Machine learning single-cell senescent cell identification and classification

Abstract

Machine learning techniques of characterizing senescence in an immobilized patient cell are disclosed. The techniques include: obtaining a patient cell image; passing the patient cell image to a trained machine learning system, where the trained machine learning system is trained using a training corpus that includes a plurality of training cell images, where the plurality of training cell images include a plurality of images of normal cells and a plurality of images of cells with senescence; obtaining an output from the trained machine learning model in response to the passing, where the output is indicative of a senescence property of the patient cell, and where the senescence property is non-binary; and providing an indication of the senescence property of the patient cell based on the output.

Description

MACHINE LEARNING SINGLE-CELL SENESCENT CELL IDENTIFICATION AND CLASSIFICATIONGovernment Support

[0001] This invention was made with government support under grants AG060903 and AG021334, awarded by the National Institutes of Health. The government has certain rights in the invention.Related Applications

[0002] This application claims the benefit of U.S. Provisional Patent Application No. 63 / 589,379, entitled “Novel Single-Cell Technology to Identify and Classify Senescent Cells,” filed October 11 , 2023, and the benefit of U.S. Provisional Patent Application No. 63 / 645,487, entitled “Machine Learning Single-Cell Senescent Cell Identification And Classification,” filed on May 10, 2024.Field

[0003] This disclosure relates generally to senescent cell identification and classification.Background

[0001] Cellular senescence is an established driver of aging. Despite its mechanistic importance, profiling senescence within cell populations is challenging. This is in part due to the limitations of current biomarkers to robustly identify senescent cells across biological settings, and the variety of forms exhibited by senescent cells.

[0002] Senescence may be characterized by stable proliferative arrest, upregulation of cyclin-dependent kinase (CDK) inhibitors such as p16 and p21 ,increased secretions of pro-inflammatory molecules, and drastic changes in cell and nuclear morphologies. Although these senescence-associated changes have shaped the current understanding of senescence across cell populations, such understanding is limited. These limitations stem from at least the following, (i) There are currently no absolute biomarkers of senescence. Instead, researchers typically rely on a handful of protein-based biomarkers, and more recently, curated gene expression signatures as proxies of senescence, which in many cases do not capture all senescent cells across biological settings, (ii) Cellular senescence is typically defined as a binary phenotype, meaning that cells are considered as senescent of not. However, this binarization comes with the implied assumption that senescent cells are in some ways equivalent and may respond similarly to biological perturbations, such as senotherapies.

[0003] Significant advances in single-cell technologies and modem data science approaches have led to limited methods to profile senescent cells. Some studies that use next-generation sequencing technologies on bulk or single-cell populations have demonstrated robust ways to profile large numbers of senescent cells based on curated gene sets. Such approaches have allowed for the simultaneous interrogation of multiple transcriptionally-defined markers and putative molecular signatures. However, performing single-cell sequencing brings a few challenges related to the high cost of sequencing and the throughput and feasibility of profiling multiple samples and conditions.

[0004] Recent image-based strategies have employed high-content microscopy coupled with machine learning to analyze senescent cells cultured on flat substrates coated with physiologically-relevant extracellular matrix (ECM) proteins. Some studies used microscopy images of senescent and non-senescent cells to train computationalmodels to provide an estimation of whether a cell is senescent or not, with downstream validation of senescence based on the expression of senescent biomarkers. See, e.g., Heckenbach, I. et al. Nuclear morphology is a deep learning biomarker of cellular senescence. Nat Aging 2, 742-755 (2022); Kusumoto, D. et al. Anti-senescent drug screening by deep learning-based morphology senescence scoring. Nat Commun 12, (2021 ); and Duran, I. et al. Detection of senescence using machine learning algorithms based on nuclear features. Nat Commun 15, 1041 (2024). While such studies provide a proof-of-concept for identifying senescence cells using imaging, they make fundamental and limiting assumptions, including: (i) the assumption that senescence is a binary phenotype (i.e., a given cell is either monotypic senescent or not senescent), and (ii) the assumption that all cells exposed to senescence inducers in the study were senescent and vehicle-treated cells were non-senescent. However, generalizations under these assumptions bias the interpretation of results, rendering them inaccurate and of limited descriptive power and use, e.g., for therapeutic purposes.Summary

[0005] According to various embodiments, a machine learning method of characterizing senescence in an immobilized patient cell is presented. The method includes: obtaining a patient cell image; passing the patient cell image to a trained machine learning system, where the trained machine learning system is trained using a training corpus including a plurality of training cell images, where the plurality of training cell images include a plurality of images of normal cells and a plurality of images of cells with senescence; obtaining an output from the trained machine learning model in response to the passing, where the output is indicative of asenescence property of the patient cell, and where the senescence property is nonbinary; and providing an indication of the senescence property of the patient cell based on the output.

[0006] Various optional features of the above method embodiments include the following. The trained machine learning system may include a neural network, where the training cell images are labeled with respective indications representative of senescence or not senescence, and where the senescence property includes a senescence score for the patient cell. The method may include clinically treating the patient with a therapeutic based on the indication of the senescence property of the patient cell, where at least one of a dosage of the therapeutic or a duration of the therapeutic is based on the indication of the senescence property of the patient cell. The patient cell image may include at least one of: an image of the patient cell, an image of a nucleus of the patient cell, or a composite image of the patient cell and the nucleus of the patient cell, and a respective training cell image may include at least one of: an image of a respective training cell, an image of a nucleus of a respective training cell, or a composite image of a respective training cell and the nucleus of the respective training cell. The senescence property may include a senescence subtype classification of the patient cell. The method may include: selecting a therapeutic based on the indication of the senescence property of the patient cell, where the therapeutic includes at least one of: a chemotherapy drug, CAR-T therapy, or a senotherapy drug; and clinically treating the patient with the treatment. The machine learning system may be trained using the plurality of training cell images by actions including: determining respective values of each of a set of morphological parameters for each of the plurality of training cell images, reducing a dimensionality of the set of morphological parameters based on results of the determining, such that a reduceddimensionality is obtained, and clustering the plurality of training cell images in the reduced dimensionality, such that a plurality of clusters are obtained, where the output is based on a closest cluster of the plurality of clusters to the patient cell, and where the senescence classification of the patient cell corresponds to the closest cluster of the plurality of clusters to the patient cell. The patient cell image may include immunofluorescence staining, and each training cell image may include immunofluorescence staining. The senescence property may include a characterization of a susceptibility to senescence of the patient cell. The method may include providing an indication of a protein expression of the patient cell based on the senescence property.

[0007] According to various embodiments, a machine learning system for characterizing senescence in an immobilized patient cell is presented. The system includes: a non-transitory computer readable medium including instructions; and at least one electronic processor that executes the instructions to perform operations including: obtaining a patient cell image; passing the patient cell image to a trained machine learning system, where the trained machine learning system is trained using a training corpus including a plurality of training cell images, where the plurality of training cell images include a plurality of images of normal cells and a plurality of images of cells with senescence; obtaining an output from the trained machine learning model in response to the passing, where the output is indicative of a senescence property of the patient cell, and where the senescence property is nonbinary; and providing an indication of the senescence property of the patient cell based on the output.

[0008] Various optional features of the above system embodiments include the following. The trained machine learning may include includes a neural network, wherethe training cell images are labeled with respective indications representative of senescence or not senescence, and where the senescence property includes a senescence score for the patient cell. The patient may be clinically treated with a therapeutic based on the indication of the senescence property of the patient cell, where at least one of a dosage of the therapeutic or a duration of the therapeutic is based on the indication of the senescence property of the patient cell. The patient cell image may include at least one of: an image of the patient cell, an image of a nucleus of the patient cell, or a composite image of the patient cell and the nucleus of the patient cell, and a respective training cell image may include at least one of: an image of a respective training cell, an image of a nucleus of a respective training cell, or a composite image of a respective training cell and the nucleus of the respective training cell. The senescence property may include a senescence subtype classification of the patient cell. The operations may further include: selecting a therapeutic based on the indication of the senescence property of the patient cell, where the therapeutic includes at least one of: a chemotherapy drug, CAR-T therapy, or a senotherapy drug, where the patient is clinically treated with the treatment. The machine learning system may be trained using the plurality of training cell images by actions including: determining respective values of each of a set of morphological parameters for each of the plurality of training cell images, reducing a dimensionality of the set of morphological parameters based on results of the determining, such that a reduced dimensionality is obtained, and clustering the plurality of training cell images in the reduced dimensionality, such that a plurality of clusters are obtained, where the output is based on a closest cluster of the plurality of clusters to the patient cell, and where the senescence classification of the patient cell corresponds to the closest cluster of the plurality of clusters to the patient cell. The patient cell image may includeimmunofluorescence staining, and each training cell image may include immunofluorescence staining. The senescence property may include a characterization of a susceptibility to senescence of the patient cell. The operations may further include providing an indication of a protein expression of the patient cell based on the senescence property.

[0009] Combinations, (including multiple dependent combinations) of the above-described elements and those within the specification have been contemplated by the inventors and may be made, except where otherwise indicated or where contradictory.Brief Description of the Drawings

[0010] Various features of the examples can be more fully appreciated, as the same become better understood with reference to the following detailed description of the examples when considered in connection with the accompanying figures, in which:

[0011] Fig. 1 is a simplified graphical illustration of an experimental workflow to induce and characterize senescent cells, as used in a study and according to various embodiments;

[0012] Fig. 2 provides representative images of immunofluorescent staining of five senescence-associated protein biomarkers across untreated (DMSO vehicle) and DOX-induced primary dermal fibroblast samples, according to various embodiments and as used in the study;

[0013] Fig. 3 illustrates box plots for Z-score fluorescence quantification relative to control of |3Gal, p16, p21 , HMGB1 , and LMNB1 for young and old dermal fibroblast samples, as used in the study and according to various embodiments;

[0014] Fig. 4 depicts a heatmap of standard scaled expression of pro- inflammatory secretions for induction conditions, as used in the study and according to various embodiments;

[0015] Fig. 5 shows a waterfall plot of cellular and nuclear morphological features enrichment to either untreated or senescence-induced samples across two sample ages, as used in the study and according to various embodiments;

[0016] Fig. 6 presents a schematic of an morphological analysis pipeline based on unsupervised machine learning clustering, as used in the study and according to various embodiments;

[0017] Fig. 7 shows a scatter plot of LIMAP space based on 87 morphological parameters for all conditions across two age groups and an overlay of 11 clusters on top of the LIMAP space with representative cellular (outer) and nuclear (inner) morphologies, according to the study and various embodiments;

[0018] Fig. 8 illustrates scatter plots of select morphological features layered on top of the LIMAP space, according to the study and various embodiments;

[0019] Fig. 9 is a flow diagram for workflow for protein imputation as used in the study and according to various embodiments;

[0020] Fig. 10 is a schematic diagram of a method of using cellular and nuclear textured images to develop single-cell resolution senescent scores, ranging from 0-1 , as used in a study and according to various embodiments;

[0021] Fig. 11 illustrates mean senescence scores of each k-means cluster overlayed on top of the LIMAP space, according to the study and various embodiments;

[0022] Fig. 12 is a heatmap describing the differential enrichment amongst the three morphological senescence subtypes across all experimental conditions, according to the study and various embodiments;

[0023] Fig. 13 depicts confusion matrices for an example XGBoost embodiment; and

[0024] Fig. 14 is a flow chart for a machine learning method of characterizing senescence in an immobilized patient cell, according to various embodiments.Description of the Examples

[0025] Reference will now be made in detail to example implementations, illustrated in the accompanying drawings. Wherever convenient, the same reference numbers will be used throughout the drawings to refer to the same or like parts. In the following description, reference is made to the accompanying drawings that form a part thereof, and in which is shown by way of illustration specific exemplary examples in which the invention may be practiced. These examples are described in sufficient detail to enable those skilled in the art to practice the invention and it is to be understood that other examples may be utilized and that changes may be made without departing from the scope of the invention. The following description is, therefore, merely exemplary.

[0026] Some embodiments provide image-based approaches that utilize machine learning to not only identify senescent phenotypes, but also go beyond binary classifications. Some embodiments provide a robust single-cell framework using imaging, morphological profiling, and machine learning to identify and classify functional subtypes of senescence across multiple biological conditions. Some embodiments use machine learning to identify and classify functional subtypes ofsenescence, while mitigating some of the current limitations with senescence identification and classification, including the oversimplified and therefore incorrect assumption that senescence may be adequately characterized by binary cell classification according to senescent versus not senescent.

[0027] Embodiments are described in reference to a study as follows. Using primary dermal fibroblasts from two donors aged 23 and 89, the inventors initially profiled the morphologies of cells at baseline, after serial passaging (replicative senescence) and eight days post-induction using four chemical inducers. By profiling single cells across each of these biological conditions, the inventors identified eleven morphologically-defined cellular subtypes out of which three were considered bona fide senescence subtypes (labeled herein as C7, C10, C11 ) based on the exhibited morphologies and the expression of multiple protein-based senescence biomarkers (up to five biomarkers).

[0028] Some embodiments use machine learning to classify a patient cell as belonging to one of a plurality of senescent subtypes based on morphological features.

[0029] According to some such embodiments that utilize unsupervised machine learning clustering, a patient cell is labeled with a senescent subtype according to proximity to a corresponding subtype cluster, e.g., using a k-nearest neighbor (KNN) approach. Embodiments may determine morphological parameters for training cell images, reducing a dimensionality of the set of morphological parameters, and perform clustering in the reduced dimensionality space. Such unsupervised machine learning and senescence classification - using a reduced dimensionality of morphological features and clustering - solves the oversimplification problem of the prior art, in which senescence is considered in a binary manner as either being present or absent.

[0030] According to other such embodiments, which utilize supervised machine learning, senescence scores, detailed below, may be used for senescence classification, e.g., according to senescence subtype and / or senescence susceptibility. According to some embodiments, senescent subtype classification allows for quantification and validation of expressions of protein-based biomarkers per cell based on their morphologies.

[0031] Some embodiments utilize supervised machine learning, e.g., a trained neural network, to assign a senescence score to a patient cell based on morphological features. A non-limiting example such embodiment is described in reference to training the Xception neural network architecture to classify cells and quantify a senescence score per cell based on raw textured images of F-actin (cell) and DNA (nuclei). The score may be used not only to identify senescence, but also classify a subtype and / or susceptibility to senescence. The supervised machine learning and senescence score described above - which relies on training cell images labeled with indications representative of senescence or not senescence, but where an output for a novel patient input cell image is non-binary, e.g., a senescence score ranging between zero and one - solves the oversimplification problem of the prior art, in which senescence is considered in a binary manner as either being present or absent.

[0032] Using a senescence characterization generated by an embodiment, the inventors showed that the senescence burden across cell populations were dependent on the mode of senescence induction, time post-induction, and the age of the donor. To establish the association between senescence, senescence subtypes, and age, the inventors used primary dermal fibroblasts from an expanded cohort of fifty healthy donors ranging in age from 20-89 years old. These associations revealed that not only did the amount of senescent cells increase with age, but also a senescence subtypeidentified herein as C10 showed the strongest age-dependence. Furthermore, the observations revealed that baseline senescence burden was indicative of the senescence score after exposure to senescence-inducer doxorubicin.

[0033] Further, the inventors determined that the identified senescence subtypes were functionally different by profiling the responses of senescence cells to a small panel of senotherapies with focused analysis on responses to Dasatinib + Quercetin (D+Q). Results showed that senescent cells responded based on four dominant patterns: short death, long death, stable viable, and unstable viable. Senescent cells treated with D+Q exhibited higher abundance in short death, long death, and unstable viable, with cells in C7 being the most sensitive to death after D+Q exposure.

[0034] Thus, embodiments may use unsupervised and / or supervised machine learning to characterize a patient cell in a non-binary manner according to senescence subtype, senescence susceptibility, and / or senescence score, rather than simply performing a binary classification, and exploit the identification to select clinical therapeutic type and / or dosage.

[0035] These and other features and advantages are shown and described herein in reference to the figures, and in reference to a study that utilized non-limiting example embodiments. A description of the study follows.

[0036] I. Comprehensive profiling of cellular senescence across multiple induction conditions and biomarkers

[0037] In general, cellular senescence can be induced across ex vivo cell populations based on a variety of methods that take advantage of core mechanisms such as DNA damage responses (DDR), replication stress and telomere shortening, oncogenic activation (i.e., p53 and RAS), and epigenetic modifications. Torecapitulate the range of physiologically-relevant senescence phenotypes, the inventors established a framework using primary dermal fibroblasts isolated from a 23- and an 89-year-old, referred to as young and old, respectively.

[0038] Fig. 1 is a simplified graphical illustration of an experimental workflow 100 to induce and characterize senescent cells, as used in the study and according to various embodiments. To profile senescence, the inventors exposed cells from both donors to optimized concentrations of four chemical senescence inducers (i.e., Bleomycin (BLEO), Doxorubicin (DOX), Atazanavir (ATV), Hydrogen peroxide (H202)), and serially passaged cells (REP) until proliferation arrest. After induction, the senescence phenotype was allowed to develop over the course of eight days, after which cells were plated at low density onto Collagen-I coated glass bottom dishes, fixed and stained for proteins / biomarkers of interest, then imaged for downstream single-cell morphology analysis.

[0039] Fig. 2 provides representative images 200 of the immunofluorescent staining of five senescence-associated protein biomarkers across untreated (DMSO vehicle) and DOX-induced primary dermal fibroblast samples (age 23 male), according to various embodiments and as used in the study. For all experiments in the study, unless otherwise stated, cells were stained with Hoechst 33343 (H33342) to delineate nuclear boundaries, Phalloidin to delineate cell boundaries, senescence biomarker p21 , and either Lamin B1 (LMNB1 ), High Mobility Group Box 1 (HMGB1 ), p16, or betagalactosidase as additional biomarkers of senescence. To quantify single-cell morphologies, baseline uninduced cells from both donors and the senescence- induced cells were segmented using a CellProfiler pipeline, followed by a postprocessing pipeline for data curation (i.e., removing mis-segmented cells and outliers) and analysis as described further herein.

[0040] Altogether, the study used an initial dataset of >50,000 single-cells, each having information on cell and nuclear morphologies with the expression of multiple protein-based senescence biomarkers for both ages across five senescence-induction conditions.

[0041] II. Establishing single-cell senescence phenotypes across multiple biological conditions

[0042] To provide in-depth profiling of cells across baseline uninduced and senescence-inducing conditions, the inventors quantified the expression of co-stained senescence biomarkers, cellular secretions, and parameters describing the average morphologies for populations of cells and their corresponding nuclei. To establish and validate the senescence phenotypes, the inventors first evaluated whether there was a halt in proliferation for senescence induced cells relative to the baseline uninduced cells. Given the observed proliferation arrest of senescence induced cell populations, the inventors then quantified the cellular expressions based of protein-based senescence biomarkers in cells from both donor across five biological conditions. As expected, an increase in the average expression of senescence biomarkers betagalactosidase, p16, and p21 , and a decrease in the expression of HMGB1 and LMNB1 for both donors relative to their baseline uninduced cells was observed, as shown in Fig. 3.

[0043] Fig. 3 illustrates box plots 300 for Z-score fluorescence quantification relative to control of |3Gal, p16, p21 , HMGB1 , and LMNB1 for young and old dermal fibroblast samples (n>400 single cells per condition, mean ± 95% Cl, One-Way ANOVA relative to age-matched control expression), as used in the study and according to various embodiments. Interestingly, the extent of increase or decrease in the biomarker expressions was dependent on the mode of senescence inductionand the donor age. Together, these results show that the mode of senescence induction matters, and that it contributes to differential abundances of senescent and non-senescent cells per condition.

[0044] To further establish a senescence phenotype, the inventors profiled the cellular secretions at baseline and eight days post induction. Secretion analysis was performed using the Bruker CodePlexTM multi-analyte secretory panel, which contained overlapping analytes (cytokines, chemokines, etc.) to those identified in the core senescence associated secretory phenotype (SASP4,18), such as IL-6, IL-7, IL- 8, IL-15, and Ml P-1 a. The abundance of secreted factors was quantified from the conditioned media across each condition by culturing a known number of cells for 48 hours, harvesting the conditioned media, and profiling the un-diluted conditioned media. From this analysis, twenty-two secreted factors across ten condition were profiled. Results showed that uninduced baseline controls for both ages clustered together and were defined by low expression of SASP-related and other pro- inflammatory proteins, as shown in Fig. 4.

[0045] Fig. 4 depicts a heatmap 400 of standard scaled expression of pro- inflammatory secretions for induction conditions (average algorithm, Euclidean linkage), as used in the study and according to various embodiments. As shown in Fig. 4, the senescence induced conditions clustered together with elevated expressions of SASP-related proteins. Consistent with the protein-based biomarkers, induction-dependent trends in the secretions were observed. For instance, IL-7 was highly secreted by cells from both the young and old donor treated with DOX but was expressed at a lower level when cells were exposed to H2O2. Together, analysis of cellular secretions provides additional confirmation of bulk senescence phenotypes across the tested induction conditions.

[0046] Further, given the fact the microscopy images of cells and nuclei were collected across all conditions, the senescence-associated changes in cell and nuclear morphologies was quantified. A significant increase in the sizes of both cells and nuclei post-induction was observed, as shown in Fig. 5.

[0047] Fig. 5 shows a waterfall plot 500 of cellular and nuclear morphological features enrichment to either untreated or senescence-induced samples across both ages (cutoffs= Log2 > 10), as used in the study and according to various embodiments. For each cell within the dataset, the inventors quantified over 250 morphological parameters, which were reduced using factor analysis to 90 morphological parameters that described the size, shape, curvature, and roughness of both cells and nuclei. Table 1 below illustrates the parameters for images of nuclei, cells, and nuclei / cell combinations.

[0048] The features listed in Table 1 are non-limiting; according to some embodiments, features extracted using a neural network such as a convolutional neural network may be used.

[0049] To push the analysis beyond the increase in cell and nuclear size, the morphological features that were most strongly associated with uninduced cells and senescence-induced cells, respectively, were analyzed. Performing differentialmorphology analysis on the induced and uninduced cells, the inventors observed that the majority of the highly differential morphological features, on both sides, were associated with nuclear morphologies, as shown in Fig. 5. However, features describing nuclear shapes were more strongly associate with the uninduced cells and features describing the nuclear sizes were more strongly associated with senescence- induced cells. These results illustrate that nuclear features strongly encode senescence information. Given this associate between cell and nuclear morphology, biomarker expressions, and senescence, the inventors analyzed whether cell and nuclear morphologies were correlated with expression features of senescence biomarkers. To assess this, the inventors performed correlation analysis across all experimental conditions for the 90 cell and nuclear parameters and the biomarker expressions. Multiple correlations were observed among features given a threshold of 0.3 for the Pearson correlation coefficient.

[0050] The study thus showed, using multiple lines of evidence, that the biological conditions considered in the study yield multiple senescence phenotypes,thereby providing a basis to develop a single-cell framework of determining morphological and functional senescence subtypes, according to various embodiments.

[0051] III. Single-cell morphologies encode heterogeneous senescence phenotypes in primary dermal fibroblasts

[0052] The capacity of cell and nuclear morphologies to encode senescence phenotypes facilitated development of a single-cell framework to identify and classify morphological subtypes of senescent and non-senescent dermal fibroblasts. To establish a single-cell framework, cells from both donors across uninduced and senescence-induced conditions were imaged using high content microscopy, then cell and nuclear boundaries were segmented using an optimized pipeline in CellProfiler. Once segmented, parameters that described cell and nuclear morphologies were computed, as well as the expression of two protein-based senescence biomarkers (p21 and a combination of either p16, beta-galactosidase, HMGB1 , or LMNBI ).

[0053] With this data, the study pooled cells from twelve biological conditions, including cells from the young and older donor at baseline (uninduced) or postinduction with optimized concentrations and durations of ATV, DOX, BLEO, H2O2, and REP, as shown in Fig. 6.

[0054] Fig. 6 presents a schematic 600 of an morphological analysis pipeline based on unsupervised machine learning clustering, as used in the study and according to various embodiments. Briefly (details appear far below), cells are imaged and segmented, then values for a large number of morphological parameters are determined, which may be summarized by a vector of such values for each cell, then the dimensionality of the vector is reduced, and finally, the reduced-dimension vector is classified according to cluster proximity. Data visualization using two-dimensionalLIMAP dimensionality reduction revealed specific cell localization patterns per condition, with cells in uninduced conditions localizing to the lower left of the manifold and cells from induced conditions localizing towards the right of the manifold, as shown in Fig. 7.

[0055] Fig. 7 shows a scatter plot 702 of LIMAP space based on 87 morphological parameters for all conditions across both age groups (about 48,000 single cells; each dot represents a cell) and an overlay 704 of 11 optimal k-means clusters on top of the LIMAP space with representative cellular (outer) and nuclear (inner) morphologies, according to the study and various embodiments. Contour lines in the overlay 704 indicate the gaussian kernel density function for each cluster. As shown in Fig. 7, the scatter plot 702 showed a higher spread of the uninduced cells from the old donor towards the right of the manifold, suggesting a slightly higher abundance of senescent-like cells relative to the young donor. Although different inducers exhibited varying degrees of spread across the manifold, there were core overlap areas for all senescence inducers. To gain a deeper understanding of these spatial patterns across the manifold, k-means clustering analysis was applied to identify morphologically distinct groups of cells. K-means analysis revealed eleven distinct morphological clusters, each having characteristic cell and nuclear morphologies, as shown in the overlay 704. These eleven clusters are labeled C1 - C11 herein. Visual inspection of the cellular morphologies indicated a pattern of increasing size from left to right and increasing linearity of the shapes from top to bottom of the manifold. These results were confirmed by color-coding the magnitude of key morphological parameters onto the LIMAP, with increased cell and nuclear area from left to right and increasing cell aspect ratio (corresponding decrease in cell circularity) from top to bottom, as shown in Fig. 7.

[0056] Fig. 8 illustrates scatter plots 800 of select morphological features layered on top of the LIMAP space, according to the study and various embodiments. Of note, a cluster labeled as C9 contained a large fraction of mis-segmented cells, as such, C9 was kept as part of the quality control step (mis-segmented cluster) but omitted cells classified in C9 from formal analysis and reporting in the results.

[0057] With the morphological clusters C1 -C11 , the study assessed the presence and abundance of cells per morphological clusters across induction conditions. Unsupervised hierarchical clustering of induction conditions per cluster revealed the emergence of three groups based on the dendrogram branches. With uninduced conditions for young and old donors and young cells induced with ATV defining group 1 , both young and old cells induced with DOX, BLEO, and H2O2 defining group 2, and young and old cells induced via serial passaging (REP) and old cells induced with ATV defining group 3. Group 2 contained young and old cells treated with DOX, BLEO, and H2O2 exhibited enrichments in clusters C6, C7, C8, C10, C11 , whereas group 1 were depleted in those clusters but enriched for C1 , C2, C4, and C5. Group 3 contained a mixture of enriched morphological clusters, with noted overlaps with both groupl and group 2, and general enrichments in C4, C5, and C8. Furthermore, senescence induction conditions exhibited higher heterogeneity based on the Shannon entropy relative to uninduced conditions for both ages, suggesting that induced cells exhibited a wider range of morphologies relative to uninduced cells.

[0058] Thus, as illustrated by the study, some embodiments provide a robust framework to define senescence phenotypes across multiple biological conditions. The morphology-defined clusters reveal functional subtypes of senescent and nonsenescent cells, as described further herein.

[0059] IV. Profiling single-cell morphologies based on senescence biomarkers

[0060] Current approaches to identify senescence across cell populations rely on profiling the presence and expression of protein-based biomarkers, including p16, p21 , HMGB1 , LMNB1 , beta-galactosidase. For instance, senescence cells tend to express increased expression of p16, p21 , beta-galactosidase, and decreased expression of HMGB1 (translocation out of the nuclei) and LMNB1 relative to nonsenescent cells. While these expression patterns have been established, many studies evaluating senescence tend to profile only one or two of these biomarkers. To provide a comprehensive assessment of the expression patterns across the biological conditions tested, the study described herein profiled all five of the biomarkers listed above. Because the imaging workflow was only capable of imaging four fluorescence channels at a time, the study implemented a multiplexed strategy to stain p21 plus one of the remaining four biomarkers. This was iterated across all samples, and biological replicates were stained with different combinations until all five biomarkers were represented across each of the twelve biological conditions. With this workflow, the localization of cells expressing elevated or depleted levels of the senescence biomarkers within the LIMAP space was quantified.

[0061] To determine the thresholds per biomarker for gross positive and negative expression (i.e. , positive = senescent), the study first plotted and overlaid the distributions of single-cell expressions in the uninduced and senescence-induced conditions. Then the thresholds were identified on the basis that senescence cells will be enriched in the induced conditions and have a low representation in the uninduced cell population. Practically speaking, thresholds per biomarker were defined based on the magnitudes of the shifts in distribution post induction. Once these thresholds weredetermined, the inventors investigated how cells expressed high and low levels of each protein-based biomarker of senescence localized across the manifold and the morphological profiles across each cluster. The presence and spread of “hot” and “cold” zones was quantified, denoting senescent (according to at least one of the multiple senescent phenotypes) and non-senescent cells. Results showed a clear separation of cells classified as senescent and non-senescent, with senescence emerging from left to right of the manifold (large cells with large nuclei were localized to the right). Note that, while the contours showed clear overlap areas among all senescence biomarkers, the unique topologies of each biomarker suggest that some biomarkers may identify cells as senescence that others may missed. For instance, the contours for senescent cells identified by p16, beta-galactosidase, and LMNB1 had less spread across the manifold relative to p21 and HMGB1 , meaning that HMGB1 and p16 will identify more cells as senescent relative to using p16, LMNB1 or betagalactosidase. Whether p21 or HMGB1 is more accurate remains to be determined, however, the union of the biomarkers likely best captures the senescence phenotypes.

[0062] The inventors also quantified the average expression of each senescence biomarker per cluster. Results showed that clusters C7, C10, and C11 had high relative expressions of p21 , p16, and beta-galactosidase and low expression of HMGB1 and LMNB1 , suggesting an enrichment for senescence within those clusters. Clusters C1 , C2, C4, and C5 had the opposite expression pattern, suggesting that they were enriched for non-senescent cells. Lastly clusters C3, C6, and C8 had intermediate expressions of the senescence biomarkers, which suggests that those clusters harbored an intermediate state or cells that may be potentially poised toward, or susceptible to, senescence. In summary, these findings highlight a strong relationship between cell and nuclear morphologies and the expression of senescencebiomarkers across the morphology-defined clusters described herein, underscoring the notion that morphology robustly encodes the heterogeneous senescence phenotypes.

[0063] V. Single-cell morphology robustly encodes senescence biomarker expressions

[0064] Given the robust link between single-cell morphologies and senescence biomarker expressions, the study investigated whether cellular morphologies may be linked to directly predict the biomarker expression for all five biomarkers per cell. To test this, the study developed and implemented an imputation-based label propagation scheme using k-nearest neighbor analysis. Briefly, the imputation scheme was constructed by leveraging cell and nuclear morphologies to determine the senescence biomarker expressions based on the proximity to cells having known expressions of the biomarker of interest, as shown in Fig. 9.

[0065] Fig. 9 is a flow diagram for workflow 900 for protein imputation as used in the study and according to various embodiments. For example, to impute the expression of p16 in a cell stained for p21 and LMNB1 (not stained for p16), the study identified the 20 nearest cellular neighbors that had measured expression of p16, then weighted the expressions based on the Euclidean distance from each of the 20 neighbors. This was iterated for every cell across all senescence biomarkers until an expression profile for each of the five senescence biomarkers per cell was obtained. This produced, for each of the 50,000 single cells, two measured and three imputed biomarker expression profiles. Results from the imputation scheme proved to be robust, with prediction accuracies over 85% for each of the senescence biomarkers relative to ground truth expressions (i.e., the nearest cells with measured expression of the biomarkers).

[0066] The study revealed that some cells had lower imputation accuracies, particularly for HMGB1 and LMNB1 ; beta-galactosidase, p21 and p16 had fewer of those cells. This may be attributed in part by those biomarkers having a lower dynamic range for senescent versus non-senescent cells, and that the distances of cells from a neighbor with known expression was larger. However, even with the subset of cells having lower accuracies, the pattern of localization for the imputed cells across the dimensionality-reduced manifold was consistent across both ages and for cells with measured expression per senescence biomarker.

[0067] Given this comprehensive dataset with all five senescence biomarkers per cell, the study considered how the biomarkers interfaced with each other and whether certain biomarkers identified senescence better than others. To answer this, the inventors took all cells, across all biological conditions, and grouped cells based on the number of senescence biomarkers that were positive based on the previously described thresholding strategy. Then within each of the comparisons (i.e. , 2, 3, 4, or 5 senescence biomarkers) the inventors quantified the fraction of cells with each of the biomarkers. This revealed that for all the cells having two positive senescence biomarkers, approximately 50% of cells were classified as senescent based on p21 and HMGB1. Of the cells classified as senescent based on three senescence biomarkers, over 70% were positive to beta-galactosidase, HMGB1 and LMNB1. Furthermore, of the cells considered senescent based on four biomarkers, betagalactosidase, HMGB1 , and LMNB1 were expression in more than 80% of them, with p16 being the forth biomarker having expression in approximately 70% of the cells. Within this framework, the inventors mapped the localization of cells with varying number of senescence biomarkers from zero (i.e., non-senescent cells) to five (senescence based on all five biomarkers). Qualitative results recapitulated the shiftfrom left to right from zero to five markers, with a tight spread corresponding to the overlap areas across all the biomarkers.

[0068] Taken together, these results show that information on the protein-based senescence biomarker expressions are nested within the senescence-associated morphologies analyzed according to the disclosed machine learning techniques, with the capacity to impute the biomarker expressions based on cell and nuclear morphologies.

[0069] VI. Machine learning strategies reveal morphological subtypes of senescence

[0070] The above describes robust associations between cell and nuclear morphologies and senescence phenotypes at single-cell resolution, which were generated from parameterizing cell and nuclear morphologies based on size, shape, curvature, and boundary roughness, among other morphological features (see Table 1 ). To extract information that may not be captured by parametrized morphological features, the inventors also implemented a supervised machine learning scheme using the raw textured images of cells (stained F-actin) and nuclei (stained DNA) across individual cells, so as to generate a non-binary senescence score per single cell.

[0071] Fig. 10 is a schematic diagram of a method 1000 of using cellular and nuclear textured images to develop single-cell resolution senescent scores, ranging in value from zero to one, as used in a study and according to various embodiments. Training was conducted on an age, induction, and biomarker balanced cohort of 6,000 single cells. First, the inventors constructed a three-layer image data instance for each individual cell, where the first layer was the cell’s nuclei, the second layer was the cell’s F-actin structure, and the final layer including a composite image both the F-actin andnuclei structures. Each of these image data instances per cell were cast onto a blank canvas having dimensions of 1526 by 1526-pixel, with the geometric centroid of each object (nuclei, cell, composite) mapping to the center of the canvas. Each data instance was intended to captured not only the ensemble of possible morphologies associating with senescence, but that it also offered unbiased insights into which properties were most important to the senescence identification and classification. Each image data instance was labeled as either ‘non-senescent’ or ‘senescent’ based on the thresholded expressions of at least two senescence biomarkers. Second, the model was trained using the Xception14 neural network architecture with a SoftMax output, by way of non-limiting example. According to other embodiments, a different image analysis neural network may be used in the alternative. The SoftMax output, ranged from ‘0’ to T based on the probability of a cell being non-senescent or senescent according to the predefined threshold biomarker expressions. The inventors implemented a balanced cohort of 6000 data instances spanning all induction conditions, both ages, and all five senescence biomarkers, with the model trained for 220 epochs. Third, the trained model was able to predict senescence phenotypes based on the inferred scores. The trained model had an average accuracy of 89% for both the training, test, and validation datasets. To further demonstrate the robustness of the model, the inventors plotted the Receiver Operator Curve (ROC), which yielded an area under the curve (AUC) of 0.95.

[0072] While previous studies using morphology to identify senescent versus non-senescent cells have often used either nuclear or cell morphology, there has been limited investigation using both cell and nuclear morphologies. To address this in an unbiased manner, the inventors converted both training and test sets to include either images of nuclei or cell only. The resulting datasets were used to iteratively train newneural network implementations for 50 epochs, and the accuracies based on proteinbased biomarker expressions were computed. Notably, both models resulted in statistically similar accuracies, indicating that both cell and nuclear morphologies both provide useful information on senescence phenotypes. To further illustrate this, the classification error between the senescence scores per cell and the senescence classification per biomarker was computed. Results showed an error of less than 0.1 across all senescence biomarkers, indicating that the Xception model accurately captured the senescence phenotype in an unbiased manner.

[0073] While the trained neural network implementation robustly quantified real- valued senescence scores ranging from 0 to 1 for each cell, a rigid threshold above or below which a cell could confidently be determined as senescent or not is useful. The inventors therefore quantified how specific scores correlate with the accuracy of the classification. In other words, the inventors addressed how the error of the classification changes for a cell having a senescent score of 0.6 versus 0.7, for example. To probe this, the inventors binned the range of senescent scores between 0 and 1 with increments of 0.02 then assessed how many cells in each bin were correctly classified based on the biomarker expressions. From this analysis, the inventors selected a score above 0.88 as senescent and below 0.1 , corresponding to a prediction error below 12%. To visualize how the model to determine senescence scores related to the morphology-defined clusters, the senescence score per cell was projected across all clusters, as shown in Fig. 11 .

[0074] Fig. 11 illustrates mean senescence scores 1100 of each k-means cluster overlayed on top of the LIMAP space, according to the study and various embodiments. Consistent with senescence phenotypes based on biomarker expressions per cluster, clusters C1 , C2, C4, C5 exhibited low senescence scores(<0.3), C7, C10, C11 exhibited high senescence scores (>0.9), and C3, C6, C7, C8 exhibited intermediate scores (0.6-0.9). Thus, combining the Xception trained model with morphological parameters and senescence biomarkers provides a robust classification of senescence, with senescence scores increasing from left to right of the manifold and clusters C7, C10, C11 exhibiting scores above 0.9 and large cell and nuclear morphologies.

[0075] Given the high senescence scores for cells in C7, C10, and C11 , the inventors re-evaluated the idea that different induction conditions result in differential senescence phenotypes.

[0076] Fig. 12 is a heatmap 1200 describing the differential enrichment amongst the three morphological senescence subtypes across all experimental conditions (average algorithm, Euclidean linkage), according to the study and various embodiments. Results indicated induction-dependent senescence phenotypes with varying abundance of cells in each of the three senescence clusters. For instance, cells from the young donor induced with BLEO harbored 22% and 21 % of cells in C7 and C10, respectively. Whereas cells from the old donor induced with H2O2 harbored 23% of cells in C11. Taken together, these results show the robust identification and classification of senescence phenotypes bases on morphology, with validation metrics based on multiple protein-based senescence biomarkers. Furthermore, the identified morphology clusters exhibit varying senescence scores, with C7, C10 and C11 having a robust senescence phenotype. In the following, and by way of non-limiting example, clusters C7, C10, and C11 are referred to at the three senescence subtypes.

[0077] VII. Profiling induction-specific kinetics to developing senescence phenotypes

[0078] While the above analysis provides a robust framework to describe morphological patterns of cellular senescence, it was developed on snapshots at a single timepoint (day 8). To establish robust temporal patterns across senescence progression, the inventors performed senescence induction on cells from the young donor using DOX and H2O2, with snapshots of morphologies collected at multiple time points, specifically at 0, 2, 5, 8, 12, and 15 days post-induction. At each of these time points, the experimental and imaging workflows described herein were repeated to collect information on cell and nuclear morphologies, as well as the single-cell expressions of the five senescence biomarkers. From this analysis, differences in the senescence progression based on whether cells were treated with DOX or H2O2 were observed. For the DOX-induced condition, cells exhibited a steady progression towards a senescence phenotype, with a decrease of cells in C1 , C2, C4, and C5 as a function of time post induction, and a complimentary increase of cells in C3, C6, C7, C10, and C11 up until day 15. H2O2-induced cells on the other hand exhibited a biphasic progression, with cells in C1 , C2, C4, and C5 decreasing up to day 5 then steadily increasing again. Simultaneously, cells in C6, C7, C10 and C11 followed the opposite pattern, with an increase up to day 5 and a decrease thereafter until day 15. Here the depletion of cells within clusters is not due to the cells dying, but due to a fractional redistribution of cells within clusters as they transform towards senescence. Furthermore, an apparent reversal of the senescence phenotype over time for cells treated with H2O2 is not that cells are going from senescent to non-senescent, but there is a small fraction of cells that do not transform and over time proliferates to recover the population.

[0079] To quantify the trends in the senescence subtypes, the inventors quantified the senescence enrichments across the cell populations in both DOX andH2O2 induced cells. This confirmed that in DOX, there was a steady increase in the senescence enrichment and a bi-phasic pattern for H2O2. Next, quantifying the temporal patterns for enrichment in each of the three senescence subtypes, the inventors observed that for DOX there was a decrease in C7 and increases in C10 and C11 to convergence of approximately 30% in each subtype, which was not as cleanly observed for H2O2. To clarify these results, the inventors also quantified the temporal patterns of expression in each of the senescence biomarkers for both DOX and H2O2 conditions. Consistently, the results show that at the population level for each inducer that at an early time point there is a deviation in the expression from the baseline uninduced cells. However, at later time points, H2O2 regains uninduced levels. Taken together, results suggest that the temporal patterns of senescence are inducer-dependent, with cell populations fractionally redistributing among senescence subtypes during the progression. Furthermore, H2O2 may be a weaker inducer relative to DOX, with a fraction of the population not entering senescence and as a result the population recovers as the cells begin to proliferate at late time points.

[0080] VIII. Profiling senescence subtypes and susceptibility as a function of cell donor chronological age

[0081] It is known that the abundance of senescence cells increases with increasing chronological age. To quantify this in the context of the senescence subtypes described herein, the inventors procured a gender-balanced cohort of primary dermal fibroblasts from fifty aging individuals with an age range of 20-89 years old. To confirm the trend of senescence accumulation with age, the inventors first profiled the morphologies of baseline uninduced cells for each of the fifty donors and plotted the relationship between average senescence score with their chronological age. Results showed a significant correlation between the senescence score and theage of the donor with a Pearson correlation coefficient of 0.33. This trend was conserved for both men and women, with a Pearson correlation coefficient of 0.32 and0.34, respectively. With this result, the inventors analyzed whether certain senescence subtypes were more strongly associated with chronological age relative to the other subtypes. Plotting the fraction of senescence cells within each senescence subtypes as a function of chronological age, the inventors found that C10 had the strongest age-association, with a Pearson correlation coefficient P=0.45, followed by C11 (P=0.30) and C7 (p=-0.16). Furthermore, other morphology clusters also exhibited age-associations, such as C8 (P=0.42). Although C10 showed the strongest age-associated relationship, it comprised a minority of the total senescence population, with fractions ranging from 2.5% to 17.5% across all ages tested. C7 on the other hand comprised the majority of the population with a range of 20% to 90%, and C11 ranging from 0.5% to 3.5%. Interestingly, from the initial induction experiments, the inventors found that C7 and C11 were the most dominant senescence subtypes, suggesting that they may be strongly associated with damage responses rather than chronological age. Together, these results indicate that the morphology-based classification of senescence recapitulated the overall age-related accumulation of senescence. However, not all senescence subtypes exhibit the same degree of age-dependence.

[0082] Given the relationship between cell donor age and senescence score, the inventors analyzed whether cells from older donors were more susceptible to senescence induction when exposed to a senescence inducer. To test this, the inventors chose to use DOX, given the clinical relevance to the treatment of cancer. From the same expanded cohort of aging samples, the inventors selected twenty samples that spanned the entire age range from 20-89 and the spread of senescencescores. With each of these aging samples, the inventors induced senescence using the same optimized induction protocols and profiled the senescence score as a function of chronological age. Plotting the average DOX-induced senescence score versus age, the inventors observed a non-significant correlation, with a Pearson correlation coefficient of 0.19. Furthermore, looking back at the relationship between the senescence scores and the age, this result is not necessarily surprising given the large donor-to-donor variability in the senescence scores, even for donors of similar chronological ages.

[0083] To further elucidate this result, the inventors re-investigated the distributions of senescence scores for cells from each donor, pre and post DOX- induction. Although the spread of distributions recapitulated the variability among donor cells, with a shift towards higher average scores post-induction, the inventors observed an inverse relationship between the spread of the senescence scores and time-point of induction (pre or post). Meaning, that if baseline uninduced cells had a tight but heavy tail distribution of senescence scores, the post-induction scores would have a wider spread of scores. Similarly, if the distribution was wider at baseline, it tended to be tight post-induction. Given this, the inventors considered that if chronological age by itself did not directly associate with senescence scores post induction, whether the baseline senescence score was indicative of senescence score post induction. Plotting the senescence scores at baseline and post induction, the inventors observed a significant correlation with P=0.54, suggesting that the baseline senescence score was indicative of post-induction score, and by extension senescence susceptibility.

[0084] Together, these results support the notion that although the senescence burden across cell populations accumulate with increasing patient age, therelationship between chronological age and senescence susceptibility is more complicated. Given that chronological age positively correlates with baseline uninduced senescence scores, but chronological age is not significantly correlated with senescence score post senescence induction. In any case, the baseline uninduced senescence score is indicative of a cell population’s susceptibility to senescence.

[0085] IX. Senescence subtypes exhibit differential responses to Dasatinib + Quercetin

[0086] With the emergence of senotherapies as clinically viable strategies to mitigate senescence-associated dysfunctions, the inventors assessed whether senescence subtypes exhibited differential responses to a small panel of senotherapies. To test this, the inventors profiled the responses of senescent cells to optimized concentrations of five clinically relevant senotherapies comprising both senolytics (D+Q, ARV-825, and Navitoclax (ABT263)) and senomorphics (Fisetin, Metformin) over the course of three days. For this experiment, the inventors induced cells from the young donor (GT22) with DOX as previously described, the cultured cells for eight days so that senescence could develop. On day 7, cells were seeded at low density onto Collagen-I coated 96-well glass bottom dishes, and on day 8 cells were stained with live cell markers (Spirochrome SPY650FastAct and SPY555-DNA to delineate cell and nuclear boundaries) and plates were mounted onto the microscope for live cell imaging. Over the course of 72 hours, images of each condition (5 senotherapies and untreated) were acquired every hour, with 60 hours of post-processing movies used for analysis. Cells within each of the conditions were segmented at every time point and were assigned to a morphology cluster and senescence subtype after projection into the LIMAP space. Live cell imaging wasperformed on the senescent cells exposed to senotherapies specifically to map the dynamic response patterns at single cell resolution over time. Across all experiments particular attention was paid to whether cells died, whether they exhibited plasticity among morphology clusters (i.e., transitioned among senescence subtypes or the other morphology clusters), or whether they were stable in their morphological cluster.

[0087] With these time-lapse movies, the inventors profiled bulk viabilities across the six tested conditions, which showed a decreased viability for cells treated with D+Q, Navitoclax, and ARV-825, with Fisetin and Metformin having similar viabilities to the untreated controls. Additionally, while most of the tested conditions had an average senescence population of ~60% post induction and treatment, D+Q had a lower fraction of ~40% and was the only treatment condition showing a bulk decrease in cell count for all three of the senescence subtypes. Given this result and the relevance to ongoing clinical trials of D+Q, the inventors performed a deeper analysis on D+Q responses. The inventors thus quantified the fraction of senescence cells at the population level and then within each of the three senescence subtypes C7, C10, and C11 as a function of exposure time for the untreated and the D+Q treated cells. Results indicated that there was a decrease of ~8% of senescence cells over the 60 hours, being largely attributed to spontaneous cell death. C10 and C11 exhibited stable trends over the 60 hours, with C7 exhibiting a decrease to ~73%, which can be explained by a combination of both spontaneous cell death and transition from C7 to other morphological clusters. On the other hand, D+Q treated senescent cells exhibited ~45% decrease in the overall fraction of senescent cells, and a complimentary decrease in the abundance of all three senescence subtypes. Interestingly, C10 had a much less decrease after D+Q treatment relative to C11 and C7, with C7 having the largest decrease over the 60-hour treatment.

[0088] Quantifying these responses further, the inventors constructed a singlecell response matrix for all senescent cells classified as C7, C10, or C11 at the initial timepoint for both untreated and D+Q conditions. Next, the inventors performed unsupervised-learning-based hierarchical clustering along the time series to identify response patterns. Qualitatively, the inventors observed unique and shared response patterns across both conditions. To allow for direct comparison among responses for untreated and D+Q senescent cells, the inventors pooled the responses of both conditions then re-clustered the time series data using hierarchical clustering. Results yielded four response patterns that were largely defined based on cell viabilities. The inventors named these responses: long death, short death, stable viable, and unstable viable. Short death referred primarily to cells cleared within the first 30 hours, long death were cells cleared after much longer time points (>30 hours), stable viable were cells that remained viable but exhibited little to no transitions to other morphology clusters or other senescence subtypes, and unstable viable were cells that remained viable but exhibited high temporal plasticity and transitions to other morphology clusters or senescence subtypes.

[0089] Next, the inventors quantified the abundance of cells within each of the response patterns from the untreated and D+Q conditions. As expected, results showed that senescent cells treated with D+Q made up most of the long and short death classified cells, with 64% and 71 %, respectively. Of the cells remaining viable over the 60 hours, the stable viable cells came mainly from the untreated condition (80%, respectively), with 55% of the unstable viable cells coming from the D+Q treatment. These results suggest that D+Q treatments drove responses of either cell death or instability across the morphological phenotypes. Investigating further the influence of starting senescence subtypes, the inventors observed that C7 wasdominate in both the long and short death, comprising 56% and 41 % of the cells. C10 was most dominant in the stable viable comprising 54% of cells, and all three of the senescence subtypes were similarly represented in the unstable viable response group. The inventors also found that for cells classified as stable viable starting in C7 exhibited some transitions primarily to C10 and C11 , on average spending 28.7% and 27.9% of the time in those clusters, respectively. Cells starting in C10 stayed mainly in C10 (63.4%) and C11 (23.9%), and cells starting in C11 either remaining in C11 (46.2%) or transitioning to c10 (39.9%). Of the unstable viable cells, senescent cells starting in C7 spent most of their time in either C7 (53.7%), C6 (21.2%), or C4 (7.9%). Senescent cells starting in C10 were morphologically very heterogeneous, spending most of the time in either C7 (30.5%), C10 (23.6%), C6 (17%), or C11 (9.3%). Lastly, senescent cells starting in C11 spent most of their time in C6 (37.1 %), C11 (20.1 %), C7 (19.5%), or C10 (5.1 %).

[0090] Taken together, these results indicate that senescent cells within each of the three senescence subtypes C7, C10, and C11 respond differently to D+Q, establishing that these are indeed functional subtypes of senescence. Notably, senescent cells in C7 were most responsive to D+Q, with cells either dying or becoming morphologically unstable after treatment. C10 senescent cells were most resilient to D+Q treatment. These findings also demonstrate the utility that ex vivo morphological profiling and senescence subtyping is a powerful tool to screen and map response patterns to senotherapies.

[0091] X. Discussion Regarding the Study Results

[0092] Cellular senescence is a heterogeneous phenotype. The study included the development of a robust single-cell approach to classify functional subtypes of senescence. Leveraging single-cell profiling of cellular morphology with machinelearning, the study classified a panel heterogeneous aging dermal fibroblast across multiple induction conditions into eleven morphology clusters, out of which three bona fide senescence subtypes, referred to herein as C7, C10, and C11 , were identified. These findings robustly demonstrate that: (a) the expressed senescence phenotypes are dependent on the mode of senescence induction and the time post senescenceinduction; (b) single-cell morphologies encode senescence phenotypes and can be used to determine the protein-based biomarker expressions using imputation via k- nearest neighbor approaches; (c) it is not appropriate to assume that all cells exposed to senescence inducers are senescent, since not all cells transform towards senescence and that cell populations can recover (e.g., H2O2) over time postinduction; (d) senescence subtypes do not all exhibit similar degrees of cell-donor agedependence, with C10 being most strongly associate with chronological age and most resilient to D+Q treatments; (e) chronological age by itself does not determine a cell’s susceptibility to senescence induction donor-to-donor variabilities that drives heterogeneity of senescence; and (f) senescence subtypes exhibit differential responses to senotherapy D+Q. The study thus provides a robust and promising approach to ex vivo senescence profiling and potential applications in precision medicine.

[0093] Prior studies typically consider senescence as a binary cellular state (cell as senescence or not) and not as a non-binary heterogeneous collection of phenotypes. Building from this, the inventors investigated whether senescence subtypes existed, and whether their existence could help explain the heterogeneous and context-dependence of senescence phenotypes. Importantly, the inventors’ approach provides deep insights and practical applications of ex vivo senescence phenotypes and cellular morphologies. Furthermore, because the inventors’ approachis high-throughput and cost-effective, with the capacity to interrogate both snapshot and dynamic responses, it has direct applications for next-generation screening across a wide range of pro- and anti-senescence perturbations. Furthermore, it can be directly integrated with other image-based or next-generation sequencing-based technologies to identify subtype-dependent molecular vulnerabilities to target senescence phenotypes. Yet further, coupling an embodiment with modem spatial multi ‘omic’ technologies can lead to new putative senotherapies and the development of novel approaches, for instance using engineered cell products like CAR-T cells to target senescence subtypes based on specific single cell targets.

[0094] The study also characterized senescence subtypes as morphology clusters having high average senescence scores (>0.9). While non-senescent clusters (C1 , C2, C4, C5) did harbor low senescence scores (<0.3), there were a few clusters, namely C3, C6, and C8 having intermediate senescence scores ranging between 0.6 to 0.82. These cells may be poised towards, or susceptible to, senescence, exhibiting a higher tendency towards senescence transformation, and may have associated molecular features that describe these phenotypes. Thus, senescence scores themselves may be used for senescence subtype classification, according to some embodiments.

[0095] XI. XGBoost embodiments

[0096] The inventors developed two embodiments that utilized the XGBoost random forest architecture to directly classify a senescence subtype of an immobilized cell. To do so, the inventors trained the XGBoost implementations using labeled images, where the labels were determined according to the clusters identified using a unsupervised learning technique (which considered both nuclear features and cellular features) as described herein. A first implementation was trained using only nuclearfeatures. A second implementation was trained using only cellular features. After the implementations were trained, they was able to classify novel cells.

[0097] Fig. 13 depicts confusion matrices 1302, 1304 for the XGBoost implementations. On the matrices 1302, 1304, the x-axis represents the predicted cluster (C1 -C11 ) and the y-axis represents the true cluster. The confusion matrix 1302 shows the results for prediction based only on nuclear features, and the confusion matrix 1304 shows the results for prediction based only on cellular features.

[0098] XII. Comprehensive description of material and methods

[0099] This section presents certain details regarding the study described herein. Note that the details are non-limiting examples. For example, other embodiments may utilize different systems, values, and / or techniques.

[0100] Cell culture. Human dermal fibroblasts were procured from the Baltimore Longitudinal Study on Aging (BLSA) and served as the experimental basis for the study. 50 primary cell samples from a gender-balanced cohort with ages ranging from 23 to 89 years old were used. Primary cells were cultured under standard conditions on cell culture-treated culture dishes (Coming, 353136) or on Collagen-I coated glass bottom dishes (Cellvis, P24-1.5H-N) with Gibson High-Glucose Dulbecco’s modified Eagle’s medium (DMEM, Thermo Fisher, 11995073) supplemented with 15% fetal bovine serum (FBS, Thermo Fisher, A5256801 ) and 1 % penicillin-streptomycin (Thermo Fisher, 15070063). Cells were grown at 37° Celsius in a humidified incubator with 5% CO2. Cells were passages on average every 3-4 days. Once cultures reached ~70% confluency, they were plated onto 10 cm cellculture treated petri dishes (Coming, 430167) at a density of 100,000 cells / dish and allowed to adhere overnight prior to senescence induction. Low passage cells(maximum of 4 passages from stock vial) were used for all experiments in the study except when stated otherwise (i.e., for replicative senescence experiments).

[0101] Induction of senescence. All drugs used for senescence induction were diluted from their respective stock solutes in DMSO (<0.2% vol / vol) according to manufacturer guidelines and subsequently diluted in media to reach the final working concentration for senescence induction (see Table 2 for respective concentrations). Briefly, cell culture media was removed from the petri dishes and replenished with 8 mL of induction media (culture media supplemented with respective concentrations of senescence inducers) and incubated for the specified time durations (see Table 2). Following induction, cells were washed twice with 1X Phosphate-Buffered Saline (PBS, Corning, 21-031) before addition of 8 mL of fresh culture media. Senescence was allowed to develop, with cellular profiling performed at various time points depending on the experiment; (either 2, 5, 8, 12, or 15 days post induction). For each of the specified induction time points cells were subsequently plated onto 50 ug / mL collagen I (Corning, 354249) coated 24-Well glass bottom plates (Cellvis, P24-1.5H- N) at a density of 1250 cells / well and a volume of 1ml per well to allow for a sparse density and identification of single cell morphologies.Table 2: Senescence induction

[0102] The end points for replicative senescence per sample (GT22 or GT125) were determined based on a sustained plateau in proliferation rate for a one-week.

[0103] Profiling cellular secretions. To profile the secreted factors from biological conditions of interest, senescent and non-senescent cells were plated at a density of 20,000 cells per 10 cm petri dish and allowed to adhere overnight. The following day, the media was replaced with fresh media and incubated for 48 hours to allow the accumulation of cellular secretions. After 48 hours, the conditioned media was collected, and cells were counted per condition (i.e. , per 10 cm dish). Conditioned media was then profiled using Bruker CodePlex Human Adaptive Immune Panel (Bruker Cellular Analysis, CODEPLEX-2L01 -4) to quantify the expression of 22 pro- inflammatory markers that overlaps with the documented secretions of senescent cells (SASP — senescence associated secretory phenotype). The raw readouts were calibrated to instrument controls and normalized using the individual cell counts to allow for direct comparisons of secreted factors across biological conditions.

[0104] Immunofluorescent staining. Prior to any immunofluorescence staining and quantifications, cells per condition were seeded at low density (1250 cells per well of a 24-well plate) and allowed to completely adhere overnight. Once cells had adhered, they were subsequently washed with 1X PBS and fixed in freshly prepared 4% (w / v) paraformaldehyde (PFA, Electron Microscopy Services) for 12 minutes. Cells were then rinsed 3x with PBS before being permeabilized by incubation in a 0.1 % Triton-X in PBS solution for 10 min. Cells were again rinsed 3x with PBS to remove any residual Triton-X solution, then blocked with 2% bovine serum albumin (BSA, weight / volume in 1X PBS) for at least 30 minutes at room temperature. Following this blocking step, cells were incubated with various primary antibodies, including p16 (Abeam, ab108349, 1 :250), LMNB1 (Abeam, ab229025, 1 :500), HMGB1(Abeam, ab18256, 1 :500), or gamma-H2AX (Abeam, ab81299, 1 :250) in 1 % BSA overnight at 4 degrees Celsius. The next day, cells were washed 3x with PBS and incubated in p21 (Santa Cruz, SC-817, 1 :200) in 1 % BSA for 1 hour and 15 minutes at room temperature. Cells were washed three times with 1x PBS to remove any residual primary antibodies, then they were subsequently incubated in a cocktail of secondary antibodies and small molecule fluorophores, including Hoechst (Thermo Fisher, H3570 ,1 :250), Alexa Fluor-488 phalloidin (Thermo Fisher, A12379,1 :200), goat anti-rabbit IgG-Alexa Fluor 568 (Thermo Fisher, A-11036,1 :500), and goat antimouse IgG-Alexa Fluor 647 (Thermo Fisher, A-21240,1 :500) in 1 % BSA. Lastly, cells were washed 3x with 1X PBS and imaged using a Leica Stellaris 5 Confocal Microscope. All samples were either imaged immediately after staining or with 24 hours of secondary antibody incubation with intermediate storage at 4 degrees Celsius.

[0105] Senescence-associated beta galactosidase fluorescence staining was performed using a similar workflow as described herein. Based on manufacturer’s protocol (Thermo Fisher, C10851 ) cells were incubated with fluorescence SA [3-gal prior to the addition of the p21 primary antibody. Following the incubation with p21 primary antibody, the experimental workflow was consistent with the workflow described above.

[0106] High-content microscopy and quantification of morphological features. Fluorescent images of all conditions were acquired using a Leica Stellaris 5 Confocal Microscope at 20X magnification using 4 laser lines (405 Diode, 488 Diode, 568 Diode, and 647 Diode). Images were taken at 1024 x 1024 pixel resolution with 0.568 microns per pixel. Individual nuclei and cell Boundaries were segmented from raw TIFF files using CellProfilerTM in combination with in-house curation pipelines toensure well-segmented single cells. Briefly, an immunofluorescence-focused segmentation algorithm used the DAPI stain to delineate the nucleus boundaries and the Phalloidin stain to delineate the general cell shape. Next, an in-house curation pipeline was used to in-silico remove cells from high density regions (more than three toughing cells) to ensure the quantification of single cells. Touching cells made up a very small portion of cells based on optimized cell density and plating. From the generated masks, morphological features primarily describing the sizes, shapes, boundary roughness, and boundary curvatures were computed (Table 1 ). These masks were leveraged to also quantify the cellular expressions of protein-based biomarkers of senescence per cell. Over 250,000 single cells were profiled from snapshot images spanning all the biological conditions described herein (not including single cell tracking data).

[0107] With the masks generated as outputs from the pipelines, a feature quantification pipeline was used to automatically compute ~218 morphological parameters of each cell. These morphological parameters ranged from basic geometric features such as area, perimeter, and curvature to more complex interpretations of morphology such as skeletal features. To identify key parameters driving variance within the cell sample set, primary factor analysis was performed across all cells and all morphological parameters. Parameters expressing a commonality value below 0.2 were excluded for subsequent analysis (note a higher commonality value indicates a parameter capturing more of the variance across a given cell population). This factor analysis resulted in a reduction of relevant morphological features from 218 to 88 morphological features, which was used for all subsequent analysis.

[0108] Analysis of protein-based biomarkers. Fluorescently labeled proteinbased biomarkers were analyzed at single-cell resolution by quantifying the pixel intensity within the cells and nuclei. Specifically, HMGB1 and LMNB1 were quantified using the mean nuclear intensity (integrated intensity per object (i.e., cell or nuclei) divided by the total number of pixels (i.e., object size), p21 was quantified by total nuclear expression (i.e., summation of the pixel intensities per nuclei), and p16 and [3GAL were quantified by total cellular expression (i.e., summation of the pixel intensities per cell, included nuclear region). To correct of any effects from potential batch-to-batch variations ReCombatTM software was used for all datasets across biological replicates. Protein biomarker expressions for all cells across induction condition were log-transformed prior to z-score normalization. The control distributions for each cell were used to compute the z-score values for cells based on their individual protein expression (performed per age sample).

[0109] T o determ ine the relative cutoffs for senescence (any senescent subtype vs. non-senescence), the distributions of cells for baseline (mostly non-senescent cells) and induced cells were plotted, and the intersection point that defined the crossover for each of the distributions, i.e., cells with low values in the baseline distributions but were high in the induced conditions were denoted as a senescence threshold per biomarker, were identified. This allowed for a determination of the average intersection point of the control distribution and the distribution of H2O2 and Bleomycin senescence induction conditions for the five biomarkers (Bleomycin and H2O2 were selected as they were strong, orthogonal inducers). These values were determined to be approximately 1.6 for young (GT22) and 1.3 for old (GT125), with the older having a smaller z-score cutoff likely due to more baseline cells expressing thesenescent phenotype at baseline. Note these values correlate for approximately 10% of the healthy GT22 cells being classified as senescent and 15% for GT125 cells.

[0110] UMAP and k-means clustering. The reduced dimensionality framework was constructed using Uniform Manifold Approximation and Projection (UMAP) across all biological conditions and replicates (n=3) for GT22 (23-year-old) and GT125 (89-year-old). All morphological parameters were independently log- normalized and standard scaled (z-score). This preprocessed morphological parameter dataset was subsequently used to construct a 2D- UMAP space. UMAP is a nonlinear dimensionality reduction algorithm that captures and projects the structure of high dimensional data in a lower-dimensionality space (for the study, the 88-vector space was reduced to two with Euclidean convergence). Note that in general any dimensionality reduction technique may be used according to various embodiments, not limited to UMAP. Each point in the UMAP space represents an individual cell. Given that UMAP is a nonlinear reduction across multiple morphological parameters, there is no direct equivalent with each of the constituent dimensions, however UMAP- 1 correlated well with cell and nuclear size and UMAP-2 with cell and nuclear shapes. As a complement to the UMAP morphological analysis, k-means clustering was performed to discretize and identify distinct morphologically-defined groups within the dataset. K-means clustering analysis was performed using eleven cluster, which was determined based on the inertia and silhouette values. Here the goal was to identify the number of clusters that would minimize the inertia (typically at the elbow) and maximize the silhouette values (i.e. , stability of the respective clusters). Note that any unsupervised clustering technique may be used according to various embodiments, not limited to k-means clustering. Across the eleven clusters bona-fide senescent and non-senescent cell populations were identified, however, cluster 9 harbored a largefraction of mis-segmented cells. For initial analysis, cluster 9 was kept for cell classification but was dropped from analysis when reporting actual findings of the data due to its association with incorrect segmentation. To quantify morphological heterogeneity, the Shannon entropy was calculated using the following equation for the ten validated k-means clusters (i.e. , eleven clusters with the removal of cluster 9). log(pi )

[0111] Here, S is the Shannon entropy (greater magnitude signifies a more heterogeneous population) and p, is the fraction of cells classified within morphological cluster / (here / range from 1 -10).

[0112] Protein biomarker imputation. Given the interconnectedness of morphology and senescence-associated protein biomarker expression, a computational pipeline to determine biomarker expression based on cell and nuclear morphologies parameters alone was developed and implemented. This is particularly useful since using the disclosed imaging workflow, staining may be performed for a maximum of 2 biomarkers at a time. This biomarker imputation workflow allowed for the determination of the expression of 5 biomarkers per single cell across all the biological conditions of interest. Principal component analysis was performed on the 88 log-normalized, standard-scaled morphological parameters that was used to construct the 2D-UMAP and identify the 5 principal components capturing 95% of the sample variance. For a given cell with an unknown biomarker, the 20 nearest neighbor cells that were stained for that biomarker if interest were identified from the weighted Euclidean distance of the 5 principal components. For computational streamlining, nearest neighbors were restricted to the same cell line (GT22 or GT125) within the same k-means clusters. After nearest neighbors were identified, the expression wascomputed using a weighted average of the biomarker. The weights were determined by Euclidean distance to the imputed cell, with cells in closer proximity being weighted more. These steps were iterated for each of the biomarkers.

[0113] Kinetics of senescence experiments. Human dermal fibroblast samples were induced with either H2O2 or doxorubicin, and the senescence phenotypes were allowed to develop over 2, 5, 8, 12, or 15 days in separate dishes. At these time points, cells were seeded at low density on collagen-l coated 24-well glass-bottom plates. Cells were fixed, immunofluorescence stained for senescent associated protein biomarkers, and imaged as described herein. This data was analyzed for biomarker expression behavior and morphological cluster enrichment as a function of senescence induction time.

[0114] Machine learning approach to quantify senescence subtypes using Xception algorithm. By way of non-limiting example, Xception, a 48-layer convolutional neural network architecture for image classification, was used to identify non-parameterized trends in morphological signatures associated with senescence. This architecture was largely based on the original namesake, with two key supplements and 1 ,000,000 total trainable parameters: (1 ) weights were initialized from ImageNet and (2) additional pooling, dropout, and SoftMax layers were appended to the end of the architecture (to allow for a senescence probability as a final output of the model). To construct the input samples to the model, a 3-layer dataset including of 1536 x 1536-pixel images were created for individual cells, where the first layer was the textured nuclei image (grayscale H33342 image), the second layer was the textured actin architecture (grayscale Phalloidin image), and the third layer was the composite overlay of both the nuclei and actin (combined H33342 and Phalloidin images). Note that embodiments are not limited to Xception; other neural network orother supervised machine learning architectures capable of performing regression on images may be used according to various embodiments. Each image was geometrically centered and had pixel intensities normalized between 0-1 . This dataset was constructed from the same cells used to create the LIMAP manifold (GT22 and GT125 dermal fibroblast samples across the different induction conditions). Every layer of the input sample set was overlaid against a neutral background (pixel value of zero). Each sample of the training dataset was tagged as either senescent or nonsenescent based on protein biomarker expression z-score cutoffs (described herein). Specifically, cells differentially expressed in both biomarkers were tagged as senescent whereas cells not differentially expressed in either biomarker were tagged as healthy.

[0115] Given the surplus of senescent-tagged cells compared to controls (because of 5 senescent induction conditions as opposed to one control), the training set was subsampled from the larger dataset at large to balance both senescent and healthy cells. Furthermore, the training set was balanced for age and the senescent- tagged cells were further balanced for induction conditions to minimize potential bias. Non-senescent tagged cells were biased towards control samples (>80% of samples were control conditions in either age group). A generative model was applied to augment the dataset by implementing non-distortive transformations (i.e., flips and rotations) to maintain the integrity of size and shape effects of the training set. Training of the model was conducted in two parts, with an 80-20 train-test split. The first 200 epochs were trained with ImageNet initialized weights, frozen backbone, Adam’s Learning rate of 1 x1 O’4, and accuracy-based optimization. Next, the backbone was unfrozen and trained for an additional 20 epochs with an Adam’s Learning Rate of 1 xl O’3, which allowed for a finer convergence of algorithm accuracy. The accuracy ofmodel along with the area under the curve (AUC) of the receiver operating curve (ROC) were used to evaluate model performance. The following evaluations were used. AccuracyTN False Positive Rate = 1 -TN + FP TP True Positive Rate = -TP + FN ROC — Graph of False Positive Rate vs True Positive Rate

[0116] In the above, TP is true positive, TN is true negative, FP is false positive, and FN is false negative.

[0117] Evaluating the effects of aging on senescence experiments. Fortyeight additional primary dermal fibroblast samples of known gender and chronological age were cultured and fixed at low densities using experimental workflows described herein. Morphologies of each patient at baseline were analyzed and their senescent score and their enrichment within morphologically-defined senescence subtypes was determined. Using a univariate linear regression model, how the abundance of senescence cells and the senescence score correlated with donor age as analyzed. 18 samples were further selected to assess chronological age-associated response to senescence induction. Samples were either treated to DMSO or Doxorubicin (8-day induction period) and subsequently fixed for morphological analysis. Morphologies of cells in both conditions were analyzed by the trained Xception model for score distributions and analyzed by the k-means algorithm to determine senescent subtype distribution. Using this two-pronged orthogonal approach, how the abundance ofsenescence within each of the three bona-fide senescence subtypes changes with increasing donor age was quantified.

[0118] Senotherapy response and live-cell imaging. GT22, human dermal fibroblast samples were induced using a 8-day doxorubicin induction protocol and plated at a density of 300 cells / well in a 96 well Collagen-I coated glass-bottom plate (Cellvis, P96-1.5H-N). After allowing the cells to adhere overnight, wells were replenished with fresh media containing Spirochrome 555-DNA (Spirochrome, SC201 ,1 :1000) and Spirochrome 650-FastAct (Spirochrome, SC505, 1 :1000) and incubated for four hours at physiological conditions in a humidified cell culture incubator. A baseline series of image was taken of the cells prior to treatment with serotherapeutic drugs. Following the acquisition of the baseline images, wells were supplemented with either Dasatinib (SelleckChem, S1021 , 1 pM) and Quercetin (SelleckChem, S2391 , 10 pM), Metformin (SelleckChem, S1950, 5 pM), Navitoclax (SelleckChem, S1001 , 5 pM), Fisetin (SelleckChem, S2298, 25 pM), or ARV-825 (SelleckChem, S8297, 10 nM) as a one time, bolus treatment while plates were on the microscope. Images were taken once every hour for a span of 60 hours to capture the response dynamics per drug. Similar to the fixed images, cells in each frame were segmented using an optimized CellProfilerTM workflow, and the trackpy Python package was used to ascribe cell identities across frames of the same imaged well. Segmented images were analyzed using the same LIMAP and k-means analysis described herein to determine the morphological clusters of individual cells as a function of time.

[0119] Dasatinib + quercetin single-cell responses and viability. Dasatinib+Quercetin treatment was used as the focus for single-cell viability as it was the most effective senotherapy based on cell killing from the inventors’ smallsenotherapy screen. To determined dead or dying cells based on the time-lapse movies, two criteria were assessed based on its morphological progression: (1 ) the cell has lost tracking and (2) in the frame before the cell has lost tracking, it must be in either cluster 1 , 2, or 3 (indicative of a senescent dying phenotype). Although clusters 1 , 2, and 3 were identified as non-senescent clusters at baseline, cells belonging to these clusters after senescence induction and senotherapy treatment exhibited enhanced expression of DNA dama as denoted by gamma H2AX. Cells were classified by their initial morphological subtype using the pre-treatment images and we analyzed their as a function of senotherapy time to assess subtype-specific responses. Given an uneven number of cells in each senescent morphological cluster, the inventors normalized cell counts for each subtype. To identify how cell morphology transitions as function of treatment, a hierarchal clustering algorithm was fed a dataset of all senescent morphological subtype trajectories (i.e. , a cell with a baseline starting in one of the senescent morphological subtypes (C7, C10, C11 ) for both untreated and D+Q treatment conditions). The hierarchical clustering used the ward algorithm to discretize four response patterns: short death, long death, stable viable, and unstable viable.

[0120] Statistics and reproducibility. To establish statistical comparisons between multiple groups we performed One-way ANOVA and post-hoc pairwise Tukey tests. Correlations of linear regression fits were evaluated using R and p-values, respectively. Data was log-transformed, where applicable, to normalize distributions for gaussian-based models. No data was excluded from the reports herein. All experiments were performed with at least 3 biological replicates with at least 2 in-plate technical replicates.

[0121] Software. All single cell segmentations were performed using optimized workflows in CellProfiler. All dynamic tracking data was performed using an custom algorithm combining CellProfiler outputs with TrackPy algorithum. All analyses was performed in Python with the following software specifications: python 3.8.12, tensorflow-gpu 2.5.0, scipy 1.7.1 , scikit-learn 1.1.1 , pandas 1.4.3, matplotlib 3.4.2, seaborn 0.12.0, plotly 5.10.0, trackpy 0.5.0, umap-learn 0.5.3, and numpy 1.21.2.

[0122] Fig. 14 is a flow chart for a machine learning method 1400 of characterizing senescence in an immobilized patient cell, according to various embodiments. The method may be implemented using computer hardware as described herein.

[0123] At 1402, the method 1400 includes obtaining a patient cell image. According to various embodiments, the patient cell image may include one or more of: an image of the patient cell, an image of a nucleus of the patient cell, and / or a composite image of the patient cell and the nucleus of the patient cell. According to some embodiments, the patient cell image may include immunofluorescence staining.

[0124] At 1404, the method 1400 includes passing the patient cell image to a trained machine learning system. The trained machine learning system may include a supervised machine learning system (e.g., a neural network or a random forest) and / or an unsupervised machine learning system (e.g., a k-means or other clustering technique), as described in detail herein. The trained machine learning system may trained using a training corpus that includes a plurality of training cell images, e.g., as shown and described herein. The training cell images may include a plurality of images of normal cells and a plurality of images of cells with senescence. Each training cell image may include one or more of: an image of a training cell, an image of a nucleus of a training cell, and / or a composite image of a training cell and thenucleus of the training cell. According to some embodiments, each training cell image may include immunofluorescence staining.

[0125] According to some embodiments, the trained machine learning system includes a neural network or a random forest. The trained machine learning system may be trained to output a senescence score. According to such embodiments, each training cell image may be labeled with a respective indication representative of senescence or not senescence.

[0126] According to some embodiments, the trained machine learning system may include a k-means or other clustering technique. The training of such a trained machine learning system may include: determining respective values of each of a set of morphological parameters for each training cell image, reducing a dimensionality of the set of morphological parameters (e.g., using LIMAP or another dimensionality reduction technique), and clustering the training cell images in the reduced dimensionality, from which a plurality of clusters are obtained.

[0127] At 1406, the method 1400 includes obtaining an output from the trained machine learning model in response to the passing the patient cell image to the trained machine learning system. The output may be indicative of a senescence property of the patient cell. The senescence property may be any of a variety of non-binary senescence properties, such as, by way of non-limiting examples: a senescence score (e.g., a numerical score in a range such as, by way of non-limiting example, zero to one), a senescence subtype classification, or a numerical or categorical characterization of susceptibility to senescence.

[0128] According to embodiments in which the trained machine learning system includes unsupervised clustering, the output may be based on a closest cluster to the patient cell in the reduced dimensionality. According to such embodiments, thesenescence classification of the patient cell may correspond to an identification of the closest cluster.

[0129] At 1408, the method 1400 includes providing an indication of the senescence property of the patient cell based on the output. The indication may be provided by display on a monitor, via a computer network, and / or to a medical records system for storage in association with an identification of the patient.

[0130] At 1410, the method 1400 includes clinically treating the patient based on the senescence property indication. For example, the patient may be clinically treated with a therapeutic based on the indication of the senescence property of the patient cell by adjusting a dosage and / or a duration of the therapeutic. According to some embodiments, the patient may be clinically treated by selecting a therapeutic based on the indication of the senescence property of the patient cell, where the therapeutic includes one or more of: a chemotherapy drug, CAR-T therapy, or a senotherapy drug. Any of the specific chemotherapy drugs, CAR-T therapies, or senotherapy drugs disclosed herein may be so selected.

[0131] According to various embodiments, the method 1400 may further include providing an indication of a protein expression of the patient cell based on the output.

[0132] Certain examples can be performed using a computer program or set of programs. The computer programs can exist in a variety of forms both active and inactive. For example, the computer programs can exist as software program (s) comprised of program instructions in source code, object code, executable code or other formats; firmware program(s), or hardware description language (HDL) files. Any of the above can be embodied on a transitory or non-transitory computer readable medium, which include storage devices and signals, in compressed or uncompressed form. Exemplary computer readable storage devices include conventional computersystem RAM (random access memory), ROM (read-only memory), EPROM (erasable, programmable ROM), EEPROM (electrically erasable, programmable ROM), flash memory, and magnetic or optical disks or tapes.

[0133] Aspects of the present disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented using computer readable program instructions that are executed by an electronic processor.

[0134] These computer readable program instructions may be provided to a processor of a general-purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the electronic processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and / or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises an article of manufacture including instructions which implement aspects of the function / act specified in the flowchart and / or block diagram block or blocks.

[0135] In embodiments, the computer readable program instructions may be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, configuration data for integrated circuitry, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++, or the like, and procedural programming languages, such as the C programming language or similar programming languages. The computer readable program instructions may execute entirely on a user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server.

[0136] As used herein, the terms “A or B” and “A and / or B” are intended to encompass A, B, or {A and B}. Further, the terms “A, B, or C” and “A, B, and / or C” are intended to encompass single items, pairs of items, or all items, that is, all of: A, B, C, {A and B}, {A and C}, {B and C}, and {A and B and C}. The term “or” as used herein means “and / or.”

[0137] As used herein, language such as “at least one of X, Y, and Z,” “at least one of X, Y, or Z,” “at least one or more of X, Y, and Z,” “at least one or more of X, Y, or Z,” “at least one or more of X, Y, and / or Z,” or “at least one of X, Y, and / or Z,” is intended to be inclusive of both a single item (e.g., just X, or just Y, or just Z) and multiple items (e.g., {X and Y}, {X and Z}, {Y and Z}, or {X, Y, and Z}). The phrase “at least one of” and similar phrases are not intended to convey a requirement that each possible item must be present, although each possible item may be present.

[0138] The techniques presented herein are referenced and applied to material objects and concrete examples of a practical nature that demonstrably improve the present technical field and, as such, are not abstract, intangible or purely theoretical. Further, if any claims appended to the end of this specification contain one or more elements designated as “means for [perform]ing [a function]...” or “stepfor [performing [a function]...”, it is intended that such elements are to be interpreted under 35 U.S.C. § 112(f). However, for any claims containing elements designated in any other manner, it is intended that such elements are not to be interpreted under 35 U.S.C. § 112(f).

[0139] While the invention has been described with reference to the exemplary examples thereof, those skilled in the art will be able to make various modifications to the described examples without departing from the true spirit and scope. The terms and descriptions used herein are set forth by way of illustration only and are not meant as limitations. In particular, although the method has been described by examples, the steps of the method can be performed in a different order than illustrated or simultaneously. Those skilled in the art will recognize that these and other variations are possible within the spirit and scope, e.g., as defined in the following claims and their equivalents.

Claims

What is claimed is:1 . A machine learning method of characterizing senescence in an immobilized patient cell, the method comprising: obtaining a patient cell image; passing the patient cell image to a trained machine learning system, wherein the trained machine learning system is trained using a training corpus comprising a plurality of training cell images, wherein the plurality of training cell images comprise a plurality of images of normal cells and a plurality of images of cells with senescence; obtaining an output from the trained machine learning model in response to the passing, wherein the output is indicative of a senescence property of the patient cell, and wherein the senescence property is non-binary; and providing an indication of the senescence property of the patient cell based on the output.

2. The method of claim 1 , wherein the trained machine learning system comprises a neural network, wherein the training cell images are labeled with respective indications representative of senescence or not senescence, and wherein the senescence property comprises a senescence score for the patient cell.

3. The method of claim 2, further comprising clinically treating the patient with a therapeutic based on the indication of the senescence property of the patient cell, wherein at least one of a dosage of the therapeutic or a duration of the therapeutic is based on the indication of the senescence property of the patient cell.

4. The method of claim 1 , wherein the patient cell image comprises at least one of: an image of the patient cell, an image of a nucleus of the patient cell, or a composite image of the patient cell and the nucleus of the patient cell, and wherein a respective training cell image comprises at least one of: an image of a respective training cell, an image of a nucleus of a respective training cell, or a composite image of a respective training cell and the nucleus of the respective training cell.

5. The method of claim 1 , wherein the senescence property comprises a senescence subtype classification of the patient cell.

6. The method of claim 5, further comprising: selecting a therapeutic based on the indication of the senescence property of the patient cell, wherein the therapeutic comprises at least one of: a chemotherapy drug, CAR-T therapy, or a senotherapy drug; and clinically treating the patient with the treatment.

7. The method of claim 5, wherein the machine learning system is trained using the plurality of training cell images by actions comprising: determining respective values of each of a set of morphological parameters for each of the plurality of training cell images,reducing a dimensionality of the set of morphological parameters based on results of the determining, whereby a reduced dimensionality is obtained, and clustering the plurality of training cell images in the reduced dimensionality, whereby a plurality of clusters are obtained, wherein the output is based on a closest cluster of the plurality of clusters to the patient cell, and wherein the senescence classification of the patient cell corresponds to the closest cluster of the plurality of clusters to the patient cell.

8. The method of claim 7, wherein the patient cell image comprises immunofluorescence staining, and wherein each training cell image comprises immunofluorescence staining.

9. The method of claim 1 , wherein the senescence property comprises a characterization of a susceptibility to senescence of the patient cell.

10. The method of claim 1 , further comprising providing an indication of a protein expression of the patient cell based on the senescence property.11 . A machine learning system for characterizing senescence in an immobilized patient cell, the system comprising: a non-transitory computer readable medium comprising instructions; and at least one electronic processor that executes the instructions to perform operations comprising: obtaining a patient cell image;passing the patient cell image to a trained machine learning system, wherein the trained machine learning system is trained using a training corpus comprising a plurality of training cell images, wherein the plurality of training cell images comprise a plurality of images of normal cells and a plurality of images of cells with senescence; obtaining an output from the trained machine learning model in response to the passing, wherein the output is indicative of a senescence property of the patient cell, and wherein the senescence property is non-binary; and providing an indication of the senescence property of the patient cell based on the output.

12. The system of claim 11 , wherein the trained machine learning system comprises a neural network, wherein the training cell images are labeled with respective indications representative of senescence or not senescence, and wherein the senescence property comprises a senescence score for the patient cell.

13. The system of claim 12, wherein the patient is clinically treated with a therapeutic based on the indication of the senescence property of the patient cell, wherein at least one of a dosage of the therapeutic or a duration of the therapeutic is based on the indication of the senescence property of the patient cell.

14. The system of claim 11 , wherein the patient cell image comprises at least one of: an image of the patient cell, an image of a nucleus of the patient cell, or a composite image of the patient cell and the nucleus of the patient cell, andwherein a respective training cell image comprises at least one of: an image of a respective training cell, an image of a nucleus of a respective training cell, or a composite image of a respective training cell and the nucleus of the respective training cell.

15. The system of claim 11 , wherein the senescence property comprises a senescence subtype classification of the patient cell.

16. The system of claim 15, wherein the operations further comprise: selecting a therapeutic based on the indication of the senescence property of the patient cell, wherein the therapeutic comprises at least one of: a chemotherapy drug, CAR-T therapy, or a senotherapy drug, wherein the patient is clinically treated with the treatment.

17. The system of claim 15, wherein the machine learning system is trained using the plurality of training cell images by actions comprising: determining respective values of each of a set of morphological parameters for each of the plurality of training cell images, reducing a dimensionality of the set of morphological parameters based on results of the determining, whereby a reduced dimensionality is obtained, and clustering the plurality of training cell images in the reduced dimensionality, whereby a plurality of clusters are obtained,wherein the output is based on a closest cluster of the plurality of clusters to the patient cell, and wherein the senescence classification of the patient cell corresponds to the closest cluster of the plurality of clusters to the patient cell.

18. The system of claim 17, wherein the patient cell image comprises immunofluorescence staining, and wherein each training cell image comprises immunofluorescence staining.

19. The system of claim 11 , wherein the senescence property comprises a characterization of a susceptibility to senescence of the patient cell.

20. The system of claim 11 , wherein the operations further comprise providing an indication of a protein expression of the patient cell based on the senescence property.

Images