer location or cell of origin. Exemplary cancers include but are not limited to carcinomas, sarcomas, leukemias, lymphomas, myelomas, or central nervous system cancers. In some embodiments, the cancer is a carcinoma such as adenocarcinoma, squamous cell carcinoma, basal cell carcinoma, Merkel cell carcinoma, or transitional cell carcinoma. In some embodiments, the cancer is a sarcoma such as osteosarcoma, liposarcoma, leiomyosarcoma, or angiosarcoma. In some embodiments, the cancer is a leukemia such as acute lymphoblastic leukemia, acute myeloid leukemia, chronic lymphocytic leukemia, chronic myeloid leukemia. In some embodiments, the cancer is a lymphoma such as Hodgkin lymphoma, non-Hodgkin lymphoma. In some embodiments, the cancer is a central nervous system such as glioma, medulloblastoma, astrocytoma, oligodendroglioma. In some embodiments, the cancer may be a virus associated cancer such as a cancer associated with HPV.

[0185] Additional exemplary cancers also may include but are not limited, bladder cancer, bone cancer, brain and spinal, cord tumors, breast cancer, cervical cancer, colorectal cancer, endometrial cancer, esophageal cancer, gallbladder cancer, gastric (stomach) cancer, head and neck cancer, Hodgkin lymphoma, kidney cancer, liver cancer, lung cancer, multiple, myeloma, non-Hodgkin lymphoma, ovarian cancer, pancreatic cancer, prostate cancer, skin cancer (nonmelanoma), soft tissue sarcoma, testicular cancer, thyroid cancer, or uterine cancer. .

[0186] In some embodiments, the cancer may include a cancer with an approved ICI therapy (e.g. FDA- approved). In some embodiments, the cancer may be non-small cell lung cancer (NSCLC), small cell lung cancer (SCLC), melanoma, renal cell carcinoma (kidney cancer), bladder cancer (urothelial carcinoma), head and neck squamous cell carcinoma (HNSCC), hepatocellular carcinoma (liver cancer), gastric cancer (stomach cancer), esophageal cancer, colorectal cancer (e.g. with MSI-H / dMMR mutations), endometrial cancer (e.g. with MSL H / dMMR mutations), Triple-Negative Breast Cancer (TNBC), Hodgkin’s Lymphoma (Classical Hodgkin Lymphoma), primary mediastinal large B-cell lymphoma, Merkel cell carcinoma, cutaneous squamous cell carcinoma (CSCC), cervical cancer, prostate cancer (e.g. with MSI-H / dMMR mutations), biliary tract cancer (Cholangiocarcinoma), thymic carcinoma, soft tissue sarcoma, and or mesothelioma.

[0187] In some embodiments, the cancer is melanoma, renal cell carcinoma, squamous cell carcinoma or Merkel cell carcinoma. In some embodiments, the cancer is Trippie negative breast cancer or bladder cancer.41MF-367064958Attorney Docket No.: 279362000640B. ICI therapy and irAEs

[0188] The methods described herein relate to ICI treatments for cancer and sorting individuals receipting ICI therapeutics or individuals in need thereof (e.g. patients with cancer) based on predicted responses to ICI. ICI therapy refers to therapeutics for treating cancer comprising immune checkpoint inhibition / blockade regimens either alone or combined with additional treatments. In some embodiments, the ICI therapy (e.g. ICI therapeutics) is administered as an advanced cancer therapy. In some embodiments, the ICI therapy is administered as an adjuvant therapy. In some embodiments, the ICI therapy is administered in cycles, e.g. 4 cycles.

[0189] In some embodiments, the ICI comprises administration of an anti-PDl therapy. In some embodiments, the anti-PDl therapy comprises Nivolumab, Pembrolizumab, or Cemiplimab. In some embodiments, the ICI comprises administration of an anti-PDLl therapy. In some embodiments, the anti-PDLl therapy comprises Avelumab or Durvalumab. In some embodiments, the ICI comprises administration of an anti-CTLA4 therapy. In some embodiments, the anti-CTLA4 therapy comprises Ipilimumab or Tremelimumab. In some embodiments, the ICI comprises administration of an anti-LAG3 therapy. In some embodiments, the anti-LAG3 therapy comprises Relatlimab.

[0190] In some embodiments, individuals who receive ICI therapy may experience one or more irAEs. irAEs development may reflect off-target impacts of immune activation by the ICI therapy. irAEs may be associated with improved outcomes for individuals who receive a treatment comprising an ICI therapeutic. ICI drug efficacy in activating an anti-cancer immune response may be linked to an auto-inflammatory / auto-immune reaction. In some embodiments, irAEs may result from an interaction of the ICI treatment (drug type, dose, regimen), immune system factors, genetics, and / or the microbiome.

[0191] The methods provided herein comprise machine learning models trained to predict and / or monitor the probability of an irAE. The machine learning models may be trained to predict and / or monitor the severity of an irAE. In some embodiments, the irAE may be a severe (e.g. high grade) irAE or a mild irAE (e.g. low grade, non-severe). In some embodiments, the models may be trained to predict onset of an irAE. The onset may be predicted as a probability of an irAE before a defined timepoint, such as any of the defined timepoints described herein.

[0192] In some embodiments, the irAE may be an irAE as described in Martins et al., Adverse effects of immune-checkpoint inhibitors: epidemiology, management and surveillance, Nat Rev. Oncol. (2019). An irAE may present in an individual who has received an ICI therapy as one or more of a dermatologic, gastrointestinal, endocrine system, pulmonary,42MF-367064958Attorney Docket No.: 279362000640Musculoskeletal, neurological, cardiovascular, or neurological symptom. In some embodiments, the dermatological symptoms comprise a rash itching or vitiligo. In some embodiments, the gastrointestinal symptoms include colitis, diarrhea or hepatitis. In some embodiments that endocrine symptoms may be Hypophysitis, Thyroiditis, adrenal insufficiency or diabetes mellitus. In some embodiments, the pulmonary symptom may be pneumonitis. In some embodiments, the musculoskeletal symptoms may be arthritis or myositis. In some embodiments, the neurological symptoms may be peripheral neuropathy, meningitis or encephalitis. Then somebody means the cardiovascular symptoms may be myocarditis or pericarditis. In some embodiments, the hematological symptoms may be Thrombocytopenia or hemolytic anemia. In some embodiments, the symptoms may contribute to organ failure and / or death. In some embodiments, the indication of the irAE refers to any of the symptoms described herein.

[0193] In some embodiments, the irAE is a severe (e.g. high grade) irAE. In some embodiments, a severe irAE may result in organ failure and / or death of the individual. In some embodiments, the irAE is a mild (e.g. low grade) irAE. In some embodiments, an irAE may be treated with or prevented by administration of an immunosuppressant drug. In some embodiments, the severity is based on Common Terminology Criteria for Adverse Events (CTCAE). In some embodiments, a severe irAE has a CTCAE of greater or equal to 3 and a non-sever irAE has a CTCAE of less than 3. Administration of an immunosuppressant drug may have a negative impact on the efficacy of the ICI therapy.

[0194] In some embodiments, the methods comprise use of one or more indications of an irAE for a plurality of ICI recipients as described herein. In some embodiments, the indication is a binary indication based on if the ICI recipient presented with an irAE at the respective timepoint. In some embodiments, the indication of the irAE is a symptom of an irAE as described herein.

[0195] In some embodiments, the indication of the irAE comprises a severity indication of the irAE. In some embodiments, the severity indication is severe or non-severe. In some embodiments, the severity indication of the irAE is a CTACE score. In some embodiments, a severe irAE has a CTACE of greater than or equal to 3. In some embodiments, a non-severe irAE has a CTACE of less than 3.

[0196] In some embodiments, more than one indication of an irAE comprise more than one indication of an irAE from each ICI recipient in the plurality of ICI recipients. The one or more indications may represent different characteristics of the irAE or be indications from different respective time points. For example, the one or more indication of the irAE for an ICI recipient 43MF-367064958Attorney Docket No.: 279362000640may comprise an indication of the irAE 1 day after treatment, 1 month after treatment, and / or 6 months after treatment. The different respective timepoint may be any of the timepoints described herein. In some embodiments, the time points may be at baseline or between cycles of the treatment with the ICI. In some embodiments, the timepoints corresponding to the irAE indication may be after two cycles of the ICI treatment and / or after treatment. In some embodiments of. Cycles of the ICI treatment are the cycles of ICI treatment for the Understanding ImmunE-related toXicities by multifACeTed profiling (EXACT) trial.

[0197] In some embodiments, multiple factors contribute to the onset of an irae. The multiple factors may comprise germline genetics, family history of autoimmune disease, circulating cytokines, T cell tolerance, B cell tolerance, fecal microbiome, ICI drug / ICI dose, tumor type, subclinical autoimmunity, or a personal history of autoimmune disease. In some embodiments, machine learning models, (e.g. adverse event model) described here in can be trained with additional information about the factors contributing to the onset of the irAE in the ICI recipient. Training the models with this information may aid in the predictions of an irAE and a candidate ICI recipient.VI. Positive Outcomes and Clinical outcomes

[0198] The methods provided herein comprise machine learning models trained to predict and / or monitor the probability of a positives outcome or a numerical quantification of a clinical outcome. The machine learning models may be trained to predict and / or monitor when a positive outcome or clinical outcome is likely to occur. The numerical quantification of the clinical outcome may represent a timepoint of when the clinical outcome occurs or is predicted to occur. As described herein, predicting and / or monitoring positive outcomes and clinical outcomes may be used for sorting candidate ICI recipients as responders or non-responders or for monitoring response to an ICI therapy.

[0199] In some embodiments, the positive outcome may relate to cancer stage, cancer progression, cancer severity, cancer remission, health-related quality of life, symptom management, a RECIST 1.1 response criteria or survival. In some embodiments, the positive outcome may be a reduction in cancer stage. In some embodiments, the positive outcome may be a decrease in cancer progression or cancer severity. In some embodiments, the positive outcome may be the maintenance of a cancer stage, cancer progression or cancer severity. In some embodiments, the positive outcome may an indication of progress for health-related quality of life or symptom management. In some embodiments, the positive outcome may be44MF-367064958Attorney Docket No.: 279362000640a positive change in one or more RECIST 1.1 response criteria. The RECIST 1.1 criteria are clinical parameters used to evaluate response to treatment for solid tumors. A positive change in one or more RECIST 1.1. response criteria may be determined by a clinician or pathologist. In some embodiments, the positive outcome may be survival.

[0200] In some embodiments, the positive outcome may be an immunological outcome, known in the art to relate to response to the ICI treatment. In some embodiments, the immunological outcome may related to immune activation. Immune activation can be achieved by modulation of the immune system, either by direct activation or through inhibiting immune suppression. Immune activation may be anti-cancer immune activation and may associated with clinical outcomes such as but no limited to reduction in tumor size. In some embodiments, the immune activation may be activation of one or more immune cells related to a response to the ICI therapeutic. In some embodiments, immune activation may be antigen-specific activation, for example as the abundance of activation marker expression (e.g. 4-1BB+, CD69+ as % of total) on surface of CD4 T cells, CD8 T cells. In some embodiments, immune activation may be measured as absolute lymphocyte count, CD8+ cells expressing activation or exhaustion markers such as PD-1, TIM-3, and LAG-3. In some embodiments, immune activation may be a measurement of T cell response and / or B cell response.

[0201] In some embodiments, the positive outcome may be an immunological outcome related to the ICI therapy. The positive outcome may be changes in expression or detectability of PD1, PDL1, CTLA4, and / or LAG3. In some embodiments, the immunological outcome related to the ICI therapy measure efficacy of the binding of the ICI therapeutic to cells. In some embodiments, the immunological outcome related to the ICI therapy may be a binding measurement of an anti-human igGl. In some embodiments, the anti-human igGl may be used to detect binding of an anti-PDl ICI therapeutic to cells from the candidate ICI recipient or ICI recipient from the plurality of ICI recipients.

[0202] In some embodiments, the indication of the positive outcome is a categorical indication of the outcome, such as presence or absence of the positive outcome. In some embodiments, the more than one indication of a positive outcome comprises more than one indications from each ICI recipient in the plurality of ICI recipients. The one or more indications may represent different characteristics of the outcome or be indications of the outcome from different respective time points. For example, the one or more indication of the positive for an ICI recipient may comprise an indication of the positive 1 day after treatment, 1 month after treatment, and / or 6 months after treatment. The different respective timepoint may be any of45MF-367064958Attorney Docket No.: 279362000640the timepoints described herein. In some embodiments, the respective timepoints may be before treatment with an ICI therapy, during treatment with an ICI therapy, or after treatment.

[0203] In some embodiments, the time points may be at baseline (e.g. before treatment) or between cycles of the treatment with the ICI. In some embodiments, the timepoints corresponding to the positive outcome indication may be after one cycle of the ICI treatment, after two cycles of the ICI treatment, after three cycles and / or after treatment. In some embodiments of. Cycles of the ICI treatment are the cycles of ICI treatment for the Understanding ImmunE-related toXicities by multifACeTed profiling (EXACT) trial.

[0204] In some embodiments, the methods comprise machine learning models trained to predict a numerical quantification of a clinical outcome. In some embodiments, a clinical outcome may be a diagnostic or prognostic outcome from a physical. The clinical outcome may relate to cancer stage, cancer progression, cancer severity, or cancer remission. In some embodiments, the clinical outcome is related to any of the positive outcomes described herein. In some embodiments, the clinical outcome relates to a RECIST 1.1 response criteria. In some embodiments, the clinical outcome may be a cancer related outcome metric used in clinical trials for assessing the effect of the ICI treatment. In some embodiments, the outcome may be a quality of life outcome, such as but not limited to an outcome related to symptom management or health-related quality of life. In some embodiments, the clinical outcome may be a change in a biomarker such as an immune biomarker.

[0205] In some embodiments, the numerical quantification of the clinical outcome is a continuous variable related to a clinical outcome. In some embodiments, the numerical quantification is expressed as a time increment. In some embodiments, the numerical quantification relates to a time since diagnosis or a time since treatment. In some embodiments, the numerical quantification is expressed in increments of hours, days, months, or years.

[0206] In some embodiments, the numerical quantification of the clinical outcomes relates to a time to a clinical outcome. In some embodiments, the numerical quantification of the clinical outcome is progression-free survival time (PFS), overall survival (OS) time, or time to a change in a biomarker.

[0207] In some embodiments, the one or more numerical quantification of a clinical outcome comprises more than one numerical quantification of a clinical outcome from each ICI recipient in the plurally of ICI recipients. The more that one numerical quantification of a clinical outcome may relate to more than one clinical outcome for the ICI recipient. This more than one numerical quantification may comprise numerical quantification of a clinical outcome obtained from an ICI recipient at different respective timepoints.46MF-367064958Attorney Docket No.: 279362000640

[0208] For example, the numerical quantification of the clinical outcome may be a numerical quantification of a clinical outcome 1 day after treatment, 1 month after treatment, and / or 6 months after treatment. The different respective timepoint may be any of the timepoints described herein. In some embodiments, the respective timepoints may be before treatment with an ICI therapy, during treatment with an ICI therapy, or after treatment.VII. Machine Learning Model Architectures and training methods

[0209] In some embodiments, the machine learning model (e.g. adverse event machine learning model, response machine learning model, multioutput machine learning model, numerical quantification machine learning model) is a time sensitive machine learning model. A time sensitive machine learning model may be a machine learning model capable of processing data associated with temporal metadata. In some embodiments, temporal metadata may related to when the data was collected or a different time based parameter. In some embodiments, the data associated with temporal metadata may comprise an indication of an irAE, an indication of positive outcome or a numerical quantification of a clinical outcome collected at two or more timepoints as described herein. In some embodiments, the temporal metadata is when a sample was collected, when an irAE presented and / or when a positive outcome presented. In some embodiments, quantitative data can be input onto the machine learning model as a numeric feature or as part of a time-encoding scheme (e.g., sinusoidal encodings for Transformers, or simply by feeding time deltas into an LSTM). In some embodiments, the outputs of multiple time-sensitive machine learning models are combined as described herein to sort a candidate ICI recipient as a responder or to monitor response to ICI therapy.

[0210] In some embodiments, the data associated with temporal metadata comprises an indication of an irAE, an indication of positive outcome, and / or a numerical quantification of a clinical outcome at different respective timepoints. In some embodiments, the respective time points are encoded into the time sensitive machine learning model using transformer-based architecture. In some embodiments, the transformer-based architecture comprises a temporal transformer. The transformer may process the data in a sequence to capture how features (e.g. cell classifications) evolve from one timepoint to the next.

[0211] In some embodiments, the time sensitive machine learning model may be a recurrent neural network (RNN). In some embodiments, the RNN comprises one or more hidden nodes. In some embodiments, the one or more hidden nodes can be configured as a gated recurrent47MF-367064958Attorney Docket No.: 279362000640unit (GRU), a long short-term memory (LSTM) or a combination thereof. In some embodiments, the RNN takes advantage of input data collected at two or more timepoints to enhance decision support for medical interventions, such as ICI therapy. In some embodiments, the RNN can discern patterns indicative of the onset of an outcome, such as an immune cell activation or an irAE. In some embodiments, the training of the RNN entails systematic adjustment of the network internal weights and biases to minimize prediction errors on training data, ensuring that the model accurately forecasts clinical metrics. It is appreciated that other models known in the art for processing data associated with temporal metadata data can be used. Such models may include but are not limited to AMIRA, exponential smoothing, and Temporal convolutional neural networks. RNN models and LSTM models may process data as a sequence and capture how features (e.g. cell classifications) evolve from one timepoint to the next.

[0212] In some embodiments, the time sensitive machine learning model comprises a feature extraction model and a longitudinal data model. In some embodiments, the longitudinal data model comprises a RNN, LSTM, a transformer, or a GRU model as described herein. In some embodiments the feature extraction model comprises a convolutional neural network (CNN) or a graphical neural network (GNN). In some embodiments, the extracted features may be fed into the longitudinal data model. In some embodiments, a feed-forward network may be used to incorporate additional training data to improve performance.

[0213] In some embodiments, the time sensitive machine learning model is a classification model. A classification model can be used to predict the probability of an irAE or a probability of a positive outcome as described herein. The predicted probability may be related to the confidence of a classification task in the classification model, such as classification of the candidate ICI recipient cell classifications as more similar to cell classifications from ICI recipients (in the training data) who experienced an irAE. In some embodiments, the time sensitive machine learning model is a regression model. A regression model may be used to predict a numerical outcome, such as a numerical quantification of a clinical outcome such as overall survival (OS).

[0214] In some embodiments, the time sensitive machine learning model is able to take in cell classification collected from the candidate ICI recipient at multiple time points. In some embodiments, the time sensitive machine learning model can generate updated predictions based on the input of cell classifications from new time points. In some embodiments, the time sensitive machine learning model reuses many of the same layers as a time sensitive machine learning model that takes in cell classifications from a single time point while also including 48MF-367064958Attorney Docket No.: 279362000640additional recurrent or transformation layers for the input temporal data. An RNN or transformer may be used to handle variable-length sequences between the timepoints. In some embodiments, the time sensitive machine learning model may output a trajectory based prediction. In some embodiments, a trajectory based output may comprise an average of the outputs of the time sensitive machine learning model for multiple timepoints. In some embodiments, an attention mechanism over hidden representations of features (e.g. cell classification) from samples at multiple time points may be used to generate a trajectory based output.

[0215] In some embodiments, the time sensitive machine learning model is a multioutput machine learning model. The multioutput machine learning model comprises a multi-task architecture. In some embodiments, the muti-task architecture allows the machine learning model to output a predicted probability of an irAE and a predicted probability of a positive outcome. In some embodiments, the multioutput machine learning model outputs a classification (predicted probability of an irAE / positive outcome) and a regression result (numerical quantification of a clinical outcome). In some embodiments, the multi-task architecture outputs multiple responses from a same network or different heads of a same network. In some embodiments, the multi-task implicitly “combines” the different tasks’ learned features by sharing the initial layers and specialized “heads” for each task. The final prediction may be a single prediction influenced by hidden layers trained for an alternative prediction.

[0216] In some embodiments, the weighting factors, bias values, and threshold values, or other computational parameters of the model (e.g. neural network), can be “taught” or “learned” in a training phase using one or more sets of training data. For example, the parameters may be trained using the input data from a training data set and a gradient descent or backward propagation method so that the output value(s) that the neural network predicts are consistent with the examples included in the training data set. The adjustable parameters of the model may be obtained using, e.g., a back propagation neural network training process.

[0217] In some embodiments, the time sensitive machine learning model comprises a plurality of nodes. In some embodiments, the plurality of nodes (i.e., the number of individual machine learning models in the ensemble) comprises at least 1000, 1200, 1400, 1600, 1800, 2000, 2200, 2400, 2600, 2800, 3000, 3200, or 3400 nodes.

[0218] In some embodiments, the time sensitive machine learning model may suffer from overfitting. In some embodiments, hard-parameter sharing may be used to help with mitigation of loss inherent to overfitting models. In some embodiments, hard-parameter sharing may be 49MF-367064958Attorney Docket No.: 279362000640implementing a deep neural network in which the deeper hidden layers are shared between all tasks (which learn and simultaneously reduce the dimensionality of the information contained in the features), while each target (or biologically-informed group of targets) has dedicated output layers in the network architecture which serve to pseudo-independently predict the expected clinical metrics from the encoded information of the hidden layers.

[0219] In some embodiments, a convolutional approach may be employed to capture the inter dependencies. Convolutional neural networks have been shown to be a powerful approach in image analysis, in which pixel-to-pixel correlations are inherent and form shapes that constitute the meaning of the image. CNNs have also been adapted to non-image data, in which the order of the "pixels" (or features and samples in ML nomenclature) do not contain useful information. These approaches require the network to be made agnostic to the order of the input data.

[0220] In some embodiments, a multilevel Mixture of Experts approach may be employed where a subset of previously trained models can be used to vote, and a second later discriminatory machine learning model makes the final prediction.

[0221] In some embodiments, the time sensitive machine learning model is compiled with an optimized with an optimizer. In some embodiments, the optimized by adjusting the network’s weights to minimize the categorical cross-entropy loss function. In some embodiments, the optimizer is Adam.

[0222] In some embodiments, the data used to train the time sensitive machine learning model may be normalized before it is used to train the time sensitive machine learning model. In some embodiments, normalizing the data comprises ensuring the data have similar scales. In some embodiments, normalizing the data comprises one-hot-encoding any categorical variables.

[0223] In some embodiments, the time sensitive machine learning models comprise data transformation layers within a network. The transformation layers may be used for dimensionality reduction of normalization of the features (e.g. cell classifications) before / or within a temporal layer of this model. In some embodiments, the transformation layers comprise autoencoder modules or attention layers.

[0224] In some embodiments, training the time sensitive machine learning model comprises feeding the data used to train the model into the network, allowing it to adjust its weights over multiple epochs to minimize the loss function. In some embodiments, each epoch represents a full pass through the training dataset. In some embodiments, a validation set is used concurrently to monitor the model’s performance and adjust hyperparameters, such as the learning rate and the number of epochs. In some embodiments, early stopping is implemented 50MF-367064958Attorney Docket No.: 279362000640to prevent overfitting; training is halted if the validation loss does not improve over several epochs.

[0225] In some embodiments, training the machine learning model (e.g. adverse event machine learning model, response machine learning model) comprises logistic regression based feature selection. Accordingly, In some embodiments, the machine learning models are referred to as logistic regression machine learning models.

[0226] FIG. 5 provides an exemplary process for training and testing a logistic regression machine learning model as described herein. In some embodiments, the methods and systems comprise, data preparation, 502-508, feature selection 510-516, model cross validation (cv), 522-526 and 532-536, and performance assessment, 528-530 and 540-542. The blocks with dotted outlines in the system represent optional steps in the workflow.

[0227] In some embodiments, at 502, data preparation can begin with subsetting the samples, 504, (i.e., ICI recipients in training data) to include only those relevant to the prediction task, based on sampling timepoints and the availability of outcome information. At 506, relevant features (i.e., cell classifications) can be identified from the immunophenotyping data from those samples. In some embodiments, filtering can be applied to retain only features derived from a mean of more than 15 cell events. Each individual cell event may represent a single cell quantified as it passed through a flow cytometer. In some embodiments, filtering features captured from few cell events can reduce noise and ensure stability in predictive models. In some embodiments, 506, comprises removing features with low variance or high intercorrelation to reduce redundancy and noise. The resulting test dataset, 508, can be used for univariate feature selection at 510, and in the test and permutation phase of cross validation at 522 and 532.

[0228] In some embodiments, feature selection is performed using bootstrap resampling. In some embodiments, feature selection is performed using 100 bootstrap resamples. Within each resample, at 510, univariate feature selection can be conducted using multiple t-tests or linear models against the outcome (e.g., response to ICI). At 512, the top subset of features can be selected. In some embodiments, the top subset of features in 512 are defined as 10% of the number of available samples. In some embodiments, selecting the top features can minimize overfitting. At 514, stepwise logistic regression can be applied to the selected features from 512. In some embodiments, stepwise logistic regression is used to identify a most informative subset of features, 516. Following all of the bootstrap resampling iterations, at 518, informative features can be aggregated across bootstraps according to their frequency of selection, and the top stable candidates, 520, can be carried forward for modelling.51MF-367064958Attorney Docket No.: 279362000640

[0229] In some embodiments, model performance can be assessed through cross-validation (CV) with permutation testing. In some embodiments, CV comprises 100 iterations with a 80:20 data spilt for training and testing. At 522 the test data from 508 is split randomly into 80:20 train and test data. At 524, a stepwise logistic regression model can be fit to the training set using the previously selected top selected features, 520, and performance can be evaluated on the held-out test set, 526. Following the 100 test iterations, the performance metrics can be aggregated at 528. Aggregating performance metrics at 528, may comprise calculating balanced accuracy and ROC-AUC, and then quantifying the mean and standard deviation of these metrics across all iterations. Models with ROC-AUC values closest to the mean of the 100 iterations can be selected as representative at 530.

[0230] For permutation testing, a procedure can be repeated using randomly selected features and shuffled outcome labels, providing a baseline for comparison for the representative test model, 530. At 532, labels and features are shuffled from the test data 508. Processes 534, 536, and 538 can be performed according to the same processes at 522, 524, and 526. At 540, the performance metrics for the permutation iterations can be aggregated and a representative permutation model can be created at 542 following the processes of 528 and 530. In some embodiments, model quality for the representative test model, 530, can be assessed by comparing the performance metrics to the representative permutation model, 542.

[0231] FIG. 6 provides an exemplary process for training and testing a logistic regression machine learning model as described herein. In some embodiments, the methods and systems are logistic regression machine learning models are predictive models for predicting patient outcomes from flow cytometry immune phenotyping data as described herein. In some embodiments, to account for limited sample size and imbalanced outcome classes in the data, the methods comprise Monte Carlo cross validation to generate stratified repeated train-test splits. In some embodiments, for each training split, optional hyperparameter tuning is performed, with internal cross validation, followed by selection of the most appropriate parameters to build a final model, fitted to the full training data using a small number of immunophenotyping machine learning output features as described herein. Validation of the model can be performed by applying the model to unseen test data for evaluation of classification accuracy, as measured through AUC and balanced accuracy. In some embodiments, after all train-test splits are evaluated, a distribution of accuracy metrics is generated, yielding an estimation of model performance on unseen data. In some embodiments, a permutation approach is employed, where outcome labels are randomized in order to generate52MF-367064958Attorney Docket No.: 279362000640an expected null distribution for assessing model performance. The blocks with dotted outlines in the system represent optional steps in the workflow.

[0232] In some embodiments, cell classifications inferred from fluorescent intensity data is used as input data, 604. In some embodiments, the cell classifications inferred from fluorescent intensity data is transformed using a centered-log ratio transformation, 602. In some embodiments, the input data is subset by individual samples based on the prediction choice, sample timepoint, and available outcome information that was relevant for the respective predictive model, 606. In some embodiments, at 608 a log Ip transformation is performed.

[0233] At 610, monte carlo cross-validation / holdout is performed, where k (e.g., 50) repeated stratified train-test splits are generated (612 and 614 respectively), then the below steps are performed on each split, resulting in AUC and balanced accuracy metrics of length k.

[0234] In some embodiments, hyperparameter selection methods, 616, are performed. A randomized search over logistic regression hyperparameters is conducted. For each hyperparameter selection, a 3-fold cross validation was repeated twice i.e. 6 total fits for each hyperparameter selection (618), In some embodiments, the 3-fold cross validation comprises scaling the data, 620, performing feature selection using L2 regularized logistic regression model. 622. In some embodiments, a maximum of 25 features are selected. At 624, a logistic regression model is fit using the selected features and hyperparameter value. In some embodiments, the best model is extracted at 626 and all hyperparameter values are evaluated, In some embodiments, the models with the highest mean AUC value across validation splits are taken forward to be used in the final model (628).

[0235] In some embodiments, the hyperparameter selection methods are not performed and the model is trained without tuning. In some embodiments, the training data are scaled, 630, feature selection is performed using L2 regularized logistic regression, 632, and the logistic regression model is fit, 634.

[0236] The best fit model, 624, was used to predict outcomes in the testing data (632).Probability estimates and model predictions were extracted for each sample, which were then compared to the true labels to generate AUC and balanced accuracy metrics. The evaluation metrics were combined from each monte carlo split to generate a distribution which was used to estimate model stability.

[0237] In some embodiments, the methods comprise evaluating the logistic regression machine learning models from 628 and / or 634 using the testing data. In some embodiments, evaluating the logistic regression machine learning models comprise using the logistic regression machine53MF-367064958Attorney Docket No.: 279362000640learning modes from 628 and / or 634 to predict outcomes in the testing data, probability estimates and model predictions are extracted for each sample, which are then compared to the true labels to generate AUC and balanced accuracy metrics. In some embodiments, evaluation metrics are combined from each monte carlo split to generate a distribution, useful in estimating model stability.

[0238] Logistic regression machine learning models as described herein (e.g., training according to the processes in FIG. 5 and / or FIG. 6) can be beneficial for predicting probabilities of response and / or adverse events following ICI. Logistic regression models are flexible and can be designed for binary outcomes e.g. responder / non-responder, estimating the probability of the outcome. In some embodiments, logistic regression models can effectively incorporate and may not be negatively affected by additional covariates which may independently be associated with clinical outcomes such as but not limited to cancer type, age, and sex.

[0239] In some embodiments, logistic regression model as described herein can be used to effectively select features from feature groups that may be larger than the number of training samples. In some embodiments, threshold, correlation and variance filtering of features can be used to limit features to the minimal informative set ahead of testing. In some embodiments, univariate screening can be used to eliminate the bulk of features by testing vs. outcome. In some embodiments, a stepwise approach to logistic regression can be used to only retains features with additional informative power. In some embodiments, bootstrap aggregation can be used to ensure selected features are stable across resamples of the data. In some embodiments, boostrap aggregation can also be used to reduce selection of noisy features, thus controlling overfitting and / or minimizing false discovery.

[0240] In some embodiments, overfitting can be controlled and / or false discovery can be minimized by performing permutation testing. In some embodiments, permutation testing comprises quantifying test performance above random fluctuations. In some embodiments, various performance metrics can be tested and applied to the logistic regression model training workflow to ensure the models are robust to class imbalance.

[0241] In some embodiments, the logistic regression machine learning models described herein can be beneficial for predicting probabilities of response and / or adverse events following ICI because all available data can be used in training the models. In some embodiments, by bootstrapping feature selection, all available data can be accessed in training. In some embodiments, training the logistic regression machine learning models described 54MF-367064958Attorney Docket No.: 279362000640herein comprise test and permutation iterations to evaluate model performance across multiple partitions and ensure robustness to small sample sizes.

[0242] In some embodiments, the logistic regression machine learning models described herein can be beneficial for predicting probabilities of response and / or adverse events following ICI because the feature sets used for inference are interpretable. In some embodiments, the training pipelines comprise reproducibility checks to ensure stability. In some embodiments, performance aggregation over multiple test and permute iterations can be used to quantify stability and enable selection of the most representative model to take forward.

[0243] In some embodiments, the machine learning model (e.g. adverse event machine learning model, response machine learning model) described herein comprises an ensemble model. In some embodiments, the ensemble model is an Adaptive Best Subset Ensemble (ABSS) model.

[0244] In some embodiments, the ABSS models described herein can be beneficial for predicting probabilities of response and / or adverse events following ICI because of the high dimensionality of the cell classification data used for training and inference. The ABSS models described herein prevent data leakage in training, validate interactions by performing train / test consistency checks during interaction term creations, allow for unbiased evaluation based on reserved data in training.

[0245] In some embodiments, ABSS is an iterative, multilayer classifier specifically designed for identifying signatures within flow cytometry data, particularly when complex interactions and correlations, such as high-dimensional immune profile datasets, do not follow the assumptions of many conventional classification methodologies. In some embodiments the methodology, integrated into a comprehensive pipeline can addresses critical limitations of traditional feature selection methods by implementing intelligent sampling strategies, rigorous multistep validations, and adaptive learning mechanisms optimized for biomedical applications. In some embodiments, ABSS can be designed to include and account for additional clinical or demographic measurements, e.g. age, sex, and comorbidities.

[0246] In some embodiments, training an ABSS model comprises iteratively identifying optimal feature combinations through intelligent weighted sampling of immune profiles (e.g., cell classifications as described herein) to delineate between samples based on the outcome of interest (e.g., response and / or irAE). In some embodiments, the selected features subsequently form ensemble blocks that undergo a comprehensive validation approach, employing shuffled outcome labels and stability assessment across multiple data partitions for further refinement.55MF-367064958Attorney Docket No.: 279362000640In some embodiments, a set of blocks with optimal risk scores can be selected for external validation.

[0247] In some embodiments, training an ABSS model comprises data preparation, multilayer feature selection, ensemble model election, and validation based on unseen data. FIG. 7 provides an exemplary process for training and an ABSS model as described herein. Input dataset, 702, comprises cell classification data as described herein for patients samples. In some embodiments, samples from the input dataset, 702, are processed at 704. In some embodiments, sample processing comprises one or more user-defined preprocessing operations to select relevant samples according to predefined criteria, such as measurement quality or association with clinical outcomes. In some embodiments, the features are also preprocessed at 706. In some embodiments, feature preprocessing comprises one or more statistical filtering operation is performed to remove features lacking sufficient reliability. Such filtering may include removal of features with minimal variance, excessive sparsity, negligible coefficient of variation between outcome groups, and high correlation with redundant variables. In some embodiments, sample preprocessing comprises a synthetic feature creation operation is executed to generate interaction terms among features. Candidate interaction terms can be statistically evaluated, for example by applying t-tests, and a subset of meaningful interaction terms can be retained in the test dataset at 708. At 710, the test dataset from 708 is partitioned into subsets. The subsets comprise the training set, 712, which may comprise about 75% of the samples in the test dataset. The test set, 714, which may comprise the about 25% of the samples can be further divided into internal training and internal test subsets; a reserve subset, 716, employed for ensemble model selection; and a holdout subset, 718, reserved exclusively for final validation and not utilized during training. In some embodiments, the reserve set from 716 can be assessed for use in training at 720 and can either be included in for feature selection at 722 or reserved for model selection at 724.

[0248] At 722, multi-layer feature selection can then be applied to the training set, 712, optionally including the reserve set, 716, and using an initial weighted random selection of features. In some embodiments, the initial weights are determined based on mutual information between features and the outcome of interest, which are then iteratively updated in response to performance outcomes, using parameters including a bad threshold, a white threshold, a reward value, and a penalty value. In some embodiments, multi-layer feature selection, 722, comprises several iterations within each layer. In some embodiments, at each iteration, the selected features are evaluated by training a lightweight classifier, such as a Logistic Regression (LR) or light weight support vector classifier (SVC), on an internal training subset and testing 56MF-367064958Attorney Docket No.: 279362000640performance on an internal validation subset. The SVC can be replaced with a different machine learning model, such as but not limited to a neural network. A composite balanced accuracy score is calculated using both internal training and test outcomes. The score is compared against the bad threshold and white threshold. If the score exceeds the white threshold, the associated feature weights are increased by the reward value, thereby increasing the likelihood of reselection in subsequent iterations. Conversely, scores below the bad threshold result in a penalty being applied to the weights, reducing the likelihood of reselection. At 722, repetition of this process over thousands of iterations across multiple layers produces a refined feature set optimized for classification performance.

[0249] In some embodiments, strategies are employed to mitigate overfitting. In some embodiments, method incorporates a seed update threshold parameter together with batchbased optimization. In some embodiments, these adjustments enable dynamic reconfiguration of internal training and testing subsets without altering reserve or holdout sets, thereby preventing overfitting to a single partition. In certain implementations, the ABSS training comprises down sampling of training data. In some embodiments, down sampling training data is advantageous when the dataset comprises thousands of samples, as it can substantially reduce computational time without materially compromising predictive performance.

[0250] The output of each layer of 722 comprises a set of models and, the features integrated into them. In some embodiments, three approaches are used to identify important features from a given layer: top N, drop point, and weighted drop point. The “top N” method selects the N top models based on their scores (balanced accuracy) and then identifies the N most frequent features. The “drop point” method first determines the top models exhibiting maximum variation and subsequently selects the N most frequent features. The “weighted drop point”, default, method is similar to the drop point method, but features are weighted by their importance scores.

[0251] The outcome models from each layer of feature selection, 726, can be input for an evaluation pipeline, 728. In some embodiments, the evaluation pipeline comprises testing the output models, 726, across multiple splits of the training set, 712, as well as on dedicated reserve and holdout sets, 724. In some embodiments, tests are performed using randomly permuted labels (i.e., a permutation of clinical outcomes). In some embodiments, the evaluation process enables the calculation of a risk score for each model, derived from metrics obtained from the (internal) training, (internal) test, reserve, and permuted sets (excluding the holdout set). The outcome model, 726, can be ranked by their lowest risk score and subsequently exported as ensemble blocks, 730. The outcome model, 726, has a flexible architecture. In 57MF-367064958Attorney Docket No.: 279362000640some embodiments, the outcome model, 726 comprises a support vector classifier (SVC) model, a logistic regression model, or a multilayer perceptron (MLP) model.

[0252] In some embodiments, inference using the ABSS model comprises ensemble prediction based on the models in the ensemble blocks. In some embodiments, a final prediction is based on majority, average, or weighted vote. In some embodiments, majority vote prediction comprises converting each model’s probability from the ensemble models into a binary vote using a threshold, predicting 1 if at least 50% of the models vote 1. In some embodiments, a final prediction is based on an average probability. In some embodiments, the average probability is based on averages the probabilities of classes from individual predictors. The average probability can be compared to a decision threshold for classification. In some embodiments, a weighted votes strategy can be used. In some embodiments, the probabilities from each of the ensemble models are transformed into a logit-margin score around a threshold. The scores are then summed across models, and a classification can be made based on the sum, for example a classification into one class if the sum is greater than or equal to 0.

[0253] The stability of an ABSS models described herein can be assessed. Stability can be calculated by randomly partitioned the training dataset into S splits (default: 200). The ensemble model generated by ABSS training is then trained independently on each of these splits, resulting in 200 trained models stored on disk. Subsequently, these models are applied to the reserved and holdout sets, and the balanced accuracy is computed.VIII. Systems

[0254] In some instances, the systems may comprise e.g., one or more processors, and a memory unit communicatively coupled to the one or more processors and configured to store instructions that, when executed by the one or more processors, cause the system to perform any of the embodiments disclosed herein.

[0255] Similarly, non-transitory computer-readable storage media are disclosed that may comprise instructions for operating a system configured to perform any of the disclosed methods for sorting a candidate ICI recipient as a responder or a non-responder, monitoring response to an ICI therapy in a candidate ICI recipient or for training a machine learning model to predict the probability of an irAE, positive outcome or numerical quantification of a clinical outcome as described herein.58MF-367064958Attorney Docket No.: 279362000640A. Computer Processors & Systems

[0256] Provided herein are non-transitory computer-readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by one or more processors of a system, cause the system to perform any of the methods described herein.

[0257] FIG. 8 illustrates an exemplary computing system, in accordance with some implementations. Computing system 800 can be a component of a system for sorting a candidate ICI recipient as a responder or a non-responder, a system for monitoring response to ICI therapy in a candidate ICI recipient or a system for training a machine learning model to predict the probability of an irAE, positive outcome or numerical quantification of a clinical outcome as described herein.

[0258] Computing system 800 can include a host computer connected to a network. Computing system 800 can be a client computer or a server. As shown in FIG. 8, computing system 800 can comprise any suitable type of microprocessor-based device, such as a personal computer; workstation; server; or handheld computing device, such as a phone or tablet. The computer can include, for example, one or more of processor 810, input device 820, output device 830, memory storage 840, and communication device 860.

[0259] Input device 820 can be any suitable device that provides input, such as a touch screen or monitor, keyboard, mouse, or voice-recognition device. Input device 820 can be configured to receive one or more use defined threshold. The user defined threshold may be any of the predetermined thresholds as described herein. Output device 830 can be any suitable device that provides output, such as a touch screen, monitor, printer, disk drive, or speaker.

[0260] Memory storage 840 can be any suitable device that provides storage, such as an electrical, magnetic, or optical memory, including a RAM, cache, hard drive, CD-ROM drive, tape drive, or removable storage disk. Communication device 860 can include any suitable device capable of transmitting and receiving signals over a network, such as a network interface chip or card. The components of the computer can be connected in any suitable manner, such as via a physical bus or wirelessly. Memory storage 840 can be a non-transitory computer-readable storage medium comprising one or more programs, which, when executed by one or more processors, such as processor 810, cause the one or more processors to execute any of the methods described herein.

[0261] Software 850, which can be stored in memory storage 840 and executed by processor 810, can include, for example, the programming that embodies the functionality of the present59MF-367064958Attorney Docket No.: 279362000640disclosure (e.g., as embodied in the methods, systems, computers, servers, and / or devices as described above). In some embodiments, software 850 can be implemented and executed on a combination of servers such as application servers and database servers.

[0262] Software 850 can also be stored and / or transported within any computer-readable storage medium for use by or in connection with an instruction execution system, apparatus, or device, such as those described above, that can fetch and execute instructions associated with the software from the instruction execution system, apparatus, or device. In the context of this disclosure, a computer-readable storage medium can be any medium, such as storage 840, that can contain or store programming for use by or in connection with an instruction execution system, apparatus, or device.

[0263] Software 850 can also be propagated within any transport medium for use by or in connection with an instruction execution system, apparatus, or device, such as those described above, that can fetch and execute instructions associated with the software from the instruction execution system, apparatus, or device. In the context of this disclosure, a transport medium can be any medium that can communicate, propagate, or transport programming for use by or in connection with an instruction execution system, apparatus, or device. The transport-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, or infrared wired or wireless propagation medium.

[0264] Computing system 800 may be connected to a network, which can be any suitable type of interconnected communication system. The network can implement any suitable communications protocol and can be secured by any suitable security protocol. The network can comprise network links of any suitable arrangement that can implement the transmission and reception of network signals, such as wireless network connections, T1 or T3 lines, cable networks, DSL, or telephone lines.

[0265] Computing system 800 can implement any operating system suitable for operating on the network. Software 850 can be written in any suitable programming language, such as C, C++, Java, or Python. In various embodiments, application software embodying the functionality of the present disclosure can be deployed in different configurations, such as in a client / server arrangement or through a Web browser as a Web-based application or Web service, for example.60MF-367064958Attorney Docket No.: 279362000640EXEMPLARY EMBODIMENTS

[0266] Embodiment 1. A method for sorting a candidate immune checkpoint inhibition (ICI) recipient as a responder or non-responder, comprising:fluorescently labeling cells contained within a sample from the candidate ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel,generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer;inferring cell classifications based on the fluorescent intensity data;providing at least a subset of the cell classifications as input to an adverse event machine learning model, wherein the adverse event machine learning model has been trained to predict a probability of an immune-related adverse event (irAE) with training data comprising at least:(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints,(b) one or more indications of the irAE for the plurality of ICI recipients at respective timepoints;outputting from the adverse event machine learning model a predicted probability of the irAE for the candidate ICI recipient; andsorting the candidate ICI recipient as a responder or a non-responder based at least on the predicted probability of the irAE.

[0267] Embodiment 2. The method of embodiment 1, wherein the one or more indication of the irAE comprise one or more severity indications of the irAE.

[0268] Embodiment 3. The method of embodiment 2, wherein the severity indication of the one or more severity indications of the irAE is based on Common Terminology Criteria for Adverse Events (CTCAE).

[0269] Embodiment 4. The method of embodiment 3, wherein the severity indication of the irAE is severe if the CTCAE greater than or equal to 3 and the severity indication of the irAE is non-severe if the CTCAE is less than 3.61MF-367064958Attorney Docket No.: 279362000640

[0270] Embodiment 5. The method of any of embodiments 1-4, wherein the respective timepoints for the one or more indications of the irAE are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0271] Embodiment 6. The method of any of embodiments 1-5, wherein the sample is a first sample from a first timepoint.

[0272] Embodiment 7. The method of embodiment 6, wherein the method comprises inferring cells classification for a second sample at a second timepoint by:fluorescently labeling cells contained within a second sample from the candidate ICI recipient, by contacting at least an aliquot of the second sample with at least one immunophenotyping fluorescent labeling panel,generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer;inferring cell classifications based on the fluorescent intensity data.

[0273] Embodiment 8. The method of embodiment 7, comprising providing at least a subset of the cell classifications from the second sample as input to the adverse event machine learning model.

[0274] Embodiment 9. The method of embodiment 8, wherein the adverse event machine learning model has been trained to average over the hidden representations of the first sample and the second sample or applies an attention mechanism over the hidden representations of the first sample and the second sample to predict a trajectory-based prediction.

[0275] Embodiment 10. The method of any of embodiments 1-9, wherein the candidate ICI recipient is sorted as a responder if the predicted probability of the irAE is less than a predetermined adverse event threshold and sorted as a non-responder if the predicted probability of the irAE is greater than the predetermined adverse event threshold.

[0276] Embodiment 11. The method of embodiment 10, wherein the predetermined adverse event threshold is about 25%, about 50%, about 75%, about 95%, or about 99%.

[0277] Embodiment 12. The method of any of embodiments 1-11, wherein the adverse event machine learning model is a time sensitive machine learning model.

[0278] Embodiment 13. A method for sorting a candidate immune checkpoint inhibition (ICI) recipient as a responder or non-responder, comprising:fluorescently labeling cells contained within a sample from the candidate ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel,62MF-367064958Attorney Docket No.: 279362000640generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer;inferring cell classifications based on the fluorescent intensity data;providing at least a subset of the cell classifications as input to a response machine learning model, wherein the response machine learning model has been trained to predict the probability of a positive outcome with training data comprising at least:(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints;(b) one or more indications of the positive outcome for the plurality of ICI recipients at respective timepoints;outputting from the response machine learning model, a predicted probability of the positive outcome for the candidate ICI recipient; andsorting the candidate ICI recipient as a responder or a non-responder based at least on the predicted probability of the of positive outcome.

[0279] Embodiment 14. The method of embodiment 13, wherein the positive outcome relates to cancer stage, cancer progression, cancer severity, cancer remission, health-related quality of life, symptom management, a RECIST 1.1 response criteria or survival.

[0280] Embodiment 15. The method of embodiment 13 or 14, wherein the respective timepoints for the one or more indications of the positive outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0281] Embodiment 16. The method of any of embodiments 13-15, wherein the sample is a first sample from a first timepoint.

[0282] Embodiment 17. The method of embodiment 16, wherein the method comprises inferring cells classification for a second sample at a second timepoint by:fluorescently labeling cells contained within a second sample from the candidate ICI recipient, by contacting at least an aliquot of the second sample with at least one immunophenotyping fluorescent labeling panel,generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer;inferring cell classifications based on the fluorescent intensity data.

[0283] Embodiment 18. The method of embodiment 17, comprising providing at least a subset of the cell classifications from the second sample as input to the response machine learning model.63MF-367064958Attorney Docket No.: 279362000640

[0284] Embodiment 19. The method of embodiment 18, wherein the response machine learning model has been trained to average over the hidden representations of the first sample and the second sample or applies an attention mechanism over the hidden representations of the first sample and the second sample to predict a trajectory-based prediction.

[0285] Embodiment 20. The method of any of embodiments 13-19, wherein the candidate ICI recipient is sorted as a responder if the predicted probability of the positive outcome is greater than a predetermined positive outcome threshold and sorted as a non-responder if the predicted probability of the positive outcome is less than the predetermined positive outcome threshold.

[0286] Embodiment 21. The method of embodiment 20, wherein the predetermined positive outcome threshold is about 25%, about 50%, about 75%, about 95%, or about 99%.

[0287] Embodiment 22. The method of any of embodiments 13-21, wherein the response machine learning model is a time sensitive machine learning model.

[0288] Embodiment 23. A method for sorting a candidate immune checkpoint inhibition (ICI) recipient as a responder or non-responder, comprising:fluorescently labeling cells contained within a sample from the candidate ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel,generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer;inferring cell classifications based on the fluorescent intensity data;providing at least a subset of the cell classifications as input to an adverse event machine learning model and a response machine learning model;wherein the adverse event machine learning model has been trained to predict the probability of an immune-related adverse event (irAE) with a first set of training data comprising at least:(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints,(b) one or more indications of the irAE for the plurality of ICI recipients at respective timepoints;wherein the response machine learning model has been trained to predict the probability of a positive outcome with a second set of training data comprising at least:(a) the plurality of cell classifications,64MF-367064958Attorney Docket No.: 279362000640(b) one or more indications of the positive outcome for the plurality of ICI recipients at respective timepoints;outputting from the response machine learning model, a predicted probability of the positive outcome and from the adverse event machine learning model, a predicted probability of the irAE for the candidate ICI recipient; andsorting the candidate ICI recipient as a responder or a non-responder based at least on the predicted probability of the of positive outcome and the predicted probability of the irAE.

[0289] Embodiment 24. The method of embodiments 21, wherein sorting the candidate ICI recipient comprising combining the predicted probability of the positive outcome and the predicted probability of the irAE using voting, stacking, averaging, or Logit blending methods.

[0290] Embodiment 25. The method of embodiment 23 or 24, wherein the respective timepoints for the one or more indications of the irAE are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0291] Embodiment 26. The method of any of embodiments 23-25, wherein the respective timepoints for the one or more indications of the positive outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0292] Embodiment 27. The method of any of embodiments 23-26 wherein the adverse event machine learning model and the response machine learning model are time sensitive machine learning models.

[0293] Embodiment 28. A method for sorting a candidate immune checkpoint inhibition (ICI) recipient as a responder or non-responder, comprising:fluorescently labeling cells contained within a sample from the candidate ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel,generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer;inferring cell classifications based on the fluorescent intensity data;providing at least a subset of the cell classifications as input to a multioutput machine learning model, wherein the multioutput machine learning model has been trained to predict the probability of a positive outcome and / or a probability of an immune-related adverse event (irAE) with training data comprising at least:65MF-367064958Attorney Docket No.: 279362000640(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints;(b) one or more indications of the positive outcome for the plurality of ICI recipients at respective timepoints;(c) one or more indications of the irAE for the plurality of ICI recipients at respective timepoints;outputting from the multioutput machine learning model, a predicted probability of the positive outcome and / or a predicted probability of the irAE for the candidate ICI recipient; andsorting the candidate ICI recipient as a responder or a non-responder based at least on the predicted probability of the of positive outcome and / or the predicted probability of the irAE.

[0294] Embodiment 29. The method of any of embodiments 23- 28, wherein the one or more indication of the irAE comprise one or more severity indications of the irAE.

[0295] Embodiment 30. The method of embodiment 29, wherein the severity indication of the one or more severity indications of the irAE is based on Common Terminology Criteria for Adverse Events (CTCAE).

[0296] Embodiment 31. The method of embodiment 30, wherein the severity indication of the irAE is severe if the CTCAE greater than or equal to 3 and the severity indication of the irAE is non- severe if the CTCAE is less than 3.

[0297] Embodiment 32. The method of any of embodiments 23-31, wherein the positive outcome relates to cancer stage, cancer progression, cancer severity, cancer remission, health-related quality of life, symptom management, a RECIST 1.1 response criteria or survival.

[0298] Embodiment 33. The method of any of embodiments 28-32, wherein the respective timepoints for the one or more indications of the irAE are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0299] Embodiment 34. The method of any of embodiments 28-33, wherein the respective timepoints for the one or more indications of the positive outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0300] Embodiment 35. The method of any of embodiments 23-34, wherein the sample is a first sample from a first timepoint.

[0301] Embodiment 36. The method of embodiment 35, wherein the method comprises inferring cells classification for a second sample at a second timepoint by:66MF-367064958Attorney Docket No.: 279362000640fluorescently labeling cells contained within a second sample from the candidate ICI recipient, by contacting at least an aliquot of the second sample with at least one immunophenotyping fluorescent labeling panel,generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer;inferring cell classifications based on the fluorescent intensity data.

[0302] Embodiment 37. The method of embodiment 36, comprising providing at least a subset of the cell classifications from the second sample as input to the multioutput machine learning model, the adverse event machine learning model and / or the response machine learning model.

[0303] Embodiment 38. The method of embodiment 37, wherein the multioutput machine learning model has been trained to average over the hidden representations of the first sample and the second sample or applies an attention mechanism over the hidden representations of the first sample and the second sample to predict a trajectory-based prediction.

[0304] Embodiment 39. The method of any of embodiments 28-38, wherein the candidate ICI recipient is sorted as a responder if the predicted probability of the irAE is less than a predetermined adverse event threshold and / or the predicted probability of the positive outcome is greater than a predetermined positive outcome threshold and sorted as a non-responder if the predicted probability of the irAE is greater than the predetermined adverse event threshold and the predicted probability of the positive outcome is less than the predetermined positive outcome threshold.

[0305] Embodiment 40. The method of embodiment 39, wherein the predetermined adverse event threshold is about 25%, about 50%, about 75%, about 95%, or about 99%.

[0306] Embodiment 41. The method of embodiment 39 or 40, wherein the predetermined positive outcome threshold is about 25%, about 50%, about 75%, about 95%, or about 99%.

[0307] Embodiment 42. The method of any of embodiments 28-41, wherein the multioutput machine learning model is a time sensitive machine learning model.

[0308] Embodiment 43. A method for sorting a candidate immune checkpoint inhibition (ICI) recipient as a responder or non-responder, comprising:fluorescently labeling cells contained within a sample from the candidate ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel,67MF-367064958Attorney Docket No.: 279362000640generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer;inferring cell classifications based on the fluorescent intensity data;providing at least a subset of the cell classifications as input to a numerical quantification machine learning model, wherein the numerical quantification machine learning model has been trained to predict a numerical quantification of a clinical outcome with training data comprising at least:(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints;(b) one or more numerical quantifications of a clinical outcome for the plurality of ICI recipients at respective timepoints;outputting from the numerical quantification machine learning model, a predicted numerical quantification of the clinical outcome for the candidate ICI recipient; and sorting the candidate ICI recipient as a responder or a non-responder based at least on the predicted numerical quantification of the clinical outcome.

[0309] Embodiment 44. The method of embodiment 43, wherein the one or more numerical quantifications of a clinical outcome comprise progression-free survival time (PFS), overall survival (OS) time, or time to a change in a biomarker.

[0310] Embodiment 45. The method of embodiment 43 or 44, wherein the respective timepoints for the one or more numerical quantification of a clinical outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0311] Embodiment 46. The method of any of embodiments 43-45, wherein the sample is a first sample from a first timepoint.

[0312] Embodiment 47. The method of embodiment 39, wherein the method comprises inferring cells classification for a second sample at a second timepoint by:fluorescently labeling cells contained within a second sample from the candidate ICI recipient, by contacting at least an aliquot of the second sample with at least one immunophenotyping fluorescent labeling panel,generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer;inferring cell classifications based on the fluorescent intensity data.68MF-367064958Attorney Docket No.: 279362000640

[0313] Embodiment 48. The method of embodiment 47, comprising providing at least a subset of the cell classifications from the second sample as input to the numerical quantification machine learning model.

[0314] Embodiment 49. The method of embodiment 48, wherein the numerical quantification machine learning model has been trained to average over the hidden representations of the first sample and the second sample or applies an attention mechanism over the hidden representations of the first sample and the second sample to predict a trajectorybased prediction.

[0315] Embodiment 50. The method of any of embodiments 43-49, wherein the candidate ICI recipient is sorted as a responder if the predicted numerical quantification of the clinical outcome is longer than a predefined time threshold and sorted as a non-responder of the predicted numerical quantification of the clinical outcome is shorter than the predefined time threshold.

[0316] Embodiment 51. The method of embodiment 50, wherein the predefined time threshold is about 1 month, 6 month, 1 year, 5 years or 10 years.

[0317] Embodiment 52. The method of any of embodiments 43-49, wherein the numerical quantification machine learning model is a time sensitive machine learning model.

[0318] Embodiment 53. The method of any of embodiments 1-52, wherein the candidate ICI recipient has received ICI therapy.

[0319] Embodiment 54. The method of embodiment 53, wherein the ICI recipients in the plurality of ICI recipients have received the same ICI therapy as the candidate ICI recipient.

[0320] Embodiment 55. The method of embodiment 53 or 54, comprising terminating ICI therapy if the candidate ICI recipient is sorted as a non-responder.

[0321] Embodiment 56. The method of any of embodiments 1-55, administering the ICI therapy to the candidate ICI recipient if they are sorted as a responder.

[0322] Embodiment 57. The method of any of embodiments 53-56, wherein the ICI therapy comprises administration of an anti-PDl therapy, administration of an anti-PDLl therapy, administration of an anti-CTLA4 therapy or administration of an anti-LAG3 therapy.

[0323] Embodiment 58. The method of any of embodiments 1-57, wherein the sample comprises peripheral blood cells.

[0324] Embodiment 59. The method of any of embodiments 1-58, wherein the sample comprises isolated peripheral blood mononuclear cells (PBMCs).

[0325] Embodiment 60. A method of monitoring response to an ICI therapy in candidate immune checkpoint inhibition (ICI) recipient, comprising:69MF-367064958Attorney Docket No.: 279362000640administering an ICI therapy to the candidate ICI recipient;at a first timepoint, sorting the candidate ICI recipient as a responder according to any one of embodiments 1-59;at a second timepoint, sorting the candidate ICI recipient as a responder according to any one of embodiments 1-59;monitoring the response based on the candidate ICI recipient being sorted as a responder or a non-responder at the first timepoint and the second timepoint.

[0326] Embodiment 61. The method of embodiment 60, wherein monitoring the response comprises maintaining a treatment plan if the candidate ICI recipient is sorted as a responder at the first timepoint and the second timepoint and modifying the treatment plan if the candidate ICI recipient is sorted as a non-responder at either the first timepoint or the second timepoint.

[0327] Embodiment 62. A method of training an adverse event machine learning model to predict a probability of an immune-related adverse event, comprising:obtaining for each ICI recipient of a plurality of ICI recipients:(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, at least two cell classifications corresponding to different respective timepoints,(b) one or more indications of the irAE at respective timepoints; training an adverse event to predict the probability of the immune-related adverse event with training data comprising at least: (a) the plurality of cell classifications and (b) one or more indications of the irAE.

[0328] Embodiment 63. The method of claim 62, wherein the respective timepoints for the one or more indications of the irAE are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0329] Embodiment 64. The method of embodiment 62 or 63, wherein the adverse event machine learning model is a time sensitive machine learning model.

[0330] Embodiment 65. A method of training a response machine learning model to predict a probability of a positive outcome, comprising:obtaining for each ICI recipient of a plurality of ICI recipients:(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, at least two cell classifications corresponding to different respective timepoints,(b) one or more indications of the positive outcome at respective timepoints;70MF-367064958Attorney Docket No.: 279362000640training a response machine learning model to predict the probability of a positive outcome with training data comprising at least: (a) the plurality of cell classifications and (b) one or more indications of the positive outcome.

[0331] Embodiment 66. The method of embodiment 65, wherein the respective timepoints for the one or more indications of the positive outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0332] Embodiment 67. The method of embodiment 65 or 66, wherein the response machine learning model is a time sensitive machine learning model.

[0333] Embodiment 68. A method of training a multioutput machine learning model to predict a probability of a positive outcome and / or a probability of an immune-related adverse event (irAE), comprising:obtaining for each ICI recipient of a plurality of ICI recipients:(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, at least two cell classifications corresponding to different respective timepoints,(b) one or more indications of the positive outcome at respective timepoints; (c) one or more indications of the irAE at respective timepoints; training a response machine learning model to predict the probability of a positive outcome with training data comprising at least: (a) the plurality of cell classifications and (b) one or more indications of the positive outcome and one or more indications of the irAE.

[0334] Embodiment 69. The method of embodiment 68, wherein the respective timepoints for the one or more indications of the positive outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0335] Embodiment 70. The method of embodiment 68 or 69, wherein the respective timepoints for the one or more indications of the irAE are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0336] Embodiment 71. The method of any of embodiments 68-70, wherein the multioutput machine learning model is a time sensitive machine learning model.

[0337] Embodiment 72. A method of training numerical quantification machine learning model to predict a numerical quantification of a clinical outcome, comprising:obtaining for each ICI recipient of a plurality of ICI recipients:(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, at least two cell classifications corresponding to different respective timepoints,71MF-367064958Attorney Docket No.: 279362000640(b) one or more numerical quantifications of a clinical outcome at respective timepoints;training a response machine learning model to predict the probability of a positive outcome with training data comprising at least: (a) the plurality of cell classifications and (b) one or more numerical quantifications of a clinical outcome.

[0338] Embodiment 73. The method of embodiment 72, wherein the respective timepoints for the one or more numerical quantifications of a clinical outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

[0339] Embodiment 74. The method of embodiment 72 or 73, wherein the numerical quantification machine learning model is a time sensitive machine learning model.

[0340] Embodiment 75. The method of any of embodiments 62-74, wherein the cell classifications of the plurality of cell classification have been obtained by:obtaining fluorescent intensity data, generated from a plurality of fluorescently labeled cells from the ICI recipient;inferring cell classifications based on the fluorescent intensity data.

[0341] Embodiment 76. The method of any of embodiments 1-75, wherein the different respective timepoints comprise timepoints before administration of an ICI therapy, during administration of an ICI therapy and / or after administration of an ICI therapy.

[0342] Embodiment 77. The method of any of embodiments 12, 22, 27, 42, 52, 64, 67, 71, and 74, wherein the different respective timepoints are encoded into the time sensitive machine learning model using transformer-based architecture.

[0343] Embodiment 78. The method of embodiment 77, wherein the transformer-based architecture comprises a temporal transformer.

[0344] Embodiment 79. The method of any of embodiments 12, 22, 27, 42, 52, 64, 67, 71, 74, 77 and 78, wherein the time sensitive machine learning model comprises a recurrent neural network (RNN), a long short-term memory (LSTM) model, or a gated recurrent units (GRUs) model.

[0345] Embodiment 80. The method of any of embodiments 12, 22, 27, 42, 52, 64, 67, 71, 74, 77 and 78, wherein the time sensitive machine learning model comprises a feature extraction model and a longitudinal data model.

[0346] Embodiment 81. The method of embodiment 80, wherein the feature extraction model comprises a convolutional neural network (CNN) or a graph neural network (GNN).72MF-367064958Attorney Docket No.: 279362000640

[0347] Embodiment 82. The method of embodiment 80 or 81, wherein the longitudinal data model comprises a recurrent neural network (RNN) a long short-term memory (LSTM) model, a transformer, or a gated recurrent units (GRUs) model.

[0348] Embodiment 83. The method of any of embodiment 75, wherein the fluorescent intensity data, generated from a plurality of fluorescently labeled cells is generated by a method comprising;fluorescently labeling cells contained within a sample from the ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel;generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer.

[0349] Embodiment 84. The method of embodiment 83, wherein the sample was collected before administration of the ICI therapy, during administration of the ICI therapy and / or after administration of the ICI therapy.

[0350] Embodiment 85. The method of any one of embodiments 1-61, and 83-84, wherein one of the at least one immunophenotyping fluorescent labeling panel comprises a panel of fluorescent-labeled antibodies directed to cell surface proteins associated with antigen-presenting cells (APCs) and / or a panel of fluorescent-labeled antibodies directed to intracellular proteins.

[0351] Embodiment 86. The method of embodiment 85, wherein the panel of fluorescently-labeled antibodies comprises fluorescently-labeled antibodies directed to CD3, CD4, CD8, CD25, CD45, CD19, CD27, IgD, IgM, CD56, CD16, CD14, HLA-DR, CDllc, CD56, TCRgd, TCR Va7.2. TCR V61, TCR V62, TCR Va24-Jal8, CCR10, CD103 / ITGAE, CD122 / IL2RB, CD161 / KLRB1, CD223 / LAG-3, CD274 / PD-L1, CD335 / NKp46, CD43, CD10, CD138, CD141, CD183 / CXCR3, CD185 / CXCR5, CD194 / CCR4, CD197 / CCR7, CD279 / PD-1, CD28, CD294 / CRTH2, CD337 / NKp30, CD38, CD39, CD5, CD62L, CD86, CD95, ICOS, TIGIT, TIM-3, CD40, KLRG1, CD69, CD196 / CCR6, CDlc, CD24, CD267 / TACI, CD303 / BDCA-2 / CLEC4C, CD31, CD319, CD57, CD127, CD45RO, CD45RA, CCR2, CCR6, CXCR4, CX3CR1, Ki67, Granzyme B, TBET, GATA3, EOMES, BLIMP 1, CTLA4, TCF1, TOX, BATE, IRF4, LEF1, ZEB2 or any combination thereof.

[0352] Embodiment 87. The method of embodiment 85 or 86, wherein the panel of fluorescently labeled antibodies comprise fluorescently-labeled antibodies directed to cell surface markers that are indicative of live cells, dead cells, or both.73MF-367064958Attorney Docket No.: 279362000640

[0353] Embodiment 88. The method of any one of embodiments 1-61, 75 and 83-87, wherein the fluorescent intensity data comprises mean fluorescent intensity (MFI) data.

[0354] Embodiment 89. The method of any of embodiments 1-61, 75 and 83-88, wherein inferring the cell classifications comprises inputting the fluorescent intensity data into a cell classification machine learning model and outputting cell classifications.

[0355] Embodiment 90. The method of any of embodiments 1-61, 75 and 83-89, wherein the cell classifications comprise cell ownership into at least 10, at least 20, at least 30, at least 40, at least 50, at least 60, at least 70, at least 80, at least 90, or at least 100 cell populations.

[0356] Embodiment 91. The method of embodiment 90, wherein the cell populations comprise distinct immune cell subpopulations.

[0357] Embodiment 92. The method of embodiment 91 , wherein the distinct immune cell subpopulations comprise white blood cells (WBC), Eosinophils, Eosinophil / CD5+, Neutrophils, Neutrophils / big, Neutrophils / CD5+, Neutrophils / small. B-cells, B-cells / CD5-CD27-, Monocytes / CD56+, Monocytes / CD56-, NK-cells, Dendritic cells (DC), T-cells, iNKT cells, gamma delta T-cells (total GD), Vdl cells, Vd2 cells, Vdx cells, Mucosal-associated invariant T (MAIT) cells, TEMRA cells, CD4 naive cells, T helper cells, CD4 effector memory cells, Treg cells, Leukocytes, Helper T cells, Non-T / Non-NK B cells, Naive T cells, Memory T cells, Naive Cytotoxic T cells, Memory Cytotoxic T cells, Granulocytes, Activated T cells, Non-T / Non-B / Non-NK activated cells, CD3+CD4+FOXP3+ (regulatory T cells), or any combination thereof.

[0358] Embodiment 93. The method of any one of embodiments 1-61 and 83-92, wherein the flow cytometer is configured for at least about 5, at least about 10, at least about 15, at least about 20, at least about 30, at least about 40, at least about 50, at least about 60, at least about 70, at least about 80, at least about 90, or at least about 100 fluorescence detection channels.

[0359] Embodiment 94. The method of any one of embodiments 1-61 and 83-93, wherein the flow cytometer is a full spectrum flow cytometer.

[0360] Embodiment 95. The method of any one of embodiments 1-61 and 83-94, wherein the flow cytometer outputs mean fluorescent intensity (MFI) data.

[0361] Embodiment 96. The method of any one of embodiments 1-12, 23-27, 53-64, 75-76, and 83-95, wherein the adverse event machine learning model comprises a logistic regression machine learning model.

[0362] Embodiment 97. The method of any one of embodiments 1-12, 23-27, 53-64, 75-76, and 83-95, wherein the adverse event machine learning model comprises a Adaptive Best Subset Ensemble (ABSS) machine learning model.74MF-367064958Attorney Docket No.: 279362000640

[0363] Embodiment 98. The method of embodiment 97, wherein the ABSS machine learning model comprises a logistic regression predictor and a multi-layer perceptron (MLP) model.

[0364] Embodiment 99. The method of any one of embodiments 13-27, 53-61, 65-67, 75-76, and 83-95, wherein the response machine learning model comprises a logistic regression machine learning model.

[0365] Embodiment 100. The method of any one of embodiments 13-27, 53-61, 65-67, 75-76, and 83-95, wherein the response machine learning model comprises an Adaptive Best Subset Ensemble (ABSS) machine learning model.

[0366] Embodiment 101. The method of embodiment 100, wherein the ABSS machine learning model comprises a logistic regression predictor and a multi-layer perceptron (MLP) model.

[0367] Embodiment 102. A system comprising:one or more processors;a memory communicative couples to the one or more processors and configures to store instructions that, when executed by the one or more processors cause the system to perform the method of any one of embodiments 1-101.

[0368] Embodiment 103. A non-transitory computer readable storage medium storing instructions which, when executed by one or more processors of a system, cause the system to perform the method of any one of embodiments 1-101.EXAMPLES

[0369] The following examples are included for illustrative purposes only and are not intended to limit the scope of the present disclosure.Example 1 : Analysis of blood samples from ICI recipientsSample Collection and Preparation:

[0370] For this example, multiple modalities of subject (e.g. ICI recipient) samples containing immune cells can be used individually or in combination, including but not limited to whole blood samples or subsets of whole blood samples, such as peripheral blood mononuclear cells (PBMCs). These samples may be analyzed with or without cryopreservation. The methods in this example can be used to analyze a sample collected from the ICI recipient before or after administration of the ICI. The methods in this example can also be used to analyze samples collected from the plurality of ICI recipients used in training machine learning algorithms.75MF-367064958Attorney Docket No.: 279362000640

[0371] Fresh Peripheral Blood: Peripheral blood samples are collected from subjects. The use of fresh blood samples is viable if samples are kept at ambient temperature (approximately 15°C-25°C) and processed within 54 hours; otherwise, samples can be cryopreserved.

[0372] Cryopreserved Peripheral Blood: Collected peripheral blood samples can be processed to isolate peripheral blood mononuclear cells (PBMCs), which are stored at -80°C if processed within 1 week or stored in liquid or vapor phase nitrogen for longer-term storage. Additionally, fresh whole blood samples can be frozen by simply mixing with a basic cryopreservative without the need for PBMC isolation where the latter is not clinically viable. Frozen cell number, starting blood volume, and time from blood draw to preservation are essential metrics to record for downstream processing.Flow Cytometry Analysis

[0373] Full-spectrum flow cytometry is employed to analyze the immune cells from all sample types and infer cell classifications. One or more panels f markers to generate a comprehensive immune profile covering but not limited to:a) Lineage markers for aPT cells, invariant T cells, y6T cells, B cells, NK cells, monocytes, macrophages, dendritic cells, neutrophils, eosinophils, and basophils. (CD3, CD4, CD8, CD25, CD45, CD19, CD27, IgD, IgM, CD56, CD16, CD14, HLA- DR, CDllc, CD56, TCRgd, TCR Va7.2. TCR V61, TCR V62, TCR Va24-Jal8, FOXP3)b) Functional markers relating to but not limited to activation, migration, exhaustion senescence and memory status (CCR10, CD103 / ITGAE, CD122 / IL2RB, CD161 / KLRB1, CD223 / LAG-3, CD274 / PD-L1, CD335 / NKp46, CD43, CD10, CD138, CD141, CD183 / CXCR3, CD185 / CXCR5, CD194 / CCR4, CD197 / CCR7, CD279 / PD-1, CD28, CD294 / CRTH2, CD337 / NKp30, CD38, CD39, CD5, CD62L, CD86, CD95, ICOS, TIGIT, TIM-3, CD40, KLRG1, CD69, CD196 / CCR6, CDlc, CD24, CD267 / TACI, CD303 / BDCA-2 / CLEC4C, CD31, CD319, CD57, CD127, CD45RO, CD45RA), CCR2, CCR6, CXCR4, CX3CR1, Ki67, Granzyme B, TBET, GATA3, EOMES, BLIMP1, CTLA4, TCF1, TOX, BATE, IRF4, LEF1, ZEB2

[0374] The final gating tree is constructed to visually represent the hierarchical relationship between the identified cell subsets. In this nonlimiting example, the following cell subsets are identified: CD45+ (Leukocytes), CD3+ (T cells), CD4+ (Helper T cells), CD8+ (Cytotoxic T cells), CD19+ (B cells), CD14+ (Monocytes), CD56+ (Natural Killer cells), CD16+ (Neutrophils), HLA-DR+ (Activated T cells), CD3-CD56+ (NK cells), CD3-CD19+ (Non-T / Non-NK B cells), CD3+CD16+ (NKT cells), CD4+CD45RA+ (Naive T cells),76MF-367064958Attorney Docket No.: 279362000640CD4+CD45RO+ (Memory T cells), CD8+CD45RA+ (Naive Cytotoxic T cells), CD8+CD45RO+ (Memory Cytotoxic T cells), CD14+HLA-DR+ (Activated Monocytes), CD16+CD45+ (Granulocytes), CD3-CD19-HLA-DR+ (Non-T / Non-B / Non-NK activated cells), CD3+HLA-DR+ (Activated T cells), CD3+CD4+FOXP3+ (regulatory T cells).

[0375] Each subset is defined by the sequential application of gates based on the specific markers identified in the staining panel.

[0376] The inferred cell classifications generated provide a high resolution overview of the immune shape of a subject. Signatures that are measured by this method can be but are not limited to single parameter increases or decreases as well as perturbations of the immune shape as dictated by changes in the combinatorial ratios of parameters.Example 2: Relationship between immune profile and ICI treatment outcome

[0377] The methods described in example 1 were performed to generate immune profiles comprising cell classifications for 300 samples collected from 87 patients with skin or renal cancer receiving treatment with an ICI (anti-PDl / anti-PDLl + / - anti-CTLA4). The samples were collected from the patients at various time points including before treatment and at various on-treatment timepoints, at the onset, peak, and resolution of irAE toxicides.

[0378] Strong associations between therapeutic outcome with immune profile data for different sampling timepoints was shown. Benefit was characterized as no relapse after at least 6 months in the adjuvant setting, and complete response, partial response or stable disease for at least 6 months in the advanced setting The association between benefit of treatment and cell classifications was determined for samples collected before treatment (baseline), the first on treatment sample, and patient matched before treatment (baseline) and first on treatment samples. FIG 9A-9C shows heatmaps of the associations and unsupervised clustering of the samples. FIG.9A shows before treatment (baseline) samples. FIG.9B shows first on treatment samples. FIG. 9C shows matched before treatment (baseline) and first on treatment samples (patient trajectory baseline to on treatment).

[0379] K-means clustering was used to evaluate if the cell classifications in each association could positively predict response. Accuracy of the prediction was measured by ROC-AUC analysis compared to the clinical annotations for response. The ROC-AUC for response was 0.757, and 0.729, for the before treatment (baseline), fist on treatment, respectively. Integrating both timepoints yielded increased accuracy of 0.833.

[0380] Strong associations between irAE severity and immune profile data for different sampling timepoints was also shown. High grade irAE was defined as having a Common 77MF-367064958Attorney Docket No.: 279362000640Terminology Criteria for Adverse Events (CTCAE) of greater than or equal to 3. Low grade irAE was defined as CTCAE of less than 3.

[0381] The association between benefit of irAE severity and cell classifications was determined for samples collected before treatment (baseline), the first on treatment sample, and irAE onset samples. FIGs. 10A-C shows heatmaps of the associations and unsupervised clustering of the samples. FIG. 10A shows before treatment (baseline) samples. FIG. 10B shows first on treatment samples. FIG. 10C shows irAE onset.

[0382] K-means clustering was used to evaluate if the cell classifications in each association could positively predict high grade irAE compared to low grade irAE or absence of irAE. Accuracy of the prediction was measured by ROC-AUC analysis compared to the clinical annotations for response. The ROC-AUC for response was 0.742, 0.887, 0.929 for the before treatment (baseline), fist on treatment, and irAE onset respectively.

[0383] Together these results highlighted the strong associations with clinical outcomes in blood samples from future or current ICI-treated cancer patients profiled using the method described in example 1. These data suggested the ability of immune phenotypes to associate with future therapeutic response and toxicity, as a pre-requisite for machine learning algorithms to perform these predictions upon new cohorts of patients. They also highlighted the clear advantages of longitudinal / repeat sampling in enhancing predictive accuracy, a key benefit of the immunophenotyping method described in Example 1 due to its low cost, non-invasive sampling and scalable infrastructure.Example 3: Predicting response to ICI with immune phenotyping- Logistic regression approachA. Introduction

[0384] Logistic regression predictive modelling can be used to build and validate predictive models of patient outcomes based on the immunophenotyping ML output of Example 1 when sample numbers are limited. It employs bootstrapping, 80:20 cross-validation and permutation to attempt to build stable and validated models given the small set. It can also handle additional clinical covariates that may also influence the outcome, potentially improving the applicability of these models to validation cohorts where the cohort breakdown is skewed.

[0385] In essence, the approach selects a small, stable set of immunophenotyping features relevant to delineating between samples based on the outcome of interest. These features are then employed in multiple iterations of logistic regression models over different 80:20 traimtest splits of the data, yielding an estimate of model performance and variability. In parallel, a78MF-367064958Attorney Docket No.: 279362000640permutation approach where features and outcome labels are randomized is employed for comparison. The model with the mean performance can then be chosen to take forward for external validation etc.B. Methodology

[0386] Data preparation began with subsetting the samples to include only those relevant to the prediction task, based on sampling timepoints and the availability of outcome information. Relevant features were then identified, and filtering was applied to retain only those derived from a mean of more than 15 cell events. Features with low variance or high intercorrelation were removed to reduce redundancy and noise.

[0387] Feature selection was performed using 100 bootstrap resamples. Within each resample, univariate feature selection was conducted using multiple t-tests or linear models against the outcome. The top subset of features was then selected, defined as 10% of the number of available samples, in order to minimise overfitting. Stepwise logistic regression was subsequently applied to these selected features to identify the most informative subset. Informative features were aggregated across bootstraps according to their frequency of selection, and the top candidates were carried forward for modelling.

[0388] Model performance was assessed through cross-validation with permutation testing. In each of 100 iterations, the data was split into training and test sets using an 80:20 stratified split by outcome. A stepwise logistic regression model was fit to the training set using the previously selected features, and performance was evaluated on the held-out test set. For permutation testing, this procedure was repeated using randomly selected features and shuffled outcome labels, providing a baseline for comparison.

[0389] Performance was summarized by aggregating the results of the test and permutation models. Balanced accuracy and ROC-AUC were calculated, and the mean and standard deviation of these metrics were quantified across all iterations. Models with ROC-AUC values closest to the mean of the 100 iterations were selected as representative.C. Application to ICI cohort

[0390] These methods were applied to generate immune profiles of 123 patients with skin or renal cancer undergoing treatment with skin or renal cancer receiving treatment with an ICI (anti-PDl / anti-PDLl + / - anti-CTLA4). The samples were collected from the patients at various time points including before treatment and at various on-treatment timepoints, at the onset, peak, and resolution of irAE toxicides.79MF-367064958Attorney Docket No.: 279362000640Treatment benefit predictions

[0391] Benefit was characterized as no relapse after at least 6 months in the adjuvant setting, and complete response, partial response or stable disease for at least 6 months in the advanced setting. The association between benefit of treatment and cell classifications was determined for samples collected before treatment (baseline) and the first on treatment sample (C2).

[0392] The logistic regression approach was employed to predict treatment benefit from baseline and C2 samples. Mean representative models successfully predicted benefit from baseline (ROC-AUC = 0.817) (FIG. 11A) and C2 (ROC-AUC = 0.85) (FIG. 11B), substantially better than permutations. These models were stable over 100 iterations of 80:20 cross validation (CV), achieving mean+SD of 0.72+0.11 (balanced accuracy) and 0.814+0.11 (ROC-AUC) from baseline and 0.775+0.102 (balanced accuracy) and 0.85+0.102 (ROC-AUC) from C2 (Table 2).Table 2: Results for Treatment BenefitirAE severity predictions

[0393] A patient was classified as max irAE severity of severe if they suffered at least one irAE having a Common Terminology Criteria for Adverse Events (CTCAE) score of greater than or equal to 3. Patients were classed as none / non-severe if they either did not develop any irAEs or if their maximum CTCA score was less than 3. The association between max irAE severity and cell classifications was determined for samples collected before treatment (baseline), the first on treatment sample (C2), and at the time of onset of their highest grade irAE (AE Dev).

[0394] The logistic regression approach was employed to predict max irAE severity from baseline, C2 and AE Dev samples. Mean representative models successfully predicted max severity from baseline (ROC-AUC = 0.835) (FIG. 12A) and C2 (ROC-AUC = 0.833) (FIG.12B), substantially better than permutations. These models were stable over 100 iterations of 80:20 CV, achieving mean+SD of 0.699+0.106 (balanced accuracy) and 0.836+0.078 (ROC-AUC) from baseline, 0.714+0.148 (balanced accuracy) and 0.82+0.126 (ROC-AUC) from C2, and 0.706+0.127 (balanced accuracy) and 0.748+0.131 (ROC-AUC) from AE Dev (Table 3).80MF-367064958Attorney Docket No.: 279362000640Table 3: Results for Max irAE

[0395] Together these results highlighted the strong associations with clinical outcomes in blood samples from future or current ICI-treated cancer patients profiled using the method described in example 1. These data suggested the ability of immune phenotypes to associate with future therapeutic response and toxicity, as a pre-requisite for machine learning algorithms to perform these predictions upon new cohorts of patients. They also highlighted the clear advantages of longitudinal / repeat sampling in enhancing predictive accuracy, a key benefit of the immunophenotyping method described in Example 1 due to its low cost, non-invasive sampling and scalable infrastructure.Example 4: Predicting response to ICI with immune phenotyping- Logistic regression approach

[0396] The methods described in Example 1 were applied to generate immune profiles of 136 (Baseline) or 120 (C2) patients with skin or renal cancer undergoing treatment with an ICI (anti-PDl / anti-PDLl + / - anti-CTLA4). The samples were collected from the patients at various time points including before treatment and at various on-treatment timepoints, at the onset, peak, and resolution of irAE toxicides.Progression free survival predictions

[0397] Progression free survival (PFS) was characterized as an absence of death, disease progression or cancer relapse at 12 months post-treatment initiation. The association between PFS and cell classifications was determined for samples collected before treatment (baseline) and the first on treatment sample (C2).

[0398] The logistic regression approach described in FIG. 6, was employed to predict PFS from baseline and C2 samples. For baseline samples, the typical ratio values were used as input, whereas for C2, centered-log ratio transformed values were used, demonstrating the modularity of input data processing. For baseline samples, median values were 0.56 (AUC, FIG. 13A) and 0.54 (balanced accuracy, FIG.13B), and for C2 samples, median values were 0.69 (AUC, FIG.14A) and 0.64 (balanced accuracy, FIG. 14B), all significantly higher than the permutation baseline of ~0.5. Together these results highlighted the associations with clinical outcomes in81MF-367064958Attorney Docket No.: 279362000640blood samples from future or current ICI-treated cancer patients profiled using the method described in Example 1.

[0399] A patient was classified as max irAE severity of severe if they suffered at least one irAE having a Common Terminology Criteria for Adverse Events (CTCAE) score of greater than or equal to 3. Patients were classed as none / non-severe if they either did not develop any irAEs or if their maximum CTCA score was less than 3. The association between max irAE severity and cell classifications was determined for samples collected before treatment (baseline), the first on treatment sample (C2), and at the time of onset of their highest grade irAE (AE Dev).irAE severity predictions

[0400] The logistic regression approach described in FIG.6 was employed to predict max irAE severity from baseline and C2 samples using hyperparameter selection. For baseline samples, median values were 0.63 (AUC, FIG. 15A) and 0.59 (balanced accuracy, FIG. 15B), and for C2 samples, median values were 0.71 (AUC, FIG. 16A) and 0.66 (balanced accuracy, FIG .16B), all significantly higher than the permutation baseline of ~0.5. Together these results highlighted the associations with clinical outcomes in blood samples from future or current ICI-treated cancer patients profiled using the method described herein.Conclusion

[0401] These data suggested the ability of immune phenotypes to associate with future therapeutic response and toxicity, an association that can be leveraged by machine learning algorithms to perform these predictions upon cohorts of patients. They also highlighted the clear advantages of longitudinal / repeat sampling in enhancing predictive accuracy, a key benefit of the immunophenotyping methods described in Example 1 due to its low cost, non-invasive sampling and scalable infrastructure.Example 5: Adaptive Best Subset Ensemble (ABSS) models

[0402] The methods described in Example 1 were applied to generate immune profiles of 123 patients with skin or renal cancer undergoing treatment with skin or renal cancer receiving treatment with an ICI (anti-PDl / lanti-PDLl + / - anti-CTLA4). The samples were collected from the patients at various time points including before treatment and at various on-treatment timepoints, at the onset, peak, and resolution of irAE toxicides.

[0403] ABSS models were trained for each timepoint and outcome. The ABSS model training included multilayer feature selection cross validated with a support vector classifier (SVC).82MF-367064958Attorney Docket No.: 279362000640Once the ensemble models were made, a majority votes strategy was employed for the ensemble predictions.Treatment benefit predictions

[0404] Benefit was characterized as no relapse after at least 6 months in the adjuvant setting, and complete response, partial response or stable disease for at least 6 months in the advanced setting. The association between benefit of treatment and cell classifications was determined for samples collected before treatment (baseline) and the first on treatment sample (C2).

[0405] The ABSS approach was employed to predict treatment benefit from baseline and C2 samples and the per-patient difference between baseline and C2.irAE severity predictions

[0406] A patient was classified as max irAE severity of severe if they suffered at least one irAE having a Common Terminology Criteria for Adverse Events (CTCAE) score of greater than or equal to 3. Patients were classed as none / non-severe if they either did not develop any irAEs or if their maximum CTCA score was less than 3. The association between max irAE severity and cell classifications was determined for samples collected before treatment (baseline), the first on treatment sample (C2), and at the time of onset of their highest grade irAE (AE Dev).

[0407] The ABSS approach was employed to predict max irAE severity from baseline and C2 samples and the per-patient difference between baseline and C2.

[0408] The results highlighted the strong associations with clinical outcomes in blood samples from future or current ICI-treated cancer patients profiled using the method described in example 1. These data suggested the ability of immune phenotypes to associate with future therapeutic response and toxicity, as a pre-requisite for machine learning algorithms to perform these predictions upon new cohorts of patients. They also highlighted the clear advantages of longitudinal / repeat sampling in enhancing predictive accuracy, a key benefit of the immunophenotyping method described in Example 1 due to its low cost, non-invasive sampling and scalable infrastructure.Example 6: ABSS Prediction of IRAE severity using a validation cohort

[0409] The methods described in Example 1 were used to generate immune profiles of patients with skin or renal cancer undergoing treatment with an ICI (anti-PDl / anti-PDLl + / - anti-CTLA4). Samples were collected at various timepoints and immune profiles were generated for 44 patients at Baseline or 122 patients at C2. Using the methods described in Example 5, the ABSS models were trained for each timepoint and outcome using a first cohort of patients83MF-367064958Attorney Docket No.: 279362000640and validated on a hold out cohort of patients. The ABSS model training included multilayer feature selection cross validated with a support vector classifier (SVC). Once the ensemble models were made, a voting strategy was employed for the ensemble predictions.irAE severity predictions

[0410] A patient was classified as max irAE severity of severe if they suffered at least one irAE having a Common Terminology Criteria for Adverse Events (CTCAE) score of greater than or equal to 3. Patients were classed as none / non-severe if they either did not develop any irAEs or if their maximum CTCA score was less than 3. The association between max irAE severity and cell classifications was determined for samples collected before treatment (baseline), the first on treatment sample (C2), and at the time of onset of their highest grade irAE (AE Dev).

[0411] The ABSS approach was employed to predict maximum irAE severity from 104 baseline and 78 C2 samples with outcome annotations. The model built from baseline samples performed with a ROC-AUC of 0.717 on the hold-out cohort of 38 patients with annotated outcomes (FIG. 17A), and classified them into benefit groups with a balanced accuracy of 0.692 (FIG. 17B). The model built from C2 samples performed with a ROC-AUC of 0.742 on the hold-out cohort of 40 patients with annotated outcomes (FIG. 18A) and classified the patients into benefit groups with a balanced accuracy of 0.722 (FIG. 18B). These results demonstrate the potential of predicting IRAE toxicity severity from pre-treatment (baseline) or early on-treatment (C2) immune phenotypic data using machine learning methods.Example 7: ABSS prediction of PFS from baseline

[0412] The methods described in Example 1 were applied to generate immune profiles of patients with skin or renal cancer undergoing treatment with an ICI (anti-PDl / anti-PDLl + / -anti-CTLA4). Samples were collected at various timepoints and immune profiles were generated for 144 patients at Baseline. Using the methods described in Example 5 ABSS models were trained for each timepoint and outcome using a first cohort of patients and validated on a hold out cohort of patients. To understand the generalisability and stability of the model performance, the train and test splits were performed five times with different random seeds, generating a set of models and a distribution of performance metrics. The ABSS model training included multilayer feature selection cross validated with a logistic regression (LR) classifier and LR for the final tuning. Furthermore, a single train / test instance was used to train an ABSS model using a multi-layer perceptron for the production model, demonstrating84MF-367064958Attorney Docket No.: 279362000640the modularity of the approach at different stages. Once the ensemble models were made, a weighted votes strategy was employed for the ensemble predictions.

[0413] Progression-free survival, a measure of treatment response, was defined as the absence of death or cancer relapse or progression at 12 months following ICI treatment initiation. Five distinct seeds were used to generate 5 distinct splits of train and test cohorts, enabling the training of 5 separate models (each using 87 samples) and validation in two hold-out sets of 11 and 28 samples, respectively. The first hold-out set was produced within the model structure (block 718 in FIG. 7) while the second hold out set was treated entirely independently, and existed outside the ABSS workflow. Using logistic regression-based classification as the final model inside the ABSS models, the first hold-out set displayed performance of ROC-AUC between 0.4 and 0.7, with a mean of 0.574 (FIGs.19A-19E). The second hold-out set displayed performance of ROC-AUC between 0.49 and 0.67, with a mean of 0.584 (FIGs. 19F-19J). A single train / test split was used to generate an ABSS model using a LR as the core predictor and multi-layer perceptron (MLP) for the final model, and achieved performance of ROC-AUC of 0.73 in both hold out sets (FIG 20A-20B). Together, this example demonstrated the ability of ABSS models with modular predictor model components to generally produce non-random prediction of progression free survival outcomes from pre-treatment baseline immune phenotypic data. The range of performance highlighted the instability of model training at this sample size and suggests that predictive performance is achievable with greater training dataset size.85MF-367064958

Claims

1. Attorney Docket No.: 279362000640CLAIMSWhat is claimed is:

1. A method for sorting a candidate immune checkpoint inhibition (ICI) recipient as a responder or non-responder, comprising:fluorescently labeling cells contained within a sample from the candidate ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel,generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer;inferring cell classifications based on the fluorescent intensity data;providing at least a subset of the cell classifications as input to an adverse event machine learning model, wherein the adverse event machine learning model has been trained to predict a probability of an immune-related adverse event (irAE) with training data comprising at least:(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints,(b) one or more indications of the irAE for the plurality of ICI recipients at respective timepoints;outputting from the adverse event machine learning model, a predicted probability of the irAE for the candidate ICI recipient; andsorting the candidate ICI recipient as a responder or a non-responder based at least on the predicted probability of the irAE.

2. The method of claim 1, wherein the one or more indication of the irAE comprise one or more severity indications of the irAE.

3. The method of claim 2, wherein the severity indication of the one or more severity indications of the irAE is based on Common Terminology Criteria for Adverse Events (CTCAE).86MF-367064958Attorney Docket No.: 2793620006404. The method of claim 3, wherein the severity indication of the irAE is severe if the CTCAE greater than or equal to 3 and the severity indication of the irAE is non-severe if the CTCAE is less than 3.

5. The method of any of claims 1-4, wherein the respective timepoints for the one or more indications of the irAE are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

6. The method of any of claims 1-5, wherein the sample is a first sample from a first timepoint.

7. The method of claim 6, wherein the method comprises inferring cells classification for a second sample at a second timepoint by:fluorescently labeling cells contained within a second sample from the candidate ICI recipient, by contacting at least an aliquot of the second sample with at least one immunophenotyping fluorescent labeling panel,generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer;inferring cell classifications based on the fluorescent intensity data.

8. The method of claim 7, comprising providing at least a subset of the cell classifications from the second sample as input to the adverse event machine learning model.

9. The method of claim 8, wherein the adverse event machine learning model has been trained to average over the hidden representations of the first sample and the second sample or applies an attention mechanism over the hidden representations of the first sample and the second sample to predict a trajectory-based prediction.

10. The method of any of claims 1-9, wherein the candidate ICI recipient is sorted as a responder if the predicted probability of the irAE is less than a predetermined adverse event threshold and sorted as a non-responder if the predicted probability of the irAE is greater than the predetermined adverse event threshold.87MF-367064958Attorney Docket No.: 27936200064011. The method of claim 10, wherein the predetermined adverse event threshold is about 25%, about 50%, about 75%, about 95%, or about 99%.

12. The method of any of claims 1-11, wherein the adverse event machine learning model is a time sensitive machine learning model.

13. A method for sorting a candidate immune checkpoint inhibition (ICI) recipient as a responder or non-responder, comprising:fluorescently labeling cells contained within a sample from the candidate ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel,generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer;inferring cell classifications based on the fluorescent intensity data;providing at least a subset of the cell classifications as input to a response machine learning model, wherein the response machine learning model has been trained to predict the probability of a positive outcome with training data comprising at least:(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints(b) one or more indications of the positive outcome for the plurality of ICI recipients at respective timepoints;outputting from the response machine learning model, a predicted probability of the positive outcome for the candidate ICI recipient; andsorting the candidate ICI recipient as a responder or a non-responder based at least on the predicted probability of the of positive outcome.

14. A method for sorting a candidate immune checkpoint inhibition (ICI) recipient as a responder or non-responder, comprising:fluorescently labeling cells contained within a sample from the candidate ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel,generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer;88MF-367064958Attorney Docket No.: 279362000640inferring cell classifications based on the fluorescent intensity data;providing at least a subset of the cell classifications as input to an adverse event machine learning model and a response machine learning model;wherein the adverse event machine learning model has been trained to predict the probability of an immune-related adverse event (irAE) with a first set of training data comprising at least:(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, for each ICI recipient of a plurality of ICI recipients, at least two cell classifications corresponding to different respective timepoints,(b) one or more indications of the irAE for the plurality of ICI recipients at respective timepoints;wherein the response machine learning model has been trained to predict the probability of a positive outcome with a second set of training data comprising at least:(a) the plurality of cell classifications,(b) one or more indications of the positive outcome for the plurality of ICI recipients at respective timepoints;outputting from the response machine learning model, a predicted probability of the positive outcome and from the adverse event machine learning model, a predicted probability of the irAE for the candidate ICI recipient; andsorting the candidate ICI recipient as a responder or a non-responder based at least on the predicted probability of the of positive outcome and the predicted probability of the irAE.

15. The method of claim 14, wherein sorting the candidate ICI recipient comprising combining the predicted probability of the positive outcome and the predicted probability of the irAE using voting, stacking, averaging, or Logit blending methods.

16. The method of claim 14 or 15, wherein the respective timepoints for the one or more indications of the irAE are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

17. The method of any of claims 14-16, wherein the respective timepoints for the one or more indications of the positive outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.89MF-367064958Attorney Docket No.: 27936200064018. A method of monitoring response to an ICI therapy in candidate immune checkpoint inhibition (ICI) recipient, comprising:administering an ICI therapy to the candidate ICI recipient;at a first timepoint, sorting the candidate ICI recipient as a responder according to any one of claims 1-17;at a second timepoint, sorting the candidate ICI recipient as a responder according to any one of claims 1-17;monitoring the response based on the candidate ICI recipient being sorted as a responder or a non-responder at the first timepoint and the second timepoint.

19. The method of claim 18, wherein monitoring the response comprises maintaining a treatment plan if the candidate ICI recipient is sorted as a responder at the first timepoint and the second timepoint and modifying the treatment plan if the candidate ICI recipient is sorted as a non-responder at either the first timepoint or the second timepoint.

20. A method of training an adverse event machine learning model to predict a probability of an immune-related adverse event, comprising:obtaining for each ICI recipient of a plurality of ICI recipients:(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, at least two cell classifications corresponding to different respective timepoints,(b) one or more indications of the irAE at respective timepoints; training an adverse event to predict the probability of the immune-related adverse event with training data comprising at least: (a) the plurality of cell classifications and (b) one or more indications of the irAE.

21. The method of claim 20, wherein the respective timepoints for the one or more indications of the irAE are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

22. The method of claim 20 or 21, wherein the adverse event machine learning model is a time sensitive machine learning model.90MF-367064958Attorney Docket No.: 27936200064023. A method of training a response machine learning model to predict a probability of a positive outcome, comprising:obtaining for each ICI recipient of a plurality of ICI recipients:(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, at least two cell classifications corresponding to different respective timepoints,(b) one or more indications of the positive outcome at respective timepoints; training a response machine learning model to predict the probability of a positive outcome with training data comprising at least: (a) the plurality of cell classifications and (b) one or more indications of the positive outcome.

24. The method of claim 23, wherein the respective timepoints for the one or more indications of the positive outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

25. A method of training a multioutput machine learning model to predict a probability of a positive outcome and / or a probability of an immune-related adverse event (irAE), comprising:obtaining for each ICI recipient of a plurality of ICI recipients:(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, at least two cell classifications corresponding to different respective timepoints,(b) one or more indications of the positive outcome at respective timepoints; (c) one or more indications of the irAE at respective timepoints; training a response machine learning model to predict the probability of a positive outcome with training data comprising at least: (a) the plurality of cell classifications and (b) one or more indications of the positive outcome and one or more indications of the irAE.

26. The method of claim 25, wherein the respective timepoints for the one or more indications of the positive outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.91MF-367064958Attorney Docket No.: 27936200064027. The method of claim 25 or 26, wherein the respective timepoints for the one or more indications of the irAE are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

28. A method of training numerical quantification machine learning model to predict a numerical quantification of a clinical outcome, comprising:obtaining for each ICI recipient of a plurality of ICI recipients:(a) a plurality of cell classifications, wherein the plurality of cell classifications includes, at least two cell classifications corresponding to different respective timepoints,(b) one or more numerical quantifications of a clinical outcome at respective timepoints;training a response machine learning model to predict the probability of a positive outcome with training data comprising at least: (a) the plurality of cell classifications and (b) one or more numerical quantifications of a clinical outcome.

29. The method of claim 28, wherein the respective timepoints for the one or more numerical quantifications of a clinical outcome are the different respective timepoints from the plurality of cell classifications or are alternative timepoints.

30. The method of any of claims 20-29, wherein the cell classifications of the plurality of cell classification have been obtained by:obtaining fluorescent intensity data, generated from a plurality of fluorescently labeled cells from the ICI recipient;inferring cell classifications based on the fluorescent intensity data.

31. The method of any of claims 1-30, wherein the different respective timepoints comprise timepoints before administration of an ICI therapy, during administration of an ICI therapy and / or after administration of an ICI therapy.

32. The method of any of claims 30, wherein the fluorescent intensity data, generated from a plurality of fluorescently labeled cells is generated by a method comprising;92MF-367064958Attorney Docket No.: 279362000640fluorescently labeling cells contained within a sample from the ICI recipient, by contacting at least an aliquot of the sample with at least one immunophenotyping fluorescent labeling panel;generating fluorescent intensity data by processing the fluorescently-labeled cells from the sample using a flow cytometer.

33. The method of claim 32, wherein the sample was collected before administration of the ICI therapy, during administration of the ICI therapy and / or after administration of the ICI therapy.

34. The method of any one of claims 1-12, 14-21 and 30-33, wherein the adverse event machine learning model comprises a logistic regression machine learning model.

35. The method of any one of claims 1-12, 14-21, and 30-33, wherein the adverse event machine learning model comprises a Adaptive Best Subset Ensemble (ABSS) machine learning model.

36. The method of any one of claims 13 and, 30-33, wherein the response machine learning model comprises a logistic regression machine learning model.

37. The method of any one of claims 13 and 30-33, wherein the response machine learning model comprises an Adaptive Best Subset Ensemble (ABSS) machine learning model.

38. The method of claim 37, wherein the ABSS machine learning model comprises a logistic regression predictor and a multi-layer perceptron (MLP) model.

39. A system comprising:one or more processors;a memory communicative couples to the one or more processors and configures to store instructions that, when executed by the one or more processors cause the system to perform the method of any one of claims 1-38.93MF-367064958Attorney Docket No.: 27936200064040. A non-transitory computer readable storage medium storing instructions which, when executed by one or more processors of a system, cause the system to perform the method of any one of claims 1-38.94MF-367064958

Images