Prediction of tumor response and clinical benefit from KRAS therapeutic agents

Abstract

A computer-implemented method of determining a tumor response to a KRAS therapeutic agent is described, the method comprising: providing, at a computing device including one or more processors for data processing, RNA sequencing data associated with a tumor, the RNA sequencing data being indicative of gene expression data of a plurality of genes involved in one or more biochemical signaling pathways at least partly related to KRAS; evaluating the RNA sequencing data with a first computational model modelling at least one of dependency of the tumor on the KRAS signaling pathway, and sensitivity of the tumor to KRAS inhibitor therapy; evaluating the RNA sequencing data with one or more additional computational models modelling activation of one or more biological signaling pathways at least partly related to KRAS; and classifying, based on the evaluation of the RNA sequencing data with the first computational model and the at least one additional computational model, the tumor as being responsive or non-responsive to, and / or the patient as deriving clinical benefit from, the KRAS therapeutic agent.

Description

PREDICTION OF TUMOR RESPONSE AND CLINICAL BENEFIT FROMKRAS THERAPEUTIC AGENTSCROSS REFERENCE TO RELATED APPLICATION(S)

[0001] This application claims the benefit of and priority to U.S. Provisional Application No. 63 / 571,813, filed on March 29, 2024, European Patent Application No. 24183371.4, filed June 20, 2024, and U.S. Provisional Application No. 63 / 697,142, filed on September 20, 2024, each of which is hereby incorporated by reference in its entirety.TECHNICAL FIELD

[0002] Generally, the present disclosure relates to a computer-implemented method of determining a tumor response to a therapeutic treatment. Specifically, the present disclosure relates to a computer-implemented method of determining a tumor response to a KRAS therapeutic agent. Further, the present disclosure relates to a computing device or system configured to perform steps of said method, to a corresponding computer program, and to a computer-readable medium storing such computer program.BACKGROUND OF THE DISCLOSURE

[0003] KRAS (Kirsten Rat Sarcoma viral oncogene homolog) is a gene that encodes a protein or monomeric G-protein (also referred to as KRAS protein), involved in cell signaling pathways regulating cell division and growth. It belongs to the RAS family of oncogenes, which are known for their involvement in cancer development.

[0004] In normal, non-cancerous cells or non-tumor cells, KRAS regulates or is involved in regulating cell growth, differentiation, and survival by transmitting signals from cell surface receptors to the cell nucleus. Therein, the protein produced based on the KRAS gene acts as a binding partner for GTP (Guanosine Triphosphate) and acts as molecular switch, turning on and off various cellular processes in response to external signals. To perform this function, the KRAS protein changes its conformation or binding activity regarding GTP. In the activated state, KRAShas the GTP molecule firmly bound. Only in this form does the protein have the possibility to interact with other signaling proteins and activate a signaling pathway or signaling transduction chain. In the inactive state, the binding partner of KRAS, the GTP, has been hydrolyzed to GDP by an enzyme from the group of GTPases.

[0005] Mutations in the KRAS gene can result in complete loss of GTPase activity, and no GTP can be hydrolyzed to GDP. Therefore, mutations in the KRAS gene, inter alia, can cause it to become permanently activated. The mutated KRAS protein drives cancer progression by promoting cell proliferation, inhibiting apoptosis (programmed cell death), and enhancing the ability of cancer cells to invade surrounding tissues and metastasize to distant organs. Moreover, KRAS mutations can also render cancer cells resistant to certain targeted therapies, making them more difficult to treat. Therein, KRAS mutations are commonly found in several types of cancer or tumors, including pancreatic cancer, colorectal cancer, and lung cancer.

[0006] Due to its central role in cancer development and progression, KRAS has been a major focus of cancer research, with efforts aimed at developing targeted therapies to inhibit mutant KRAS activity and improve treatment outcomes for patients with KRAS-driven cancers. However, recent KRAS-targeted therapies with KRAS therapeutic agents only show modest effectiveness, with patient or tumor response rates of about 29-53% in colon and lung cancer clinical trials.

[0007] Also, currently used KRAS-related biomarkers are based on DNA sequencing or polymerase chain reaction (PCR) assays to determine KRAS mutation status, either alone or as part of a genetic panel, wherein the KRAS mutation status refers to whether or not a particular cancer cell harbors mutations in the KRAS gene. However, these DNA-based biomarkers are insufficient to predict which patients or tumors will respond to targeted KRAS therapies.SUMMARY

[0008] Here, a computer-implemented method of determining a tumor response to a therapeutic treatment is disclosed.

[0009] In some embodiments, a computer-implemented method of determining a tumor response to a KRAS therapeutic agent and / or of determining an efficacy of a KRAS therapeutic agent against a tumor comprises providing, at a computing device including one or more processors for data processing, RNA sequencing data associated with a tumor, the RNAsequencing data being indicative of gene expression data of a plurality of genes involved in one or more biological signaling pathways at least partly related to KRAS; evaluating the RNA sequencing data with a first computational model modelling at least one of dependency of the tumor on the KRAS signaling pathway, and sensitivity of the tumor to KRAS inhibitor therapy; evaluating the RNA sequencing data with one or more additional computational models modelling activation of one or more biological signaling pathways at least partly related to KRAS; and classifying, based on the evaluation of the RNA sequencing data with the first computational model and the one or more additional computational models, the tumor as being responsive or non-responsive to the KRAS therapeutic agent.

[0010] In some embodiments, the first computational model is configured to model one or both of the dependency of the tumor on the KRAS signaling pathway and the sensitivity of the tumor to KRAS inhibitor therapy as continuous tumor biology parameter and / or with a continuous metric; and / or wherein the one or more additional computational models are configured to model the activation of said one or more biological signaling pathways at least partly related to KRAS as continuous tumor biology parameter and / or with a continuous metric.

[0011] In some embodiments, the first computational model and the one or more additional computational models each define a model axis of a computational multi-axis model modelling at least two different tumor biology parameters selected from the group consisting of dependency of the tumor on the KRAS signaling pathway, sensitivity of the tumor to KRAS inhibitor therapy, and activation of the one or more biochemical signaling pathways at least partly related to KRAS.

[0012] In some embodiments, the multi-axis model includes two or more model axes, the two or more model axes comprising at least a first model axis representative of at least one of the dependency of the tumor on the KRAS signaling pathway and the sensitivity of the tumor to KRAS inhibitor therapy, and a second model axis representative of the activation of the one or more biological signaling pathways at least partly related to KRAS.

[0013] In some embodiments, the multi-axis model includes at least one further model axes representative of one or more of: one or more cell or tumor intrinsic signaling pathways, one or more cell or tumor extrinsic signaling pathways, cellular composition of the tumor, cellular states of one or more tumor cells, biological activities of one or more tumor cells, molecular processes,genetic variants, synthetic lethal partners, metabolic signatures, modifications, and proteininteraction networks.

[0014] In some embodiments, the multi-axis model defines a multi-dimensional phenotypic vector space for the phenotype of the tumor; and the multi-dimensional phenotypic vector space includes at least a first region associated with a first phenotype of the tumor responsive to the KRAS therapeutic agent and a second region associated with a second phenotype of the tumor non-responsive to KRAS therapeutic agent.

[0015] In some embodiments, the method further comprises determining, based on the evaluation of the RNA sequencing data with the first computational model, one or both of a dependency score indicative of a degree of dependency of the tumor on the KRAS signaling pathway and a sensitivity score indicative of a degree of responsiveness of the tumor to KRAS inhibitor therapy; determining, based on the evaluation of the RNA sequencing data with the one or more additional computational models, at least one activation state score indicative of a degree of activation of the one or more biological signaling pathways at least partly related to KRAS; and mapping the determined at least one activation state score and at least one of the dependency score and the sensitivity score into the multi-dimensional phenotypic vector space defined by the multi-axis model.

[0016] In some embodiments, the method further comprises determining that the at least one activation state score and at least one of the dependency score and the sensitivity score are located within the first region of the multi-dimensional phenotypic vector space associated with the first phenotype of the tumor responsive to the KRAS therapeutic agent, thereby classifying the tumor as being responsive to the KRAS therapeutic agent; and / or determining that the at least one activation state score and at least one of the dependency score and the sensitivity score are located within the second region of the multi-dimensional phenotypic vector space associated with the second phenotype of the tumor non-responsive to the KRAS therapeutic agent, thereby classifying the tumor as being non-responsive to the KRAS therapeutic agent.

[0017] In some embodiments, evaluating the RNA sequencing data with the first computational model includes determining one or both of a dependency score indicative of a degree of dependency of the tumor on the KRAS signaling pathway and a sensitivity score indicative of a degree of responsiveness of the tumor to KRAS inhibitor therapy; and / or evaluating the RNA sequencing data with the one or more additional computational modelsincludes determining one or more activation state scores indicative of a degree of activation of the one or more biological signaling pathways at least partly related to KRAS.

[0018] In some embodiments, the method further comprises providing the at least one activation state score and at least one of the dependency score and the sensitivity score as inputs to a classifier for classifying the tumor as being responsive or non-responsive to the KRAS therapeutic agent.

[0019] In some embodiments, the classifier is based on logistic regression.

[0020] In some embodiments, the first computational model and the one or more additional computational models are independent computational models.

[0021] In some embodiments, the first computational model is a trained machine learning model; and / or at least one of the one or more additional computational models is a statistical model.

[0022] In some embodiments, the first computational model is a trained machine learning model configured to model dependency of the tumor on the KRAS signaling pathway and / or sensitivity of the tumor to KRAS inhibitor therapy based on model features associated with a plurality genes involved in one or more of the RAS signaling, the MAPK signaling, the PI3K signaling, and the EGFR signaling pathways.

[0023] In some embodiments, the first computational model is trained based on LI regularization and L2 regularization.

[0024] In some embodiments, at least one of the one or more additional computational models is a statistical model configured to model activation of one or more biological signaling pathways at least partly related to KRAS based on quantifying gene expression of a plurality of genes involved in one or more biological signaling pathways at least partly related to KRAS.

[0025] In some embodiments, the one or more biological signaling pathways at least partly related to KRAS include Androgen Receptor signaling pathway, Estrogen Receptor signaling pathway, Hypoxia signaling pathway, JAK-STAT signaling pathway, MAPK signaling pathway, NFkB signaling pathway, Notch signaling pathway, PI3K / AKT signaling pathway, p53 signaling pathway, RTK signaling pathway, TGFb signaling pathway, TNFa signaling pathway, Trail signaling pathway, and WNT / p-Catenin signaling pathway.

[0026] In some embodiments, the method further comprising evaluating the RNA sequencing data with at least one further computational model configured to model at least one further tumorparameter, and the at least one further tumor parameter is indicative of one or more of a microenvironment of the tumor, RNA sequencing metadata, demographic data of the patient, angiogenesis of the tumor, activity of the VEGF signaling pathway, and cellular composition of the tumor, preferably wherein the at least one further computational model is a trained machine learning model.

[0027] In some embodiments, providing the RNA sequencing data at the computing device includes one or more of accessing the RNA sequencing data with the computing device, obtaining the RNA sequencing data at the computing device, receiving the RNA sequencing data at the computing device, and retrieving the RNA sequencing data with the computing device.

[0028] In some embodiments, providing the RNA sequencing data at the computing device includes measuring the RNA sequencing data based on one or more RNA sequencing modalities including total-RNA sequencing, polyA enriched sequencing, and RNA-Exome-based sequencing.

[0029] In some embodiments, the RNA sequencing data include data associated with human tumor tissue, with human-derived tumor tissue, or preclinical tumor tissue.

[0030] In some embodiments, the RNA sequencing data include data from one or more sources including fresh tissue, biofluid, Formalin-Fixed Paraffin-Embedded, cell culture, and a cell-free source.

[0031] In some embodiments, the method further comprises normalizing the RNA sequencing data based on one or more of a counts per million normalization, a fragments per kilobase million normalization, a transcripts per million normalization, an upper quartile normalization, a normalization with respect to counts adjusted with upper quartile factors, a normalization with respect to trimmed mean of M- values, and a normalization with respect to counts adjusted with trimmed mean of M- values factors.

[0032] In some embodiments, the RNA sequencing data is acquired prior to administration of the KRAS therapeutic agent, during treatment with the KRAS therapeutic agent or after administration of the KRAS therapeutic agent.

[0033] In some embodiments, the RNA sequencing data is acquired prior to administration of the KRAS therapeutic agent, and wherein the method further comprises: obtaining further RNA sequencing data acquired after administration of the KRAS therapeutic agent; and determining, based on evaluating the further RNA sequencing data with the first computational model and theone or more additional computational models, an evolution and / or alteration of the tumor response to the KRAS therapeutic agent.

[0034] In some embodiments, the KRAS therapeutic agent is a KRAS inhibitor.

[0035] In some embodiments, the KRAS therapeutic agent is a mutation specific inhibitor or a pan-KRAS inhibitor.

[0036] In some embodiments, the KRAS therapeutic agent is a G12C inhibitor or a KRAS G12D inhibitor.

[0037] In some embodiments, the tumor is related to lung cancer, non-small cell lung cancer, pancreatic cancer, or a colorectal cancer.

[0038] In some embodiments, the first computational model is based on one or more genes selected from the group consisting of TP53, SIKE1, DUSP4, MLPH, MAP3K5, DUSP6, ETV1, ANTXR2, RASSF1, KRAS, KLHL22, RAP1B, and RAB5A.

[0039] In some embodiments, a computer program, which when executed on a computing device, instructs the computing device to perform steps of the method of any one of the preceding embodiments.

[0040] In some embodiments, a non-transitory computer-readable medium stores a computer program according to the preceding embodiment

[0041] In some embodiments, a computing device with one or more processors, the computing device being configured to perform steps of the method according to any one of the previous embodiments.

[0042] In some embodiments, RNA sequencing data are used for determining i) a tumor response to a KRAS therapeutic agent, ii) an efficacy of a KRAS therapeutic agent against a tumor and / or iii) a time on a KRAS therapeutic agent treatment, using the method according to any one of the previous embodiments.BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings, which are incorporated into and constitute a part of this specification, illustrate various exemplary embodiments and together with the description, serve to explain the principles of the disclosed embodiments.

[0044] FIG. 1 shows a computing device configured to determine a tumor response to a KRAS therapeutic agent according to an exemplary embodiment.

[0045] FIG. 2 shows a flow chart illustrating steps of a computer-implemented method of determining a tumor response to a KRAS therapeutic agent according to an exemplary embodiment.

[0046] FIG. 3 shows a flow chart illustrating aspects of model development for a computing device and method according to an exemplary embodiment.

[0047] FIG. 4 shows a heatmap of selected genes for a dependency model and / or a first computational model according to an exemplary embodiment.

[0048] FIG. 5 illustrates a plurality of computational models to determine a tumor response to a KRAS therapeutic agent according to an exemplary embodiment.

[0049] FIGs. 6A-6C illustrate steps of a computer-implemented method of determining a tumor response to a KRAS therapeutic agent according to exemplary embodiments.

[0050] FIGs. 7A-7F illustrate steps of a computer-implemented method of determining a tumor response to a KRAS therapeutic agent according to exemplary embodiments.

[0051] FIGs. 8A-8C illustrate Kaplan-Meier survival curves for clinical data and data predicted using a computer-implemented method of determining a tumor response to a KRAS therapeutic agent according to an exemplary embodiment.

[0052] FIG. 9 schematically shows an output of a multi-axes model of an embodiment of the method of the present disclosure.

[0053] FIG. 10 schematically illustrates an output of a multi-axes model to stratify a real- world cohort of sotorasib-treated NSCLC patients according to an exemplary embodiment.

[0054] FIG. 11 shows a Kaplan-Meier plot illustrating treatment durability (or time on sotorasib), grouped by response cohort and according to an exemplary embodiment

[0055] FIG. 12 shows the output of a multi-axes model illustrating clustering by biologies and hypothesized relationships to KRAS inhibitor therapy, ICI therapy or combination therapy in a CPTAC-3 (Clinical Proteomic Tumor Analysis Consortium, CPTAC) PDAC patient cohort according to an exemplary embodiment.

[0056] FIG. 13 shows the output of a multi-axes model for a plurality of computational models according to an exemplary embodiment.

[0057] FIG. 14 illustrates a Kaplan-Meier plot for predicted and observed treatment durability (or time on sotorasib) at 180 days based on 72 non-small cell lung cancer (NSCLC) patients, grouped by response cohort and according to an exemplary embodiment.

[0058] The Figures are schematic only and not true to scale. In principle, identical or like parts, elements and / or steps are provided with identical or like reference numerals in the Figures.DETAILED DESCRIPTION

[0059] Reference will now be made in detail to the exemplary embodiments of the present disclosure, examples of which are illustrated in the accompanying drawings. Wherever possible, the same reference numbers will be used through-out the drawings to refer to the same or like parts.

[0060] The systems, devices, and methods disclosed herein are described in detail by way of examples and with reference to the figures. The examples discussed herein are examples only and are provided to assist in the explanation of the apparatuses, devices, systems, and methods described herein. None of the features or components shown in the drawings or discussed below should be taken as mandatory for any specific implementation of any of these devices, systems, or methods unless specifically designated as man-datory.

[0061] Also, for any methods described, regardless of whether the method is described in conjunction with a flow diagram, it should be understood that unless otherwise specified or required by context, any explicit or implicit ordering of steps performed in the execution of a method does not imply that those steps must be performed in the order presented but instead may be performed in a different order or in parallel.

[0062] As used herein, the term "exemplary" is used in the sense of "example," rather than "ideal." Moreover, the terms "a" and "an" herein do not denote a limitation of quantity, but rather denote the presence of one or more of the referenced items.

[0063] Due to the high variability in response to KRAS-targeted therapies and due to the drawbacks of DNA-based approaches and other biomarkers, an improved method, approach and / or biomarker to determine a tumor response to a tumor-targeted therapy in general, and specifically tumor response to a KRAS therapeutic agent is needed. In particular, it may be desirable to provide for an improved method, approach and / or biomarker to identify patients likely to benefit from tumor-targeted therapy, such as KRAS therapeutic agents.

[0064] Aspects of the present disclosure relate to one or more computer-implemented methods, in particular for determining a tumor response to a KRAS therapeutic agent, to a corresponding computing device or system, to a corresponding computer program, and to a computer-readablemedium storing such computer program. It is emphasized that any disclosure presented herein with reference to one or an aspect of the present disclosure equally applies to any other aspect of the present disclosure.

[0065] According to a first aspect of the present disclosure, there is provided a computer- implemented method of determining a tumor response and / or patient response to a KRAS therapeutic agent. Alternatively or additionally, the method according to the first aspect may relate to a computer-implemented method of determining an efficacy of a KRAS therapeutic agent against a tumor. Alternatively or additionally, the method according to the first aspect may relate to a method of determining whether or not a tumor, one or more tumor cells, a patient and / or an individual responds to a KRAS therapeutic agent. In some embodiments, the method comprises: providing, at a computing device including one or more processors for data processing, RNA sequencing data associated with a tumor, a tumor tissue, a tumor sample and / or one or more tumor cells, the RNA sequencing data being indicative of gene expression data and / or variant data of a plurality of genes involved in one or more biological signaling pathways at least partly related to KRAS; evaluating, processing and / or analyzing the RNA sequencing data with a first computational model modelling and / or being configured to model at least one of dependency of the tumor on the KRAS signaling pathway and sensitivity of the tumor to KRAS inhibitor therapy; evaluating, processing and / or analyzing the RNA sequencing data with one or more additional computational models modelling and / or being configured to model activation (also referred to herein as activation state) of one or more biological signaling pathways at least partly related to KRAS; and classifying, based on the evaluation of the RNA sequencing data with the first computational model and the one or more additional computational models, the tumor as being responsive or non-responsive to the KRAS therapeutic agent.

[0066] Evaluating RNA sequencing data with the first computational model and one or more additional computational models, and modelling dependency on the KRAS signaling pathway and / or sensitivity to KRAS inhibitor therapy as well as modelling activation of one or more biological pathways at least partly related to KRAS can allow assessment of KRAS mutation status as well as predicting, determining or assessing tumor response and / or patient response to the KRAS therapeutic agent. The method according to the first aspect of the present disclosure can provide for an RNA-based computational biomarker, thereby providing an improved tool, system and / or approach to assess KRAS mutation status and assess tumor or patient response. Inturn, the method can assist in guiding clinical development and therapeutic decisions. Also, the method can advantageously allow to stratify responses including assessment of clinical benefit of patients to a KRAS therapeutic agent.

[0067] Known KRAS-related biomarkers are based on DNA sequencing or PCR assays to determine KRAS mutation status alone or as part of a gene panel. This approach is insufficient to accurately or robustly predict which patients or tumors will respond to targeted KRAS therapies, or respectively, to a KRAS therapeutic agent.

[0068] Utilizing RNA sequencing data in accordance with embodiments of the present disclosure is advantageous over DNA sequencing data. DNA tests or assays only determine the presence or absence of mutations, while RNA tests can assess gene variants and alterations, and can quantify gene expression and / or variants. Accordingly, RNA tests or RNA sequencing data can provide a more comprehensive assessment of mutation status and functional biological states of tumor cells. DNA sequencing data alone cannot capture the biological complexity needed to stratify treatment outcomes effectively, whereas RNA sequencing data can allow evaluation of the overall transcriptional landscape within the tumor, which can enable the identification of tumors responsive to tumor- and / or microenvironment-targeted therapies. Moreover, RNA-based biomarkers utilizing RNA sequencing data can allow analysis of thousands of expressed genes, enabling machine learning and statistical methods to identify key biological features as predictors of response. In contrast, DNA biomarkers utilizing DNA sequencing data, particularly those focusing on single-gene mutations, have a limited scope in developing predictive biomarkers using machine learning techniques. Overall, utilizing RNA sequencing data rather than DNA sequencing data can provide a more detailed and dynamic understanding of genetic expression, which can be crucial for effective treatment stratification and predictive analytics in oncology.

[0069] Accordingly, embodiments of the present disclosure allow stratification of responses including assessment of clinical benefit or response to KRAS therapy and / or to a KRAS therapeutic agent using RNA sequencing data as a high-dimensional and multimodal data to analyze KRAS biology, tailoring patient stratification to their unique biological characteristics. At least two key biological aspects or tumor biology parameters - KRAS dependency and activation states of one or more biological signaling pathways at least partly related to KRAS - can identify those patients and / or tumors, which may respond to and / or receive clinical benefitfrom treatment with a KRAS therapeutic agent, a task where existing biomarker solutions are limited. Embodiments of the present disclosure overcome the limitation of currently used DNA- based biomarkers, and can provide the potential to improve guidance for clinical trials, optimize therapeutic outcomes, rationalize effective combination strategies, and accelerate approvals in various medical contexts.

[0070] ‘ ‘KRAS targeted therapy” is to be construed broadly and can refer to a “KRAS therapeutic treatment” and / or a medical treatment targeting the tumor. In particular, KRAS targeted therapy refers to a KRAS therapeutic agent or a combination of KRAS therapeutic agents or a combination of a KRAS therapeutic agent with any other medical treatment targeting the tumor. The KRAS therapeutic agent can be a KRAS inhibitor or KRAS-effector pathway inhibitor. These can include targeted agents against the KRAS protein such as mutation-specific inhibitors, for example G12C- or G12D-targeting agents, or mutation-agnostic agents, for example pan-KRAS inhibitors, molecular glue or multi-component KRAS inhibitors, or KRAS degraders. In addition, a KRAS therapeutic agent may in the context of the present disclosure include a therapy or treatment which does not directly target the KRAS protein but rather targets the transcription, translation, or stability of the KRAS protein or RNA, such as e.g., antisense oligonucleotides, short interfering or short hairpin RNAs, and targeted-or-guided nucleases. Further, a KRAS therapeutic agent may include a vaccination or biologic therapy which may ultimately target aberrant KRAS expressing cells. Hence, the KRAS therapeutic agent can encompass the targeting of proteins involved in the KRAS and / or Ras signaling pathway and / or effector pathways, either as monotherapy or in combination with KRAS-specific targeting agents. The term “therapeutic agent” includes small molecules, antibodies and binding fragments thereof, non-antibody protein scaffold proteins, aptamers and nucleotide-based molecules, such as siRNAs or gRNAs, immunotherapeutic agents, such as chimeric antigen receptors, T cell receptors, immune effector cells, engineered immune effector cells, e.g., genetically engineered T cells, genetically engineered CAR T cells and immunomodulators. In some embodiments, the KRAS inhibitor is a small molecule. In some embodiments, the KRAS inhibitor is a mutation specific small molecule inhibitor. The KRAS inhibitor may only bind to the KRAS inactive state (also termed “KRAS OFF inhibitor”), e.g., KRAS G12C OFF inhibitors, such as Sotorasib or Adagrasib. The KRAS inhibitor may only bind to the KRAS active state (also termed “KRAS ON inhibitor”), for example KRAS G12C ON inhibitors, such as RMC-6291. Exemplary KRAS G12C inhibitors include Sotorasib (Amgen), Adagrasib (Mirati Therap.), JAB-21822 (Jacobio Pharm.), GDC-6036 (Roche), LY3537982 (Eh Lilly), D-1553 (InventisBio), IDQ443 (Novartis), GH35 (Suzhou Genhouse Bio), GFH925 (Zhejiang Genfleet Therap.), BI 1823911 (Boehringer Ingelheim) YL-15293 (Ahanghai YingLi Pharm.), HS-10370 (Jiangsu Hansoh Pharm.), MK-1084 (Merck), BPI-41286 (Betta Pharm). Pan-Ras inhibitors include for example RSC-1255 (RasCals Therapeutics), RMC-6236 (Revolution Medicines), BI 1701963 (Boehringer Ingelheim). An exemplary G12D inhibitor is MRTX1133 (Mirati Therap.).

[0071] As used herein, the phrase “tumor response to a KRAS therapeutic agent” can refer to the reaction of a tumor to treatment with the KRAS therapeutic agent. Such response can, for example, be quantified by, associated with and / or or accompanied by changes in size, structure, metabolic function, and / or molecular characteristics of the tumor. Generally, the tumor response to the KRAS therapeutic agent can be utilized as a metric to assess the efficacy of a treatment of a patient or tumor with the KRAS therapeutic agent. It should be noted that the terms “tumor response” and “patient response” can be interchangeably or synonymously used herein.

[0072] It should be noted that a tumor response can lead to a patient response, at least if a patient’s tumor is investigated, wherein the latter is considered as a “clinical benefit” in the context of the present disclosure. Therefore, as used herein, a tumor response may be accompanied by or lead to patient response, the latter being considered a subset of clinical benefit for the patient.

[0073] RNA sequencing data, also referred to as RNA seq data, can refer to the information obtained through an RNA sequencing (RNA-seq) measurement or test. RNA seq typically uses next-generation sequencing (NGS) to capture a snapshot of the RNA present in a sample tissue, such as tumor tissue, at a given moment. The RNA seq data, thus, can be indicative of and / or contain information about the transcriptome, the complete set and quantity of transcripts in a tumor cell or sample, allowing determination of which genes are actively being expressed and to what extent.

[0074] RNA sequencing data, particularly from a total RNA sequencing protocol, can give a comprehensive view of all the RNA molecules, including messenger RNA (mRNA), ribosomal RNA (rRNA), transfer RNA (tRNA), and non-coding RNA, present in the tumor tissue or sample. This can allow identification of which genes are being transcribed and their level or degree of expression. Alternatively or additionally, RNA sequencing data can providequantitative data on gene expression levels, which may also be referred to herein as gene expression data. For instance, by counting the number of reads that align to specific genes or regions, the expression level of those genes can be quantified, which can provide information about the cellular functions and processes active at the time of sampling.

[0075] The RNA sequencing data can be provided as raw RNA sequence data, for example in the so-called FASTQ format, as aligned RNA sequencing data, for example in the so-called BAM data format, as counts matrices, as standalone expression tables and / or quantified expression levels.

[0076] As used herein, a “computational model” may refer to a computer-implemented mathematical model configured to model or simulate the behavior of the tumor using or based on one or more computational algorithms. The computational models described herein may be implemented at least in part as software, software modules and / or software instructions on or at the computing device.

[0077] Each of the first and the one or more additional computational models may be implemented as dedicated modules at the computing device. Alternatively, at least parts of the first and one or more additional computational models can be implemented in a combined software module at the computing device.

[0078] The first computational model modelling and / or being configured to model one or more aspects of the dependency of the tumor on the KRAS signaling pathway and the sensitivity of the tumor to KRAS inhibitor therapy can represent or include one or more mathematical equations and / or algorithms that describe the relationship between at least one aspect of the dependency and the sensitivity and the tumor being modelled. Alternatively or additionally, the one or more additional computational models modelling and / or being configured to model the activation of one or more biological signaling pathways at least partly related to KRAS can represent or include one or more mathematical equations and / or algorithms that describe the relationship between activation of the one or more biological signaling pathways at least partly related to KRAS and the tumor being modelled.

[0079] As used herein, the one or more additional computational models can define a second computational model, a third computational model, and so forth. Also, it should be noted that the term ‘additional computational model’ can be interchangeably used herein with ‘further computational model’.

[0080] In the context of the present disclosure, the dependency of a tumor on the KRAS signaling pathway may refer to the reliance of cancer or tumor cells on the signals transmitted through the KRAS signaling pathway for their growth, survival, and proliferation.

[0081] The sensitivity of a tumor to KRAS inhibitor therapy may refer to the tumor's ability or inability to evade KRAS therapeutic inhibition and / or to the tumor’s ability or inability to continue growing and proliferating despite attempts to inhibit the KRAS pathway. Resistance can, for example, occur when a tumor develops mechanisms that allow it to bypass the blocked pathway or when it finds alternative pathways to sustain its growth and survival. Accordingly, resistance of the tumor against KRAS inhibitor therapy can be considered as non-dependence of the tumor on the KRAS signaling pathway and / or can be considered as one surrogate to the dependency of the tumor on the KRAS signaling pathway.

[0082] Activation of the KRAS signaling pathway may refer to the KRAS protein, encoded by the KRAS gene, being activated, being turned on and / or being stimulated to transmit signals inside one or more tumor cells.

[0083] Moreover, as used herein, a biological signaling pathway (synonymously or interchangeably used herein with biochemical signaling pathway, signaling pathway, pathway, or biological pathway) may refer to a series of molecular events involving the transmission of a signal from a tumor cell's exterior to its interior, which may lead to a specific response of the cell. These pathways or biological signaling pathways may allow cells to respond to their environment, communicate with each other, and / or execute vital functions like growth, differentiation, and apoptosis.

[0084] In the context of the present disclosure, a biological signaling pathway at least partly related to KRAS may refer to or include the KRAS signaling pathway and / or a biological signaling pathway inter-related and / or inter-connected with the KRAS signaling pathway. The latter can include one or more biological signaling pathways converging on the KRAS signaling pathway, one or more pathways diverging from the KRAS signaling pathway, one or more regulators of the KRAS signaling pathway, one or more pathways regulated by the KRAS signaling pathway, one or more pathways upstream of the KRAS signaling pathway, and / or one or more pathways downstream from the KRAS signaling pathway. Accordingly, the “one or more biological signaling pathways at least partly related to KRAS” can include at least one ofthe KRAS signaling pathway and one or more biological signaling pathways (inter-)related with the KRAS signaling pathway.

[0085] In some embodiments, classifying the tumor and / or patient as being responsive or non- responsive to the KRAS therapeutic agent may include determining and / or coming to a conclusion as to whether or not the tumor and / or patient responds, is likely to respond and / or expected to respond to the KRAS therapeutic agent. Such determination may include or refer to a binary classification into two classes, a first class associated with responsive tumors and / or patients and a second class associated with non-responsive tumors and / or patients. The present disclosure, however, is not limited to binary classification for the response, but multi-class classification in more than two classes is possible.

[0086] Optionally, classifying the tumor and / or patient as being responsive or non-responsive to the KRAS therapeutic agent may include determining, computing and / or calculating a probability and / or likelihood for the tumor and / or patient to respond to the KRAS therapeutic agent and / or KRAS therapy. This can include computing a numerical value for the probability or likelihood, such as e.g. a relative probability between zero and one. Alternatively or additionally, a categorical probability or likelihood, such as a low, medium or high likelihood to respond to respond to the KRAS therapeutic agent, may be computed.

[0087] The classification of the tumor and / or patient being responsive to the KRAS therapeutic agent or not can be based on an output of the first and the one or more additional computational models, and optionally one or more further or additional computational models, such as e.g., based on one or more dependency scores, one or more activation state scores, one or more sensitivity scores, and / or one or more resistance scores.

[0088] The first and one or more additional computational models may be executed in parallel or consecutively to process the RNA sequencing data or at least a part thereof. Also, the first and the one or more additional computational models may process or analyze the entire set of the RNA sequencing data or they may process or analyze a part thereof only. The first and the one or more additional computational models may each process or analyze a part of the RNA sequencing data, wherein the parts or subsets of data processed by the first and the one or more additional computational models may be identical or may differ from one another.

[0089] According to some embodiments, the first computational model is configured to model and / or simulate one or both the dependency of the tumor on the KRAS signaling pathway andthe sensitivity of the tumor to KRAS inhibitor therapy as a continuous tumor biology parameter. In other words, the first computational model may be configured to model and / or simulate one or both the dependency of the tumor on the KRAS signaling pathway and the sensitivity of the tumor to KRAS inhibitor therapy with a continuous metric. Alternatively or additionally, the one or more additional computational models, in particular each of the one or more additional computational models, may be configured to model and / or simulate the activation of one or more biological signaling pathways at least partly related to KRAS as continuous tumor biology parameters and / or with continuous metrics.

[0090] A continuous metric for a tumor biology parameter, such as dependency, sensitivity and activation, may refer to measuring or quantifying the respective tumor biology parameters that can take on an infinite number of values within a given range. Hence, continuous metrics can allow for any value in a given range of values, including fractions or decimals. This can be contrasted with discrete metrics, which can only take on specific, separated values, such as e.g., whole numbers. Therefore, modelling one or more of the dependency, the sensitivity, and the activation as continuous tumor biology parameters and / or with continuous metrics can allow for a higher resolution or granularity of the respective tumor biology parameters, which may, for instance, allow to identify, determine or detect comparatively small changes or differences between different values of the same tumor biology parameter. Hence, more information may be retained in and / or derived from tumor biology parameters modelled with continuous metrics, when compared to modelling on discrete metrics. Also, intercomparability of different values of the same tumor biology parameter may be enhanced.

[0091] Further, continuous metrics for tumor biology parameters, such as dependency, resistance, sensitivity and / or activation, may be advantageous from a treatment assessment perspective, because such continuous parameters or variables may move beyond a binary assessment which can provide more granularity for assigning response to permit assessment across samples (e.g., different patient biopsies), treatment conditions (e.g., cell lines versus xenografts versus patient samples), or across different drugs, treatments, therapeutic agents and / or mechanisms of action. Also, a continuous variable or parameter may serve as a more robust input metric for subsequent modelling, for example using the output of one or more of the first and the at least one additional computational model as input features for a subsequent computational model or classifier. In addition, a continuous variable or parameter may allow formore robust standardization across RNA sequencing modalities or technologies which may permit the use to scale across common clinical diagnostic tests, such as e.g. multiplex PCR or shallow transcriptomic profiling, to also encompass utility with emerging technologies, such as e.g., long read RNA sequencing or spatial transcriptomics. Moreover, continuous parameters or variables can also be binarized using thresholding, which may have particular utility in exploratory modes and testing of multiple hypotheses.

[0092] One or more of the dependencies of the tumor on the KRAS signaling pathway, the sensitivity of the tumor to KRAS inhibitor therapy, and the activation of one or more biological signaling pathways at least partly related to KRAS can be modelled as continuous biology parameters on an arbitrary scale, including a linear scale, a logarithmic scale or other scales. Therein, values of one or more of the dependencies, the sensitivity, and the activation (and / or activation states) may range from a minimum value to a maximum value, in particular from negative values to positive values. The values of one or more of the dependency, the sensitivity, activation can also be referred to herein as scores or numerical scores for the respective tumor biology parameter, i.e., the dependency, the sensitivity, and the activation. In this sense, a value and a score of a tumor biology parameter may be interchangeably or synonymously used herein.

[0093] According to an embodiment, the first computational model and the at least one additional computational models each constitute, provide and / or define a model axis of a computational multi-axis model modelling and / or being configured to model at least two different tumor biology parameters selected from the group of tumor biology parameters consisting of dependency of the tumor on the KRAS signaling pathway, sensitivity of the tumor to KRAS inhibitor therapy, and activation of one or more biological signaling pathways at least partly related to KRAS. A model axis of the computational multi-axis model can be considered as, denote or be referred to as a dimension of the multi-axis model. The multi-axis model can also be referred to herein as multi-dimensional model to predict or determine response to the KRAS therapeutic agent.

[0094] Accordingly, the dependency, the sensitivity, the resistance and the activation can constitute or can be considered as tumor biology parameters in the context of the present disclosure.

[0095] Accordingly, each of the first and the at least one additional computational models can provide, define and / or generate a dimension of the computational multi-axis model to predictresponse of the tumor to the KRAS therapeutic agent. For instance, the first computational model may define a first dimension of the multi-axis model and at least one of the one or more additional computational models, e.g. a second computational model, may define a second model axis different than the first dimension.

[0096] The model axes of the multi-axis model, as used herein, can denote abstract non- congruent axes in a multi-dimensional space spanned or created by said at least two different tumor biology parameters selected from the group of tumor biology parameters consisting of dependency of the tumor on the KRAS signaling pathway, sensitivity of the tumor to KRAS inhibitor therapy, and activation of the one or more biological signaling pathways at least partly related to KRAS.

[0097] It is emphasized that the computational multi-axis model described herein is not limited to two model axes and / or dimensions for modelling two different tumor biology parameters selected from the group of tumor biology parameters consisting of dependency of the tumor on the KRAS signaling pathway, sensitivity of the tumor to KRAS inhibitor therapy, and activation of the one or more biological signaling pathways at least partly related to KRAS. Rather, the computational multi-axis or multi-dimensional model can include more than two axes and / or dimensions, which may be provided and / or defined by one or more further or additional computational models, as described in more detail hereinbelow.

[0098] In an example, evaluating the RNA sequencing data with the first and the at least one additional computational models includes evaluating the RNA sequencing data with a computational multi-axis or multi-dimensional model modelling and / or being configured to model at least two different tumor biology parameters selected from the group of tumor biology parameters consisting of dependency of the tumor on the KRAS signaling pathway, sensitivity of the tumor against KRAS inhibitor therapy, and activation of the one or more biological signaling pathways at least partly related to KRAS. Therein, the computational multi-axis model may include or comprise at least the first computational model and the at least one additional computational model. Generally, such multi-axis or multi-dimensional model can enable a reliable and accurate determination of whether or not the tumor is responsive to the therapeutic agent, in particular when it comes to complex tumor tissue, such as e.g. human tumor tissue obtained from biopsy.

[0099] According to an embodiment, the multi-axis model includes two or more model axes, the two or more model axes comprising at least a first model axis representative of at least one of the dependency of the tumor on the KRAS signaling pathway and the sensitivity of the tumor against KRAS inhibitor therapy, and a second model axis representative of the activation of the one or more biological signaling pathways at least partly related to KRAS. Utilizing at least two model axes can allow to accurately predict response to the KRAS therapeutic agent.

[0100] According to an embodiment, the multi-axis model includes at least one further or additional model axis defined by the at least one additional computational model, the at least one further model axis being representative of one or more of: a cell or tumor intrinsic signaling pathway (e.g. PI3K), a cell or tumor extrinsic signaling pathway (e.g. TGF-beta, angiogenesis), a cellular composition of the tumor (e.g. tumor infiltrating lymphocyte (TIL)-rich), cellular states of one or more tumor cells (e.g. GO / quiescent stage or S-phase / proliferative phase), biological activities of one or more tumor cells (e.g. sustaining proliferative signaling), molecular processes within one or more tumor cells (e.g. epithelial -to-mesenchymal transition), genetic variants (e.g. TP53 somatic mutations), synthetic lethal partners or gene pairs, modifications (e.g. protein post- translational modifications or RNA post-transcriptional modifications), mutations in the KRAS protein or proteins interacting with KRAS that confer resistance to KRAS therapeutic agents, and protein-interaction networks (e.g. the p53 signaling network or pathway). By utilizing one or more additional or further model axes and / or dimensions in the multi-axis model, additional information can be taken into consideration, which can allow to further improve or enhance accuracy in the prediction or determination of the response to the KRAS therapeutic agent, and which may be advantageous for analyzing complex tumor tissue.

[0101] According to an embodiment, the multi-axis model defines a multi-dimensional phenotypic vector space (also referred to herein as phenotypic space or vector space) for the phenotype of the tumor, wherein the multi-dimensional phenotypic vector space includes at least a first region associated with a first phenotype of the tumor responsive to the KRAS therapeutic agent and a second region associated with a second phenotype of the tumor non-responsive to KRAS therapeutic agent. Therein, the phenotypic vector space can have two or more dimensions, respectively, two or more axes defined and / or generated by at least the first and at least one additional computational models. The phenotypic vector space may refer to or be represented asa Cartesian space or Euclidian space in two or more dimensions. However, the present disclosure is not limited in this respect.

[0102] In an example, the phenotypic vector space is a Cartesian space with at least two orthogonal or perpendicular axes defined or generated by at least the first and the at least one additional computational models. Accordingly, a first coordinate within the phenotypic or Cartesian space may be provided by at least one of the dependency of the tumor on the KRAS signaling pathway and the sensitivity of the tumor to KRAS inhibitor therapy. Alternatively or additionally, a second coordinate within the phenotypic or Cartesian space may be provided by the activation of the one or more biological signaling pathways at least partly related to KRAS.

[0103] In yet another example, the phenotypic or Cartesian space is defined by at least two model axes generated or defined by at least the first and the at least one additional computational models can divide a plane into four quadrants, wherein at least two quadrants and / or half planes may be associated with and / or indicative of a different phenotype of the tumor. For instance, the first and second regions of the Cartesian or phenotypic vector space can refer to or denote a first and second quadrant of the Cartesian or phenotypic vector space, respectively. Alternatively or additionally, the first and second regions of the Cartesian or phenotypic vector space can refer to or denote a first and second half plane of the Cartesian or phenotypic vector space, respectively. It should be noted, however, that the first and second regions of the Cartesian or phenotypic vector space do not necessarily have to refer to or denote quadrants of the Cartesian or phenotypic vector space, but rather can refer to dedicated or particular areas and / or regions within said space. Therein, the first and second regions of the Cartesian or phenotypic vector space can at least partly intersect each other, overlap with each other or be spaced apart from each other.

[0104] According to an embodiment, the first region of the phenotypic vector space associated with a first phenotype of the tumor responsive to KRAS therapeutic agent and the second region associated with a second phenotype of the tumor non-responsive to KRAS therapeutic agent may be defined and / or separated by a hyperplane dividing the phenotypic vector space into at least the first and second regions. Depending on the number of dimensions and / or axes considered or utilized for determining KRAS response, the hyperplane may have one or more dimensions.

[0105] In an example, the classification of the tumor as responsive or non-responsive may include determining a hyperplane dividing the phenotypic vector space into at least the first region associated with a first phenotype of the tumor responsive and the second region associated with a second phenotype of the tumor non-responsive to KRAS therapeutic agent.

[0106] According to an embodiment, the method further comprises determining, based on the evaluation of the RNA sequencing data with the first computational model, one or both a dependency score indicative of a degree or level of dependency of the tumor on the KRAS signaling pathway and a sensitivity score indicative of a degree or level of responsiveness of the tumor to KRAS inhibitor therapy. The method may further comprise determining, based on the evaluation of the RNA sequencing data with the one or more additional computational models, one or more activation state scores (also referred to herein as activity scores, biological activation state scores or activation scores) indicative of a degree or level of activation of the one or more biological signaling pathways at least partly related to KRAS. Moreover, the method comprises mapping and / or assigning the determined at least one activation score and at least one of dependency score and the sensitivity score to the multi-dimensional phenotypic vector space defined by the multi-axis model. By determining the at least one activation state score and at least one of the dependency score and the sensitivity score, and by mapping the respective scores to the multi-dimensional phenotypic vector space, the tumor and / or patient can accurately be classified as being responsive or non-responsive to the KRAS therapeutic agent. Hence, tumor and / or patient response to the therapeutic agent can be accurately stratified based on mapping a combination of dependency, sensitivity and / or activation state scores to the phenotypic vector space.

[0107] According to an embodiment, the method further comprises determining that the at least one activation state score and at least one of the dependency score and sensitivity score are located within the first region of the multi-dimensional phenotypic vector space associated with the first phenotype of the tumor responsive to the KRAS therapeutic agent, thereby classifying the tumor as being responsive to the KRAS therapeutic agent. In other words, the tumor and / or patient may be classified as being responsive to the KRAS therapeutic agent based on or by determining that the at least one activation state score and at least one the dependency score and the sensitivity score are located within the first region of the multi-dimensional phenotypic vector space.

[0108] According to an embodiment, the method further comprises determining that the at least one activation state score and at least one of the dependency score and the sensitivity score are located within a second region of the multi-dimensional phenotypic vector space associated with the second phenotype of the tumor non-responsive to the KRAS therapeutic agent, thereby classifying the tumor as being non-responsive to the KRAS therapeutic agent. In other words, the tumor and / or patient may be classified as being non-responsive to the KRAS therapeutic agent based on or by determining that the at least one activation state score and at least one of the dependency score and the sensitivity score are located within the second region of the multidimensional phenotypic vector space.

[0109] As used herein, a score may refer to a numerical measure for and / or may represent or indicate a value of the respective tumor biology parameter. Accordingly, a score can quantify the associated tumor biology parameter. Hence, the dependency score may refer to a value of or may quantify the dependency of the tumor on the KRAS signaling pathway, the sensitivity score may refer to a value of or may quantify the sensitivity of the tumor to KRAS inhibitor therapy, and the activation state score may refer to a value of or may quantify the activation of the one or more biological signaling pathways at least partly related to KRAS. One or more of the dependency score, the sensitivity score and the at least one activation state score can be unitless and / or dimensionless. Accordingly, the dependency score can provide a relative measure for the dependency of the tumor on the KRAS signaling pathway, the sensitivity score can provide a relative measure for the sensitivity of the tumor to KRAS inhibitor therapy, and / or the at least one activation state scores can provide a relative measure for the activation of the one or more biological signaling pathways at least partly related to KRAS.

[0110] The at least one activity or activation score can be positive or negative, indicating whether at least one biological signaling pathway at least partly related to KRAS is activated or inhibited, respectively. As mentioned above, the activation state score or activity score may be dimensionless, reflecting relative changes in KRAS signaling or related biological pathway activity. The value of the activity and / or the activation state score may represent the extent to which the KRAS signaling or related biological pathway is activated or inhibited, for example under certain conditions compared to a baseline or control state. An interpretation of the activation state score can be based on its magnitude and direction. In particular, a positive activation state score may indicate KRAS signaling or related biological pathway activation,wherein the pathway activity may be higher than the reference state. A negative activation state score, on the other hand, can indicate inhibition of the KRAS signaling or related biological pathway, wherein the pathway activity may be lower than the reference state. Accordingly, the activation or activation state score can be unitless and provide a relative measure of the KRAS signaling or related biological pathway activity that can be used to compare across samples, conditions, or time points, in turn facilitating the understanding of biological processes and disease mechanisms.

[0111] KRAS dependency and / or sensitivity may signify that a particular cancer survival or proliferation is dependent or non-dependent on the normal or mutated function of KRAS. Inhibition of KRAS in dependent or sensitive cells can lead to reduced viability or cell death, which may indicate KRAS as a potential therapeutic target in these cancer cells.

[0112] The dependency score and / or sensitivity score may refer to quantitative measures, for example derived from the effects observed upon gene knockdown or knockout in cell lines. These scores may reflect the reduction in cell viability compared to a control condition, for instance normalized across all tested cell lines for a given gene, such as KRAS.

[0113] The dependency score and / or sensitivity score may be dimensionless, expressed as unitless or arbitrary metrics. The dependency score and / or sensitivity score may range from positive to negative values. A negative dependency score and / or a negative sensitivity score may indicate that loss of the KRAS gene has little to no effect on cell viability, which may indicate that the cells are not dependent on the KRAS gene and / or the KRAS signaling pathway. A positive dependency score and / or positive sensitivity score may indicate that loss of the KRAS gene significantly reduces cell viability, which may indicate that the cells are dependent on the KRAS gene and / or the KRAS signaling pathway. The more positive the dependency score, the stronger the dependency, and the more positive the sensitivity score, the stronger the sensitivity to the therapeutic. Thus, a KRAS dependency score that is significantly positive across a set of cancer cells, cell lines, or in a patient biopsy may imply the cancer cells are highly reliant on KRAS for survival.

[0114] According to an embodiment, evaluating the RNA sequencing data with the first computational model includes determining, computing and / or calculating one or both a dependency score indicative of a degree of dependency of the tumor on the KRAS signaling pathway and a sensitivity score indicative of a degree of sensitivity of the tumor to KRASinhibitor therapy. Alternatively or additionally, evaluating the RNA sequencing data with the and one or more additional computational models includes determining, computing and / or calculating one or more activation state scores indicative of a degree of activation of the one or more biological signaling pathways at least partly related to KRAS. A combination of the at least one activation state score and at least one of the determined dependency score and the sensitivity score may then be used to reliably determine whether or not the tumor is responsive to the KRAS therapeutic agent and / or to stratify response to the KRAS therapeutic agent.

[0115] According to an embodiment, the method further comprises providing at least one of the determined dependency score and the sensitivity score, and providing the at least one determined activation state score as inputs to a classifier for classifying the tumor as being responsive or non-responsive to the KRAS therapeutic agent. Therein, the classifier may be a binary classifier predicting the probability that the tumor is responsive or not to the KRAS therapeutic agent, thereby allowing classification of the tumor as being responsive or not to the KRAS therapeutic agent. Generally, utilizing a subsequent classifier receiving the at least one activation state score and at least one of the dependency score and the sensitivity score as inputs may be of particular advantage when discovering or investigating response to the KRAS therapeutic agent in complex tumor tissue, such as human tumor tissue, in particular because one or more additional intrinsic and / or extrinsic tumor biology parameters can easily be taken into consideration in the classification. Also, an impact or influence of one or more additional intrinsic and / or extrinsic tumor biology parameters on the classification result may be investigated.

[0116] According to an embodiment, the classifier is based on logistic regression. In other words, the classifier can be a logistic regression classifier providing a binary classification as to whether the tumor is responsive or non-responsive to the KRAS therapeutic agent. Generally, logistic regression can provide an efficient, malleable classifying approach where the features of the classifier can more easily be identified and interpreted, when compared to more complex classification approaches. More specifically, logistic regression can be advantageous in terms of explainability. In other words, the features selected for the classifier and the weights associated with each of those features can be inferred directly. Another advantage can be that the classifier can be more efficiently trained, validated, and repurposed as compared to more complex modelsor classification approaches. This can give the advantage of being able to accurately tune the parameters, e.g. should additional features or data become available.

[0117] According to various embodiments, the classifier is based on a decision tree. In a decision tree, observations about an item (e.g. a set of input values and / or a set of input models) may be represented by branches of the decision tree, and an output value corresponding to the item may be represented by leaves of the decision tree. Decision trees may support both discrete values and continuous values as output values. If discrete values are used, the decision tree may be denoted a classification tree, if continuous values are used, the decision tree may be denoted a regression tree. A decision tree may have the advantage, e.g., of being easy interpretable, not requiring data to be linearly separable and / or being robust to outliers.

[0118] According to an embodiment, the first computational model and at the at least one additional computational models are independent and / or separate computational models. In other words, the first and the at least one additional computational models can be implemented as separate and / or independent models at the computing device. This may allow to modify one of the models without affecting the other one. Also, the first and the at least one additional computational models may be executed or run in arbitrary sequence or at least in part simultaneously, which may increase overall performance in the determination of the tumor response.

[0119] According to an embodiment, the first computational model that is configured to model at least one of dependency of the tumor on the KRAS signaling pathway and sensitivity of the tumor to KRAS inhibitor therapy is a trained machine learning model or algorithm. In other words, the first computational model may be a machine- learned model.

[0120] For instance, the first computational model may be trained to process and / or analyze the RNA sequencing data in order to compute one or more of the dependency of the tumor on the KRAS signaling pathway and sensitivity of the tumor to KRAS inhibitor therapy. This may include estimating, determining and / or computing one or more of the dependency score and the sensitivity score. In particular for determining dependency and sensitivity based on RNA sequencing data with the amount information contained therein, use of machine learning can be advantageous and allow for an accurate determination of the dependency and / or sensitivity score within reasonable timeframes.

[0121] To train the first computational model, which may also be referred to herein as dependency and / or sensitivity model, an appropriate set of training data can be selected. In a non-limiting example, RNA sequencing data from the Cancer Cell Line Encyclopedia (CCLE) dataset may be used, such as lung cell lines or other cell lines that might be of interest. In particular, data for which the CRISPR-Cas9-based viability screens were performed can be selected from the CCLE dataset as training data set, and one or more dependency and / or sensitivity scores can be calculated based thereon. At least one of the dependency score and the sensitivity score can be used as target variable for the first computational model. As noted above, the dependency score may be indicative of a cell line's reliance on KRAS for survival, whereas the sensitivity score may be indicative of the ability to respond to KRAS therapy, such as e.g. KRAS therapeutic inhibition.

[0122] Further, appropriate features of the first computational model may be selected. For this purpose, genes with stable annotations defined by one or more feature selection criteria can be prioritized. Exemplary feature selection criteria can include whether the gene has a Transcript Support Level (TSL) of one (1), and whether the gene has Matched Annotation from NCBI and EBI (MANE). For instance, TP53 has a MANE (Matched Annotation from NCBI (National Center for Biotechnology Information) and EBI (European Bioinformatics Institute)) annotation, signifying that there is a consensus reference annotation for this gene. Another exemplary feature selection criterion can include whether the gene is included in the Consensus Coding Sequence Project (CCDS). For instance, BRCA1 is included in the Consensus Coding Sequence Project (CCDS), wherein the CCDS collaboration involves several major databases, including NCBI, EBI (Ensembl), and others, aiming to identify a core set of human and mouse protein-coding regions that are consistently annotated and of high quality. Another exemplary feature selection criterion can include whether the gene is present in the Ensembl and Havana database. For instance, SOX9 is present in both the Ensembl and Havana databases. Ensembl is a comprehensive database that provides automated annotations for eukaryotic genomes, while Havana (part of the Vertebrate Genome Annotation (Vega) database) provides manually curated annotations of vertebrate genomes. The presence of SOX9 in both databases ensures that its annotations benefit from both automated computational approaches and expert manual curation. Based on one, a plurality or all of these feature selection criteria genes may be prioritized and / or selected as possible feature candidates for the first computational model. Optionally, genes thatare robustly expressed may be selected. For instance, only genes with gene expression levels of greater than 100 counts at the 80th percentile may be selected.

[0123] Further, it may be determined, which of the selected genes may be relevant to and / or related to KRAS biology. This can be empirically determined and / or based on published data. In an example, gene signatures from published papers may be combined with Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways describing one or more of MAPK, Ras, and ErbB signaling pathways.

[0124] Training of the first computational model may employ recursive feature elimination with LI and L2 regularized regression. In other words, the first computational model may be trained based on LI regularization and L2 regularization.

[0125] Hyperparameter tuning can be achieved through nested cross-validation. Validation of the first computational model can then be conducted using real-world patient data, ensuring its applicability in clinical settings.

[0126] According to an embodiment, the first computational model is a trained machine learning model configured to model dependency of the tumor on the KRAS signaling pathway and / or sensitivity of the tumor to KRAS inhibitor therapy based on model features associated with a plurality of genes involved in one or more of the RAS signaling, the MAPK signaling, the PI3K signaling, the EGFR signaling pathways, metabolic reprogramming and immune dysregulation. In other words, the first computational model may be trained to determine dependency and / or sensitivity, and / or corresponding scores, based on evaluating the RNA sequencing data with respect to model features associated with a plurality genes involved in one or more of the RAS signaling, the MAPK signaling, the PI3K signaling, the EGFR signaling pathways, metabolic reprogramming and immune dysregulation.

[0127] According to an embodiment, at least one of the one or more additional computational models that are configured to model activation of one or more biological signaling pathways at least partly related to KRAS is a statistical model. For instance, at least one of the additional computational models may include one or more statistical model components and / or mathematical equations to compute, calculate and / or determine one or more activation state scores based on RNA sequencing data. Such a statistical approach may allow to efficiently compute activation state scores based on statistically analyzing the RNA sequencing data. It should be noted that while at least one of the additional computational models may be astatistical model, one or more other computational models of the additional computational model can be a trained machine learning model.

[0128] In an exemplary implementation, at least one of the one or more additional computational models may employ a statistical approach to infer the activity or activation state of one or more biological signaling pathways (interchangeably used herein with biochemical signaling pathway) from gene expression data by analyzing the collective expression of one or more predefined sets of responsive genes associated with each pathway.

[0129] In a further exemplary implementation, the RNA sequencing data and / or the gene expression data contained therein may be normalized and scaled. This may ensure comparability across experiments.

[0130] Further, at least one of the one or more additional computational models may utilize or invoke one or more statistical model components, such as for example for regression analysis and / or principal component analysis, to integrate the expression levels of these gene sets, thereby producing and / or computing dimensionless pathway activation state scores indicative of a degree or level of activation of the one or more biological signaling pathways at least partly related to KRAS. As described hereinabove, activation state scores can in principle be positive or negative, with positive values or scores indicating the extent of pathway activation and negative values or scores indicating inhibition relative to a predefined baseline.

[0131] According to an embodiment, at least one of the one or more additional computational models is a statistical model configured to model activation of the one or more biological signaling pathways at least partly related to KRAS and / or to compute one or more activation state scores based on quantifying gene expression of a plurality of genes involved in one or more biological or biochemical signaling pathways at least partly related to KRAS.

[0132] According to an exemplary embodiment, the one or more biological or biochemical signaling pathways at least partly related to KRAS may include but are not limited to one or more of Androgen Receptor signaling pathway, Estrogen Receptor signaling pathway, Hypoxia signaling pathway, JAK-STAT signaling pathway, MAPK signaling pathway, NFkB signaling pathway, Notch signaling pathway, PI3K / AKT signaling pathway, p53 signaling pathway, RTK signaling pathway, TGFb signaling pathway, TNFa signaling pathway, Trail signaling pathway, VEGF signaling pathway, and WNT / p-Catenin signaling pathway.

[0133] According to yet another exemplary embodiment, the at least one additional computational model may be configured to compute and / or determine, based on the RNA sequencing data, gene expression data contained therein and / or variant data contained therein, one or more intermediate pathway scores indicative of the activity and / or activation of one or more signaling pathways selected from the group consisting of Androgen Receptor signaling pathway, Estrogen Receptor signaling pathway, Hypoxia signaling pathway, JAK-STAT signaling pathway, MAPK signaling pathway, NFkB signaling pathway, Notch signaling pathway, PI3K / AKT signaling pathway, p53 signaling pathway, RTK signaling pathway, TGFb signaling pathway, TNFa signaling pathway, Trail signaling pathway, VEGF signaling pathway, and WNT / p-Catenin signaling pathway. The intermediate pathway scores may be used individually or uniquely, optionally with different weights for different scores. Alternatively, the at least one additional computational model may be configured to combine, merge and / or add one or more of these intermediate pathway scores to compute the activation state score indicative of the degree or level of activation of the one or more biological signaling pathways at least partly related to KRAS. For combining a plurality of intermediate pathway scores one or more normalization, standardization and / or statistical techniques may be applied.

[0134] In an example, the at least one additional computational model may quantify the activity level of one or more biological pathways by analyzing gene expression data, wherein expression patterns of specific gene sets, referred to as "pathway signatures," may be used to indicate pathway activity and / or compute a corresponding intermediate pathway score. The RNA sequencing data and / or gene expression and / or variant data contained therein may be processed by the at least one additional computational model, which may include normalization and scaling. Further, one or more statistical methods, such as regression and / or principal component analysis, may be applied to compute one or more intermediate pathway scores. These scores, indicating activation or inhibition as a positive or negative number respectively, provide a quantitative measure of pathway activity for the individual signaling pathways relative to a baseline.

[0135] According to an embodiment, the method further comprises evaluating the RNA sequencing data with at least one further computational model configured to model at least one further tumor parameter. The at least one further tumor parameter may be a tumor intrinsic parameter indicative of tumor characteristics inherent to the tumor. Alternatively or additionally,the at least one further tumor parameter may be a tumor extrinsic parameter indicative of a tumor microenvironment. It should be noted that in the context of the present disclosure the ‘at least one further computational model’ may be part of the ‘one or more additional computational models’, as used herein.

[0136] In an example, the at least one further tumor parameter may include one or more cell or tumor intrinsic signaling pathways, one or more cell or tumor extrinsic signaling pathways, cellular composition of the tumor, cellular states of one or more tumor cells, biological activities of one or more tumor cells, molecular processes, genetic variants, synthetic lethal partners, metabolic signatures, modifications, RNA sequencing metadata (such as e.g. read length, error rate, sequencing method), demographic data of the patient, and protein-interaction networks. It is emphasized that at least one, a plurality of, or all of the aforementioned further tumor parameters may be considered.

[0137] For instance, the at least one further tumor parameter may be indicative of one or more of: a cell or tumor intrinsic signaling pathway (e.g. PI3K), a cell or tumor extrinsic signaling pathway (e.g. TGF-beta), a cellular composition of the tumor (e.g. tumor infiltrating lymphocyte (TIL)-rich), cellular states of one or more tumor cells (e.g. GO / quiescent stage or S-phase / proliferative phase), biological activities of one or more tumor cells (e.g. Sustaining Proliferative Signaling), RNA sequencing metadata, demographic data of the patient, molecular processes within one or more tumor cells (e.g. epithelial-to-mesenchymal Transition), genetic variants (e.g. TP53 somatic mutations), synthetic lethal partners, metabolic signatures (e.g. aerobic glycolysis), modifications (e.g. protein post-translational modifications or RNA post-transcriptional modifications), and protein-interaction networks (e.g. the p53 signaling network or pathway)

[0138] According to an embodiment, the method further comprises evaluating the RNA sequencing data with at least one further computational model, e.g. a further activation state model, configured to model at least one tumor parameter. The at least one further computational model may be configured to model activation of one or more biological signaling pathways and / or to compute one or more activation state scores for one or more biological signaling pathways based on quantifying gene expression of a plurality of genes involved in the respective one or more biological signaling pathways. In an example, the one or more biological signaling pathways may include one or more signaling pathways selected from the group consisting of Androgen Receptor signaling pathway, Estrogen Receptor signaling pathway, Hypoxia signalingpathway, JAK-STAT signaling pathway, MAPK signaling pathway, NFkB signaling pathway, Notch signaling pathway, PI3K / AKT signaling pathway, p53 signaling pathway, RTK signaling pathway, TGFb signaling pathway, TNFa signaling pathway, Trail signaling pathway, VEGF signaling pathway, and WNT / P-Catenin signaling pathway. The at least one further computational model may be part of the ‘one or more additional computational models’.

[0139] According to an embodiment, the method further comprises evaluating the RNA sequencing data with at least one further computational model (which may be part of or a subset of the ‘one or more additional computational models’) configured to model at least one further tumor parameter, the at least one further tumor parameter being indicative of one or more of a microenvironment of the tumor (also referred to as tumor microenvironment, TME), RNA sequencing metadata, demographic data of the patient, angiogenesis of the tumor, activity of the Vascular Endothelial Growth Factor (VEGF) signaling pathway, and cellular composition of the tumor. Therein, the at least one further computational model may be a trained machine learning model. It is emphasized that also a plurality of further computational models to model a plurality of further tumor extrinsic or intrinsic parameters can be used to predict KRAS response, which can be of particular advantage for complex tissue, such as human tumor tissue.

[0140] In the context of the present disclosure, the tumor microenvironment (TME) refers to the environment surrounding a tumor, including the surrounding blood vessels, immune cells, fibroblasts, signaling molecules, and the extracellular matrix. Angiogenesis is the physiological process through which new blood vessels form from pre-existing vessels. The VEGF (Vascular Endothelial Growth Factor) signaling pathway is a mechanism in the body that regulates the formation of new blood vessels or angiogenesis.

[0141] In an exemplary implementation, one or more further tumor parameters can constitute or define one or more further model axes or dimensions of the computational multi-axis model as described herein. In other words, the computational multi-axis model may include one or more further model axes or dimensions representative of one or more further tumor parameters, as described herein.

[0142] According to an embodiment, the method further comprises determining, based on at least one further computational model, at least one further tumor parameter indicative of one or more of a TME, RNA sequencing metadata, demographic data of the patient, angiogenesis of the tumor, activity of the VEGF signaling pathway, and cellular composition of the tumor. Therein,one or more of these further tumor parameters may be determining using one or more machine learning model as one or more further computational model.

[0143] For instance, a machine learning-based transcriptomic biomarker may be utilized as a further computational model to compute values of said one or more further tumor parameters. Such machine learning-based transcriptomic biomarker may be configured to predict therapeutic responses in various cancers and may utilize a panel algorithm trained on a signature of a plurality of genes which may be optimized across different solid tumors. The panel may categorize the TME into various biological states or classes. Alternatively or additionally, one or more activation state scores for individual signaling pathways may be computed based on one or more further or additional computational models. For instance, one or more activation state scores for one or more extrinsic-associated pathways including specific immune pathways (e.g. JAK STAT, TGF-beta) and angiogenic pathways (e.g. VEGF). Further, the cellular composition (e.g. tumor content) may be assessed, for example as a quality control metric to ensure adequacy and standardization of the biopsy or tissue sample.

[0144] According to an embodiment, the method further comprises evaluating the RNA sequencing data with at least one further computational model (which may be part of the ‘one or more additional computational models’) configured to model at least one further tumor parameter, the at least one further tumor parameter being indicative of the activity and / or activation of one or more signaling pathways selected from the group consisting of Androgen Receptor signaling pathway, Estrogen Receptor signaling pathway, Hypoxia signaling pathway, JAK-STAT signaling pathway, MAPK signaling pathway, NFkB signaling pathway, Notch signaling pathway, PI3K / AKT signaling pathway, p53 signaling pathway, RTK signaling pathway, TGFb signaling pathway, TNFa signaling pathway, Trail signaling pathway, VEGF signaling pathway, and WNT / p-Catenin signaling pathway. The activity of these pathways may form independent axes of the computational multi-axis model, which can be assessed in isolation or concurrently with other axes or dimensions, for example based on the first, the second and / or one or more further computational models described herein.

[0145] According to an embodiment, providing the RNA sequencing data at the computing device includes one or more of accessing the RNA sequencing data with the computing device, obtaining the RNA sequencing data at the computing device, receiving the RNA sequencing data at the computing device, and retrieving the RNA sequencing data with the computing device.

[0146] In an example, the computing device may include a data storage and the RNA seq data may accessed at and / or retrieved from the data storage. Alternatively or additionally, the computing device may include a communication interface communicatively couplable to an external data source, and the computing device may be configured to receive the RNA seq data via the communication interface from the external data source.

[0147] According to an embodiment, providing the RNA sequencing data at the computing device includes measuring the RNA sequencing data based on one or more RNA sequencing modalities including total-RNA sequencing, polyA enriched sequencing, and RNA-Exome-based sequencing. It should be noted that the present disclosure is not limited to a particular modality for acquiring the RNA seq data, but any modality or any combination of different modalities can be used.

[0148] According to an embodiment, the RNA sequencing data include data associated with human tumor tissue, with human-derived tumor tissue, or preclinical tumor tissue. In particular, RNA seq data may be obtained based on or from one or more tumor cells. Human tumor tissue may for example be obtained via patient biopsy, liquid biopsy, solid tumor biopsy, human- derived tumor tissue may be obtained based on primary cell lines or Patient derived xenografts (PDXs), and preclinical tumor tissue may be obtained from preclinical models, such as e.g. rat, mouse, etc. either in the form of organisms or in the form of derivatives such as cell lines.

[0149] According to an embodiment, the RNA sequencing data include data from one or more sources including fresh tissue, biofluid, Formalin-Fixed Paraffin-Embedded, cell culture, and a cell-free source.

[0150] According to an embodiment, the method further comprises normalizing the RNA sequencing data based on one or more of a counts per million normalization (CPM), a fragments per kilobase million normalization (FPKM), a transcripts per million normalization (TPM), an upper quartile normalization, a normalization with respect to counts adjusted with upper quartile factors, a normalization with respect to trimmed mean of M- values, and a normalization with respect to counts adjusted with trimmed mean of M-values factors. It should be noted that a plurality of normalization techniques may be combined to normalize the RNA sequencing data.

[0151] Based on CPM normalization, raw read counts can be scaled to account for differences in library size, wherein the number of reads per million reads in a sample can be calculated,which can e.g. allow for comparison of expression levels across samples with different sequencing depths.

[0152] Based on FPKM normalization both library size and gene length may be adjusted for, wherein the number of fragments mapped to a gene per kilobase of transcript length per million mapped reads may be computed, which can e.g. make it suitable for comparing the expression of genes within and across samples.

[0153] Based on TPM normalization the read counts may be divided by the length of each gene in kilobases, wherein these values may be normalized by scaling them so that the sum across all genes is one million, which can e.g. make the values more directly comparable across samples.

[0154] Based on Upper Quartile Normalization gene expression data may be normalized based on the upper quartile of the counts distribution, wherein the counts may be scaled so that the upper quartile is the same across samples. Thereby, the impact of highly expressed genes on the overall expression profile may be minimized.

[0155] Normalization with Respect to Counts Adjusted with Upper Quartile Factors may refer to an extension of upper quartile normalization, based on which counts can be adjusted based on factors derived from the upper quartile values, aiming to refine the normalization process by closely considering the distribution of gene expression values.

[0156] According to an embodiment, the RNA sequencing data is acquired prior to administration of the KRAS therapeutic agent, during treatment with the KRAS therapeutic agent or after administration of the KRAS therapeutic agent. Accordingly, the method may be applied irrespective of the actual administration or treatment of a patient with the KRAS therapeutic agent.

[0157] For instance, it may be determined whether or not a tumor is responsive to KRAS therapy and / or the KRAS therapeutic agent without potentially altering dependency, sensitivity and / or activation of the one or more biological signaling pathways at least partly related to KRAS by analyzing the RNA seq data acquired prior to administration of the KRAS therapeutic agent. When analyzing RNA seq data acquired during treatment, the evolution of the treatment may be analyzed. Further, by analyzing RNA seq data acquired after administration, potential adaptions to the KRAS therapeutic agent may be analyzed.

[0158] According to an embodiment, the RNA sequencing data is acquired prior to administration of the KRAS therapeutic agent, and the method further comprises obtaining further RNA sequencing data acquired after administration of the KRAS therapeutic agent, and determining, based on evaluating the further RNA sequencing data with the first computational model and the one or more additional computational models, an evolution and / or alteration of the tumor response to the KRAS therapeutic agent. Generally, this may include determining one or more modifications and / or alterations in one or more of the dependency on the KRAS signaling pathway, the sensitivity to KRAS therapy, and the activity or activation of one or more biological signaling pathways at least partly related to KRAS.

[0159] According to an embodiment, the KRAS therapeutic agent is a KRAS inhibitor.

[0160] According to an embodiment, the KRAS therapeutic agent is a mutation-specific inhibitor or a pan-KRAS inhibitor.

[0161] According to an embodiment, the KRAS therapeutic agent is a G12C inhibitor or a KRAS G12D inhibitor.

[0162] According to an embodiment, the tumor is related to lung cancer, non-small cell lung cancer, pancreatic cancer, or a colorectal cancer. Therein, the tumor may be a primary tumor or a secondary tumor. It should be noted that the present disclosure is not limited to a particular cancer or tumor type but can be used to advantage for many cancer or tumor types.

[0163] A second aspect of the present disclosure relates to a computer program, which when executed on a computing device, instructs the computing device to perform steps of the method according to the first aspect of the present disclosure.

[0164] A third aspect of the present disclosure relates to a computer-readable medium, for example a non-transitory computer-readable medium, storing a computer program according to the preceding claim.

[0165] A fourth aspect of the present disclosure relates to a computing device or system with one or more processors for data processing, wherein the computing device is configured to perform steps of the method according to the first aspect of the present disclosure.

[0166] The computing device or system may be a standalone computing device, a computing network, a cloud computing network, a server, a mobile computing device, such as e.g. a smartphone, tablet PC or laptop, or any other computational data processing device.

[0167] The computing device may include a data storage storing and / or configured to store at least the RNA sequencing data. Alternatively or additionally, software instructions and / or a computer program may be stored at the data storage or memory, which, when executed on the computing device, instructs the computing device to perform steps of the method according to the first aspect of the present disclosure.

[0168] The computing device may further comprise a communication interface for communicatively coupling the computing device to one or more external data sources and / or external computing devices.

[0169] The computing device may further comprise a user interface for receiving one or more user instructions and / or for providing information to a user.

[0170] It is emphasized that any feature, step, function, element, technical effect and / or advantage described herein with reference to one aspect of the disclosure equally applies to any other aspect of the disclosure.

[0171] Examples

[0172] Provided below is a non-exhaustive list of non-limiting examples. Any one or more of the features of the below examples may be combined with any one or more features of another example, embodiment, or aspect described herein.

[0173] Example 1A: A computer-implemented method of determining a tumor response to a KRAS therapeutic agent and / or of determining an efficacy of a KRAS therapeutic agent against a tumor, the method comprising:• providing, at a computing device including one or more processors for data processing, RNA sequencing data associated with tumor tissue, the RNA sequencing data being indicative of gene expression data of a plurality of genes involved in one or more biological signaling pathways at least partly related to KRAS;• evaluating the RNA sequencing data with a first computational model modelling at least one of dependency of the tumor on the KRAS signaling pathway, and sensitivity of the tumor to KRAS inhibitor therapy;• evaluating the RNA sequencing data with one or more additional computational models modelling activation of one or more biological signaling pathways at least partly related to KRAS; and• classifying, based on the evaluation of the RNA sequencing data with the first computational model and the one or more additional computational models, the tumor as being responsive or non-responsive to the KRAS therapeutic agent.

[0174] Example IB: A computer-implemented method of determining a tumor response to a KRAS therapeutic agent and / or of determining an efficacy of a KRAS therapeutic agent against a tumor, the method comprising:• providing, at a computing device including one or more processors for data processing, RNA sequencing data associated with tumor tissue, the RNA sequencing data being indicative of gene expression data of a plurality of genes involved in one or more biological signaling pathways at least partly related to KRAS;• evaluating the RNA sequencing data with a first computational model modelling at least one of dependency of the tumor on the KRAS signaling pathway, and sensitivity of the tumor to KRAS inhibitor therapy;• evaluating the RNA sequencing data with at least one second computational model modelling activation of one or more biological signaling pathways at least partly related to KRAS; and• classifying, based on the evaluation of the RNA sequencing data with the first computational model and the at least one second computational model, the tumor as being responsive or non-responsive to the KRAS therapeutic agent.

[0175] Example 1C: A computer-implemented method of determining a tumor response to a KRAS therapeutic agent and / or of determining an efficacy of a KRAS therapeutic agent against a tumor, the method comprising:• providing, at a computing device including one or more processors for data processing, RNA sequencing data associated with tumor tissue, the RNA sequencing data being indicative of gene expression data of a plurality of genes involved in one or more biological signaling pathways at least partly related to KRAS;• evaluating the RNA sequencing data with a first computational model modelling at least one of dependency of the tumor on the KRAS signaling pathway, and sensitivity of the tumor to KRAS inhibitor therapy; and• classifying, based on the evaluation of the RNA sequencing data with the first computational model, the tumor as being responsive or non-responsive to the KRAS therapeutic agent.

[0176] Example ID: A computer-implemented method of determining a tumor response to a KRAS -targeted therapy and / or of determining an efficacy of a KRAS -targeted therapy against a tumor, the method comprising:• providing, at a computing device including one or more processors for data processing, RNA sequencing data associated with tumor tissue, the RNA sequencing data being indicative of gene expression data of a plurality of genes involved in one or more biological signaling pathways, e.g. at least partly related to KRAS;• evaluating the RNA sequencing data with a first computational model modelling at least one of dependency of the tumor on the KRAS signaling pathway, and sensitivity of the tumor to KRAS inhibitor therapy;• evaluating the RNA sequencing data with at least one additional computational model modelling activation of one or more biological signaling pathways at least partly related to KRAS and / or modelling at least one further parameter; and• classifying, based on the evaluation of the RNA sequencing data with the first computational model and additional computational models, the tumor as being responsive or non-responsive to the KRAS-targeted therapy.

[0177] Example IE: A computer-implemented method of determining a tumor response to a tumor therapy and / or of determining an efficacy of a tumor therapy agent against a tumor, the method comprising:• providing, at a computing device including one or more processors for data processing, RNA sequencing data associated with tumor tissue, the RNA sequencing data being indicative of gene expression data of a plurality of genes involved in one or more biological signaling pathways at least partly related to the tumor;• evaluating the RNA sequencing data with a first computational model modelling at least one of dependency of the tumor on one or more signaling pathways, and sensitivity of the tumor to tumor therapy;• evaluating the RNA sequencing data with a at least one additional computational model modelling activation of one or more signaling pathways and / or at least one further tumor parameter; and• classifying, based on the evaluation of the RNA sequencing data with the first computational model and the at least one additional computational model, the tumor as being responsive or non-responsive to the tumor therapy.

[0178] Example IF: A computer-implemented method of determining a tumor response to a tumor therapy, e.g. a KRAS targeted therapy and / or a KRAS therapeutic agent, and / or a method of determining an efficacy of a tumor therapy against a tumor, the method comprising:• providing, at a computing device including one or more processors for data processing, RNA sequencing data associated with tumor tissue, the RNA sequencing data being indicative of gene expression data of a plurality of genes involved in one or more biological signaling pathways, e.g. at least partly related to KRAS and / or the tumor;• evaluating the RNA sequencing data with a first computational model modelling at least one of dependency of the tumor on one or more signaling pathways, and sensitivity of the tumor against tumor therapy;• evaluating the RNA sequencing data with at least one additional computational model modelling at least one further tumor parameter; and• classifying, based on the evaluation of the RNA sequencing data with the first computational model and the at least one additional computational model, the tumor as being responsive or non-responsive to the tumor therapy,• preferably wherein the at least one further tumor parameter includes one or more of activation of the KRAS signaling pathway, activation of one or more cell or tumor intrinsic signaling pathways, activation of one or more cell or tumor extrinsic signaling pathways, RNA sequencing metadata, demographic data of the patient, cellular composition of the tumor, cellular states of one or more tumor cells, biological activities of one or more tumor cells, molecular processes, genetic variants, synthetic lethal partners, metabolic signatures, modifications, protein-interaction networks, tumor microenvironment, angiogenesis of the tumor, activity or activation of the Vascular Endothelial Growth Factor signaling pathway, and activation of one or more biological signaling pathways selected from the group consisting of Androgen Receptor signaling pathway, Estrogen Receptor signaling pathway, Hypoxia signaling pathway, JAK-STAT signaling pathway, MAPK signaling pathway, NFkB signaling pathway, Notch signaling pathway, PI3K / AKT signaling pathway, p53 signaling pathway, RTK signaling pathway, TGFb signaling pathway, TNFa signaling pathway, Trail signaling pathway, VEGF signaling pathway, and WNT / P-Catenin signaling pathway.

[0179] Example 2: The method according to any one of the preceding examples, wherein the first computational model is configured to model one or both the dependency of the tumor on the KRAS signaling pathway and the sensitivity of the tumor to KRAS inhibitor therapy as continuous tumor biology parameter and / or with a continuous metric; and / or wherein the one or more additional computational models are configured to model the activation of said one or more biological signaling pathways at least partly related to KRAS as continuous tumor biology parameter and / or with a continuous metric.

[0180] Example 3: The method according to any one of the preceding examples, wherein the first computational model and the one or more additional computational models each define a model axis of a computational multi-axis model modelling at least two different tumor biology parameters selected from the group consisting of dependency of the tumor on the KRAS signaling pathway, sensitivity of the tumor to KRAS inhibitor therapy, and activation of the one or more biochemical signaling pathways at least partly related to KRAS.

[0181] Example 4: The method according to the preceding example, wherein the multi-axis model includes two or more model axes, the two or more model axes comprising at least a first model axis representative of at least one of the dependency of the tumor on the KRAS signaling pathway and the sensitivity of the tumor to KRAS inhibitor therapy, and a second model axis representative of the activation of the one or more biological signaling pathways at least partly related to KRAS.

[0182] Example 5: The method according to the preceding example, wherein the multi-axis model includes at least one further model axes representative of one or more of: one or more cell or tumor intrinsic signaling pathways, one or more cell or tumor extrinsic signaling pathways, cellular composition of the tumor, cellular states of one or more tumor cells, biological activities of one or more tumor cells, molecular processes, genetic variants, synthetic lethal partners, metabolic signatures, modifications, and protein-interaction networks.

[0183] Example 6: The method according to any one of examples 3 to 5, wherein the multiaxis model defines a multi-dimensional phenotypic vector space for the phenotype of the tumor;and wherein the multi-dimensional phenotypic vector space includes at least a first region associated with a first phenotype of the tumor responsive to the KRAS therapeutic agent and a second region associated with a second phenotype of the tumor non-responsive to KRAS therapeutic agent.

[0184] Example 7: The method according to any one of examples 3 to 6, further comprising:• determining, based on the evaluation of the RNA sequencing data with the first computational model, one or both a dependency score indicative of a degree of dependency of the tumor on the KRAS signaling pathway and a sensitivity score indicative of a degree of responsiveness of the tumor to KRAS inhibitor therapy;• determining, based on the evaluation of the RNA sequencing data with the one or more additional computational models, at least one activation state score indicative of a degree of activation of the one or more biological signaling pathways at least partly related to KRAS; and• mapping the determined at least one activation state score and at least one of the dependency score and the sensitivity score into the multi-dimensional phenotypic vector space defined by the multi-axis model.

[0185] Example 8: The method according to Examples 6 and 7, further comprising:• determining that the at least one activation state score and at least one of the dependency score and the sensitivity score are located within the first region of the multi-dimensional phenotypic vector space associated with the first phenotype of the tumor responsive to the KRAS therapeutic agent, thereby classifying the tumor as being responsive to the KRAS therapeutic agent; and / or• determining that the at least one activation state score and at least one of the dependency score and the sensitivity score are located within the second region of the multi-dimensional phenotypic vector space associated with the second phenotype of the tumor non-responsive to the KRAS therapeutic agent, thereby classifying the tumor as being non-responsive to the KRAS therapeutic agent.

[0186] Example 9: The method according to any one of the preceding examples, wherein evaluating the RNA sequencing data with the first computational model includes determining one or both a dependency score indicative of a degree of dependency of the tumor on the KRAS signaling pathway and a sensitivity score indicative of a degree of responsiveness of the tumor to KRAS inhibitor therapy; and / or wherein evaluating the RNA sequencing data with the one ormore additional computational models includes determining one or more activation state scores indicative of a degree of activation of the one or more biological signaling pathways at least partly related to KRAS.

[0187] Example 10: The method according to the preceding example, further comprising: providing the at least one activation state score and at least one of the dependency score and the sensitivity score as inputs to a classifier for classifying the tumor as being responsive or non- responsive to the KRAS therapeutic agent.

[0188] Example 11: The method according to the preceding example, wherein the classifier is based on logistic regression.

[0189] Example 12: The method according to any one of the preceding examples, wherein the first computational model and the one or more additional computational models are independent computational models.

[0190] Example 13: The method according to any one of the preceding examples, wherein the first computational model is a trained machine learning model; and / or wherein at least one of the additional computational models are statistical models.

[0191] Example 14: The method according to any one of the preceding examples, wherein the first computational model is a trained machine learning model configured to model dependency of the tumor on the KRAS signaling pathway and / or sensitivity of the tumor to KRAS inhibitor therapy based on model features associated with a plurality genes involved in one or more of the RAS signaling, the MAPK signaling, the PI3K signaling, and the EGFR signaling pathways.

[0192] Example 15: The method according to the preceding example, wherein the first computational model is trained based on LI regularization and L2 regularization.

[0193] Example 16: The method according to any one of the preceding examples, wherein at least one of the one or more additional computational models is a statistical model configured to model activation of one or more biological signaling pathways at least partly related to KRAS based on quantifying gene expression or variants of a plurality of genes involved in one or more biological signaling pathways at least partly related to KRAS.

[0194] Example 17: The method according to the preceding example, wherein the one or more biological signaling pathways at least partly related to KRAS include Androgen Receptor signaling pathway, Estrogen Receptor signaling pathway, Hypoxia signaling pathway, JAK-STAT signaling pathway, MAPK signaling pathway, NFkB signaling pathway, Notch signaling pathway, PI3K / AKT signaling pathway, p53 signaling pathway, RTK signaling pathway, TGFb signaling pathway, TNFa signaling pathway, Trail signaling pathway, and WNT / p-Catenin signaling pathway.

[0195] Example 18: The method according to any one of the preceding examples, further comprising: evaluating the RNA sequencing data with at least one further computational model configured to model at least one further tumor parameter, wherein the at least one further tumor parameter is indicative of one or more of a microenvironment of the tumor, RNA sequencing metadata, demographic data of the patient, angiogenesis of the tumor, activity of the VEGF signaling pathway, and cellular composition of the tumor, preferably wherein the at least one further computational model is a trained machine learning model.

[0196] Example 19: The method according to any one of the preceding examples, wherein providing the RNA sequencing data at the computing device includes one or more of accessing the RNA sequencing data with the computing device, obtaining the RNA sequencing data at the computing device, receiving the RNA sequencing data at the computing device, and retrieving the RNA sequencing data with the computing device.

[0197] Example 20: The method according to any one of the preceding examples, wherein providing the RNA sequencing data at the computing device includes measuring the RNA sequencing data based on one or more RNA sequencing modalities including total-RNA sequencing, polyA enriched sequencing, and RNA-Exome-based sequencing.

[0198] Example 21: The method according to any one of the preceding examples, wherein the RNA sequencing data include data associated with human tumor tissue, with human-derived tumor tissue, or preclinical tumor tissue.

[0199] Example 22: The method according to any one of the preceding examples, wherein the RNA sequencing data include data from one or more sources including fresh tissue, biofluid, Formalin-Fixed Paraffin-Embedded, cell culture, and a cell-free source.

[0200] Example 23: The method according to any one of the preceding examples, further comprising:• normalizing the RNA sequencing data based on one or more of a counts per million normalization, a fragments per kilobase million normalization, a transcripts per million normalization, an upper quartile normalization, a normalization with respect to counts adjustedwith upper quartile factors, a normalization with respect to trimmed mean of M- values, and a normalization with respect to counts adjusted with trimmed mean of M-values factors; and / or• aligning the RNA sequencing data to a reference genome; and / or• quantifying the RNA sequencing data.

[0201] Example 24: The method according to any one of the preceding examples, wherein the RNA sequencing data is acquired prior to administration of the KRAS therapeutic agent, during treatment with the KRAS therapeutic agent or after administration of the KRAS therapeutic agent.

[0202] Example 25: The method according to any one of the preceding examples, wherein the RNA sequencing data is acquired prior to administration of the KRAS therapeutic agent, and wherein the method further comprises:• obtaining further RNA sequencing data acquired after administration of the KRAS therapeutic agent; and• determining, based on evaluating the further RNA sequencing data with the first computational model and the one or more additional computational models, an evolution and / or alteration of the tumor response to the KRAS therapeutic agent.

[0203] Example 26: The method according to any one of the preceding examples, wherein the KRAS therapeutic agent is a KRAS inhibitor.

[0204] Example 27: The method according to any one of the preceding examples, wherein the KRAS therapeutic agent is a mutation specific inhibitor or a pan-KRAS inhibitor.

[0205] Example 28: The method according to any one of the preceding examples, wherein the KRAS therapeutic agent is a G12C inhibitor or a KRAS G12D inhibitor.

[0206] Example 29: The method according to any one of the preceding examples, wherein the tumor is related to lung cancer, non-small cell lung cancer, pancreatic cancer, or a colorectal cancer.

[0207] Example 30: The method according to any one of the preceding examples, wherein the first computational model is based on one or more genes selected from the group consisting of KRAS, RASSF1, ETV1, TP53, SIKE1, ANTXR2, DUSP6, MLPH, MAP3K5, DUSP4, RAP1B, IFI44L, RAB5A, GDE1, MPC2, TAOK2, AREG, PLAT, KLHL22, SPRY2, SPRED2, SSX2IP, MAP3K14, SLC6A6, ACTR2, MDFIC, BARD1, ATG10, GR32, MAPK9, EPH32,RNF167, MAP4K4, LPGAT1, VWA5A,CEACAM1, ERRFI1, MAP3K13, DUSP7, GPNMB and RP1A.

[0208] Example 31: The method according to any one of the preceding examples, wherein first computational model is based on one or more genes selected from the group consisting of KRAS, RASSF1, ETV1, TP53, SIKE1, ANTXR2, DUSP6, MLPH, MAP3K5, DUSP4, RAP1B, IFI44L, RAB5A, GDE1, MPC2, TAOK2, AREG, PLAT, KLHL22 and SPRY2.

[0209] Example 32: The method according to any one of the preceding examples, wherein first computational model is based on one or more genes selected from the group consisting of KRAS, RASSF1, ETV1, TP53, SIKE1, ANTXR2, DUSP6, MLPH, MAP3K5, DUSP4, RAP1B, IFI44L, RAB5A, GDE1 and KLHL22.

[0210] Example 33: The method according to any one of the preceding examples, wherein the first computational model is based on one or more genes selected from the group consisting of TP53, SIKE1, DUSP4, MLPH, MAP3K5, DUSP6, ETV1, ANTXR2, RASSF1, KRAS, KLHL22, RAP1B, and RAB5A.

[0211] Example 34: The method according to any one of the preceding examples, wherein the first computational model is based on one or more genes selected from the group consisting of TP53, SIKE1, DUSP4, MLPH, MAP3K5, DUSP6, ETV1, ANTXR2, RASSF1, and KRAS.

[0212] Example 35: The method according to any one of the preceding examples, wherein one or more genes selected from the group consisting of KRAS, RASSF1, ETV1, TP53, SIKE1, ANTXR2, DUSP6, MLPH, MAP3K5, DUSP4, RAP1B, IFI44L, RAB5A, GDE1, MPC2, TAOK2, AREG, PLAT, KLHL22, SPRY2, SPRED2, SSX2IP, MAP3K14, SLC6A6, ACTR2, MDFIC, BARD1, ATG10, GR32, MAPK9, EPH32, RNF167,MAP4K4, LPGAT1, VWA5A,CEACAM1, ERRFI1, MAP3K13, DUSP7, GPNMB and RP1A are used at least in the first computational model.

[0213] Example 36: The method according to any one of the preceding examples, wherein one or more genes selected from the group consisting of TP53, SIKE1, DUSP4, MLPH, MAP3K5, DUSP6, ETV1, ANTXR2, RASSF1, and KRAS and / or from the group consisting of TP53, SIKE1, DUSP4, MLPH, MAP3K5, DUSP6, ETV1, ANTXR2, RASSF1, KRAS, KLHL22, RAP1B, and RAB5A are used at least in the first computational model.

[0214] Example 37: The method according to any one of the preceding examples, wherein the plurality of genes involved in one or more biological signaling pathways at least partly related to KRAS comprises TP53, SIKE1, DUSP4, MLPH, MAP3K5, DUSP6, ETV1, ANTXR2, RASSF1, KRAS, and optionally further KLHL22, RAP1B, and RAB5A.

[0215] Example 38: The method according to any one of the preceding examples, wherein the plurality of genes involved in one or more biological signaling pathways at least partly related to KRAS comprises KRAS, RASSF1, ETV1, TP53, SIKE1, ANTXR2, DUSP6, MLPH, MAP3K5, DUSP4, RAP1B, IFI44L, RAB5A, GDE1, MPC2, TAOK2, AREG, PLAT, KLHL22, SPRY2, SPRED2, SSX2IP, MAP3K14, SLC6A6, ACTR2, MDFIC, BARD1, ATG10, GR32, MAPK9, EPH32, RNF167,MAP4K4, LPGAT1, VWA5A,CEACAM1, ERRFI1, MAP3K13, DUSP7, GPNMB and RP1 A.

[0216] Example 39: A computer program, which when executed on a computing device, instructs the computing device to perform steps of the method of any one of the preceding examples.

[0217] Example 40: A non-transitory computer-readable medium storing a computer program according to the preceding example.

[0218] Example 41: A computing device with one or more processors, the computing device being configured to perform steps of the method according to any one of Examples 1 to 38.

[0219] Example 42: Use of RNA sequencing data for determining a tumor response to a KRAS therapeutic agent, preferably using the method according to any one of examples 1 to 38.

[0220] Example 43: Use of RNA sequencing data for determining an efficacy of a KRAS therapeutic agent against a tumor, preferably using the method according to any one of examples 1 to 38.

[0221] Example 44: Use of RNA sequencing data for determining, predicting and / or classifying a time on a KRAS therapeutic agent treatment, preferably using the method according to any one of examples 1 to 38. For example, RNA sequencing data may be used for classifying whether a patient treated with a KRAS therapeutic agent will (likely) be on therapy for a shorter or a longer duration, preferably using the method according to any one of examples 1 to 38. This may be indicative for a respective time on disease or time to disease progression.

[0222] Example 45: Use of RNA sequencing data for stratifying patients based on their predicted response to a KRAS therapeutic agent, preferably using the method according to any one of examples 1 to 38.

[0223] Example 46: Use of RNA sequencing data for selecting and / or identifying patients likely to respond to a KRAS therapeutic agent treatment, preferably to a targeted KRAS therapy or inhibition, preferably using the method according to any one of examples 1 to 38.

[0224] Example 47: Use of RNA sequencing data for monitoring tumor response during treatment with a KRAS therapeutic agent, preferably using the method according to any one of examples 1 to 38.

[0225] Example 48: Use of RNA sequencing data according to any one of examples 42 to 47, using the method according to any one of examples 1 to 38, wherein the first computational model is based on one or more genes selected from the group consisting of TP53, SIKE1, DUSP4, MLPH, MAP3K5, DUSP6, ETV1, ANTXR2, RASSF1, KRAS, KLHL22, RAP1B, and RAB5A, preferably selected from the group consisting of TP53, SIKE1, DUSP4, MLPH, MAP3K5, DUSP6, ETV1, ANTXR2, RASSF1, and KRAS.

[0226] Example 49: Use of RNA sequencing data according to any one of examples 42 to 47, using the method according to any one of examples 1 to 38, wherein at least the first computational model is based on one or more genes selected from the group consisting of TP53, SIKE1, DUSP4, MLPH, MAP3K5, DUSP6, ETV1, ANTXR2, RASSF1, KRAS, KLHL22, RAP1B, and RAB5A, preferably selected from the group consisting of TP53, SIKE1, DUSP4, MLPH, MAP3K5, DUSP6, ETV1, ANTXR2, RASSF1, and KRAS.

[0227] Example 50: Use of RNA sequencing data according to any one of examples 42 to 47, using the method according to any one of examples 1 to 38, wherein the plurality of genes involved in one or more biological signaling pathways at least partly related to KRAS comprises TP53, SIKE1, DUSP4, MLPH, MAP3K5, DUSP6, ETV1, ANTXR2, RASSF1, KRAS, KLHL22, RAP1B, and RAB5A.

[0228] Example 51: A method for determining, predicting and / or classifying a time on a KRAS therapeutic agent treatment, comprising the method according to any one of examples 1 to 38.

[0229] Example 52: A method for stratifying patients based on their predicted response to a KRAS therapeutic agent, comprising the method according to any one of examples 1 to 38.

[0230] Example 53: A method for selecting and / or identifying patients likely to respond to a KRAS therapeutic agent treatment, preferably to a targeted KRAS therapy or inhibition, comprising the method according to any one of examples 1 to 38.

[0231] Example 54: A method for monitoring tumor response during treatment with a KRAS therapeutic agent, comprising the method according to any one of examples 1 to 38.

[0232] FIG. 1 shows a computing device or system 100 configured to determine a tumor response to a KRAS therapeutic agent and / or to KRAS therapy according to an exemplary embodiment.

[0233] The computing device 100 comprises a processing circuitry 110 with one or more processors 112 for data processing.

[0234] The computing device 100 further comprises a first computational model 114a configured to model and / or simulate at least one of dependency of the tumor on the KRAS signaling pathway, and sensitivity of the tumor to KRAS inhibitor therapy. The computing device 100 further comprises a second computational model 114b configured to model and / or simulate activation of one or more biological signaling pathways at least partly related to KRAS. The second computational model 114b is also referred to herein as additional computational model 114b. Optionally, the computing device 100 can comprise one or more further or additional computational models 114c configured to model and / or simulate at least one further tumor parameter, as described in more detail hereinabove and hereinbelow.

[0235] The first computational model 114a is configured to compute and / or determine, based on evaluating and / or analyzing RNA sequencing data associated with tumor tissue, one or more dependency scores indicative of a degree of dependence of the tumor on the KRAS signaling pathway. Alternatively or additionally, first computational model 114a is configured to compute and / or determine, based on evaluating and / or analyzing RNA sequencing data associated with tumor tissue, one or more sensitivity scores indicative of a degree of sensitivity of the tumor to KRAS therapy and / or against the KRAS therapeutic agent.

[0236] The at least one additional computational model 114b and / or the second computational model 114b is configured to compute and / or determine, based on evaluating and / or analyzing RNA sequencing data associated with tumor tissue, one or more activation state scores indicative of a level or degree of activity of one or more biological signaling pathways at least partly related to KRAS.

[0237] The optional one or more further or additional computational models 114c can be configured to determine one or more values or scores of one or more further tumor intrinsic or tumor extrinsic parameters, such as for example pathway activation state scores for one or more cell or tumor intrinsic signaling pathways and / or pathway activation state scores for one or more cell or tumor extrinsic signaling pathways. Alternatively or additionally, one or more of cellularcomposition of the tumor, cellular states of one or more tumor cells biological activities of one or more tumor cells, values indicative of molecular processes, genetic variants, synthetic lethal partners, metabolic signatures, modifications, and protein-interaction networks may be considered in one or more further computational models 114c. Alternatively or additionally, one or more further or additional computational models 114c may be configured to evaluate the RNA sequencing data to model at least one further tumor parameter and / or to determine at least one further score or value for the at least one further tumor parameter. The at least one further tumor parameter can, for example, be indicative of one or more of a microenvironment of the tumor, RNA sequencing metadata, demographic data of the patient, angiogenesis of the tumor, activity of the VEGF signaling pathway, gene variants, mutations in the KRAS protein or proteins interacting with KRAS that confer resistance to KRAS therapeutic agents, and cellular composition of the tumor.

[0238] Optionally, the computing device 100 can comprise a classifier 115 or classifier circuitry 115 configured to receive one or more outputs of one of, a plurality of or all of the first computational model 114a, the second or the at least one additional computational model 114b and the one or more further computational models 114c. For instance, one or more dependency scores, one or more sensitivity scores, one or more activation state scores, and / or one or more scores or values of one or more further tumor intrinsic or extrinsic parameters may be received as inputs by the classifier 115, and may be processed and / or analyzed by the classifier 115 to determine whether or not the tumor is responsive to the KRAS therapeutic agent and / or the KRAS therapy. Therein, the classifier 115 may compute a classification result and / or probability for a response of the tumor to the KRAS therapy and / or the KRAS therapeutic agent.

[0239] The computing device 100 further comprises at least one data storage 116 and / or memory 116 for storing data and / or software instructions, for example in the form of a computer program for instructing the computing device 100 to carry out steps of the method of determining a tumor response to a KRAS therapeutic agent, as described in more detail hereinabove and hereinbelow. The data storage 116 may also be configured to store and / or may store RNA sequencing data that can be processed, analyzed and / or evaluated by the first and second computational models 114a, b, and optionally by one or more further computational models 114c.

[0240] The computing device 100 shown in FIG. 1 further comprises a user interface 118 for controlling the computing system 100 by the user and / or for outputting data and / or information, such as the e.g., one or more dependency scores, sensitivity scores, activation state scores, intermediate pathway activation state scores, parts of the RNA seq data, a classification result as to whether the tumor or patient is responsive or non-responsive to the KRAS therapeutic agent, and the like.

[0241] Further, the computing device 100 comprises a communication interface 120 for communicatively and / or operatively coupling the computing device 100 to an external computing device 500, and / or for coupling the computing system 100 to one or more external data sources 500 or external data storages 500, for example to receive or obtain RNA sequencing data therefrom and / or to transmit data thereto, such as e.g., one or more of the aforementioned scores and / or a classification result.

[0242] The computing device 100 is in particular configured to perform steps of a method of shows a flow chart illustrating steps of a method of determining a tumor response to a KRAS therapeutic agent and / or an efficacy of a KRAS therapeutic agent against a tumor, for example as described with reference to FIG. 2.

[0243] FIG. 2 shows a flow chart illustrating steps of a computer-implemented method of determining a tumor response to a KRAS therapeutic agent and / or an efficacy of a KRAS therapeutic agent against a tumor according to an exemplary embodiment.

[0244] The method comprises a step SI of providing, at a computing device 100 including one or more processors 112 for data processing, RNA sequencing data associated with tumor tissue, the RNA sequencing data being indicative of gene expression data of a plurality of genes involved in one or more biological signaling pathways at least partly related to KRAS. Optionally, step SI may comprise one or more of receiving the RNA seq data, e.g. from a data storage 116 of the computing device 100 and / or from one or more external data sources 500. Alternatively or additionally, step SI may comprise acquiring the RNA seq data based on or using one or more RNA sequencing modalities, such as e.g. total-RNA sequencing, polyA enriched sequencing, and RNA-Exome- based sequencing. Optionally, step SI may include one or more pre-processing steps, such as alignment and / or normalization of the RNA sequencing data.

[0245] RNA sequencing data may be derived from raw RNA sequencing data, which may be provided in either raw data (FASTQ format), as aligned sequences (BAM), or as processed gene expression tables. Therein, one or a plurality of different formats may be used. Depending on the format of the raw sequencing data, different processing steps may optionally be applied in step SI. For instance, if the raw sequencing data is in FASTQ format, then alignment and / or mapping to a reference genome may be applied. Alternatively or additionally quantification, e.g. with STAR, may be applied. If the BAM format is chosen, quantification may be applied, e.g. with STAR. Alternatively or additionally, featureCounts may be applied. In all cases, including if expression tables are given as raw sequencing data, normalization of expression values may be applied as described herein.

[0246] In step S2, the method comprises evaluating the RNA sequencing data with a first computational model 114a modelling at least one of dependency of the tumor on the KRAS signaling pathway, and sensitivity of the tumor to KRAS inhibitor therapy. Optionally, one or more dependency scores and / or one or more sensitivity scores may be computed by the first computational model 114a in step S2.

[0247] In step S3, the method further encompasses evaluating the RNA sequencing data with at least one additional computational model 114b and / or a second computational model 114b modelling activation of the one or more biological signaling pathways at least partly related to KRAS. Optionally, one or more activation state scores may be computed in step S3.

[0248] The method further comprises a step S4 of classifying, based on the evaluation of the RNA sequencing data with the first computational model 114a and the second computational model 114b (and / or the at least one additional computational model 114b), e.g. by means of the classifier 115, the tumor as being responsive or non-responsive to the KRAS therapeutic agent. Optionally, step S4 may comprise computing or generating a classification result as to whether or not the tumor is responsive to the KRAS therapeutic agent. Such classification result may further optionally be output at the user interface 118 of the computing device 100, may be stored at the data storage 116 and / or may be transmitted via the communication interface 120 to one or more external data sources 500 and / or external computing devices 500.

[0249] It should be noted that the method according to FIG. 2 may include one or more additional, supplemental and / or further optional steps. In particular, one or more further tumorparameters and / or one or more further scores or values for one or more further tumor biology parameters may be computed based on or using one or more further computational models 114c.

[0250] FIG. 3 shows a flow chart or block diagram illustrating aspects of model development for a computing device 100 and / or for a computer- implemented method of determining a tumor response to a KRAS therapeutic agent and / or an efficacy of a KRAS therapeutic agent against a tumor. Specifically, FIG. 3 illustrates a KRAS biology-informed multi-module classifier framework and / or prediction pipeline.

[0251] Block S10 relates to data generation, wherein raw RNA sequencing data may be generated and / or acquired. The raw RNA sequencing data may be acquired from one or more tumor cells, such as e.g. from cancer models and / or tumor samples using next generation sequencing. Accordingly, at block S10 tumor and preclinical cancer models RNA sequencing can be performed and corresponding RNA sequencing data can be generated.

[0252] For initial model testing and development, about 2000 quality controlled high quality RNA sequencing patient biopsy samples across multiple tissues were utilized, including lung and colorectal tumor samples. For preclinical data, about 1200 cell lines in CCLE from the BROAD Institute’s DepMap Dataset were used, which is a richly characterized cell line repository with CRISPR-derived dependency scoring and expression data. Additional preclinical data was sourced from publicly available sources wherein RNA sequencing was performed on the cell lines, xenografts, etc prior to treatment with KRAS inhibitors such as Sotorasib, Adagrasib, and RMC-6291. As clinical data, Real World Patient Data obtained from Tempus Labs, including data for about 60 Sotorasib, a KRAS G12C inhibitor, treated biopsies, with annotated clinical metadata for pre- and post- treated Sotorasib (KRAS G12C inhibitor) biopsy specimens were utilized.

[0253] Block S12 represents pre-processing for quality control, wherein the raw RNA seq data may be processed for normalized counts. For instance, the raw RNA sequencing data may be normalized based on one or more of a counts per million normalization, a fragments per kilobase million normalization, a transcripts per million normalization, an upper quartile normalization, a normalization with respect to counts adjusted with upper quartile factors, a normalization with respect to trimmed mean of M- values, and a normalization with respect to counts adjusted with trimmed mean of M- values factors.

[0254] Alternatively or additionally, RNA sequencing reads may be aligned at block S12. Alternatively or additionally, gene variants may be identified and / or annotated. Alternatively or additionally, gene expression of one or more genes may be quantified at block S12. Optionally one or more of alignment, mapping to a reference genome, quantification, and normalization may be applied, as described in detail e.g. with reference to Figure 2 and in the summary part.

[0255] At block S14, development of the actual computational models may be initiated or started. Block S14 includes feature or gene filtering and batch normalization based on or using the pre-processed and / or normalized RNA sequencing data. Therein, the normalized RNA seq data may be adjusted and normalized for batch effects, and potential features or feature candidates of one or more computational models 114a-c, which feature candidates may at least partly correspond to one or more genes, may be annotated and assessed for robust and dynamic expression. Accordingly, at block S14, gene expression normalization, batch effect adjustment and / or gene filtering by stable annotation and robust expression may be performed.

[0256] Block S16 relates to feature selection, wherein one or more model features may be selected. Therein, model features, e.g. biologic features or genes of one or more computational models 114a-c, can be selected through manual curation and / or based on identifying synthetic lethal partners for one or more targeted genes or gene sets, such as KRAS-specific gene sets. Therein, synthetic lethality (SL) refers to a genetic interaction in which the simultaneous perturbation of two genes leads to cell or organism death, whereas viability is maintained when only one of the pair is altered. Accordingly, at block SI 6, KRAS biologic features and / or KRAS- specific synthetic lethal gene pairs can be selected.

[0257] As non-limiting example, the following gene sets may be selected for developing the one or more computational models 114a-c and / or the overall multi-axis model. KEGG MAPK, KEGG RAS, KEGG ErbB, GSEA KRAS DN, GSEA_KRAS_UP, and GSEA PI3K may be used for tumor-intrinsic pathway gene sets. Alternatively or additionally, one or more of GSEA G2M, GSEA E2F, GSEA WNT, GSEA IMMUNE, GSEA EMT, GSEA ANGIO, GSEA APOPTOSIS, GSEA GLYCOLYSIS may be used.

[0258] At block S18 one or more computational or biologic models 114a-c may be generated, created or developed. Therein, statistical and machine learning methods may be utilized, e.g. after model fitting and testing including KRAS-specific dependency model and assigning KRAS-focused pathway activation state scores. Specifically, one or more dependency models orthe first computational models 114a may be created or generated, cross-validated and / or evaluated with respect to model performance. This may include one or more of feature reduction and selection, regularization and regression.

[0259] In an example, a plurality of first computational models 114a or dependency models may be generated or created based on machine learning, wherein different dependency models may differ in the number and / or type of genes or gene sets considered for determining and / or computing the dependency score. In a non-limiting example, the first computational models 114a can be based on one or more of the following genes, which are also referred to as features of the respective model 114a, and which are listed according to their relevance and / or importance for modelling dependency. KRAS, RASSF1, ETV1, TP53, SIKE1, ANTXR2, DUSP6, MLPH, MAP3K5, DUSP4, RAP1B, IFI44L, RAB5A, GDE1, MPC2, TAOK2, AREG, PLAT, KLHL22, SPRY2, SPRED2, SSX2IP, MAP3K14, SLC6A6, ACTR2, MDFIC, BARD1, ATG10, GR32, MAPK9, EPH32, RNF167, MAP4K4, LPGAT1, VWA5A, CEACAM1, ERRFI1, MAP3K13, DUSP7, GPNMB, RP1 A. From these genes or features, a subset of genes may be selected that should preferably be stable across multiple iterations of cross-validation. However, the aforementioned 41 genes may generally be used for the first computational model 114a modelling dependency on the KRAS signaling pathway and / or sensitivity to KRAS therapy.

[0260] The aforementioned 41 genes or model features are also shown in FIG. 4 as heatmap. Therein, the genes and / or features of the first computational model 114a are listed on the lefthand side of FIG. 4 and are ranked over the course of multiple iterations of cross-validation of the first computational model 114a. The color delineates the weight or importance of each gene, wherein a darker color indicates higher weight or importance. In order to assess the performance of the model and stability of the features in a correct way on the same dataset, cross-validation can be applied, where data is split into a training set (used to train the model) and a validation set (to assess the performance). In the heat map of FIG. 4, the top most frequently selected genes are shown in 30 iterations of cross-validation where it is demonstrated that some genes may be selected regardless of how the data is split, whereas some other genes can sometimes be replaced by others. The first or top ten genes in FIG. 4 may provide good or satisfying overall performance. This is reflected by the top ten genes of the heat map harboring a darker color indicating a higher weight across multiple iterations. In a similar way, using cross-validation, the performance metrics of the first computational model 114a can be measured. While many genescan be tested, only the most stable genes or genes that are stable across multiple iterations of cross-validation genes may be used in the first computational model 114a, which may e.g. be the case for the top ten or first ten genes in FIG. 4 from top to bottom. As such, the remaining genes may not be informative for the final model.

[0261] The following table compares model performance as area under the receiver operating characteristic curve (ROC AUC) of two exemplary dependency or first computational models 114a that are based on different numbers and sets of the aforementioned 41 genes.

[0262] Therein, model 1 includes 13 genes, namely TP53, SIKE1, KLHL22, RAP IB, RAB5A, MAP3K5, DUSP4, MLPH, DUSP6, ETV1, ANTXR2, RASSF1, and KRAS. Model 2 includes 10 (ten) genes, namely TP53, SIKE1, DUSP4, MLPH, MAP3K5, DUSP6, ETV1, ANTXR2, RASSF1, and KRAS.

[0263] Accordingly, in some embodiments the first computational model 114a, is based on at least 1, at least 2, at least 3, at least 4, at least 5, at least 6, at least 7, at least 9, at least 10, at least 11, at least 12, at least 13, at least 14, at least 15 genes selected from the group consisting of KRAS, RASSF1, ETV1, TP53, SIKE1, ANTXR2, DUSP6, MLPH, MAP3K5, DUSP4, RAP1B, IFI44L, RAB5A, GDE1, MPC2, TAOK2, AREG, PLAT, KLHL22, SPRY2, SPRED2, SSX2IP, MAP3K14, SLC6A6, ACTR2, MDFIC, BARD1, ATG10, GR32, MAPK9, EPH32, RNF167,MAP4K4, LPGAT1, VWA5A,CEACAM1, ERRFI1, MAP3K13, DUSP7, GPNMB, RP1 A. In some embodiments the first computational model 114a, is based on at least 1, at least 2, at least 3, at least 4, at least 5, at least 6, at least 7, at least 9, at least 10, at least 11, at least 12, at least 13, at least 14 genes, at least 15 genes selected from the group consisting of KRAS, RASSF1, ETV1, TP53, SIKE1, ANTXR2, DUSP6, MLPH, MAP3K5, DUSP4, RAP1B, IFI44L, RAB5A, GDE1, MPC2, TAOK2, AREG, PLAT, KLHL22, SPRY2. In some embodiments the first computational model 114a, is based on at least 1, at least 2, at least 3, at least 4, at least 5, at least 6, at least 7, at least 9, at least 10, at least 11, at least 12, at least 13, genes selected from the group consisting of TP53, SIKE1, KLHL22, RAP1B, RAB5A, MAP3K5, DUSP4, MLPH, DUSP6, ETV1, ANTXR2, RASSF1. In other embodiments the first computational model 114a,is based on at least 1, at least 2, at least 3, at least 4, at least 5, at least 6, at least 7, at least 9, at least 10 genes selected from the group consisting of TP53, SIKE1, DUSP4, MLPH, MAP3K5, DUSP6, ETV1, ANTXR2, RASSF1, and KRAS. In a specific embodiment the first computational model 114a is based on TP53, SIKE1, KLHL22, RAP1B, RAB5A, MAP3K5, DUSP4, MLPH, DUSP6, ETV1, ANTXR2, RASSF1. In another embodiment, the first computational model 114a is based on TP53, SIKE1, KLHL22, RAP1B, RAB5A, MAP3K5, DUSP4, MLPH, DUSP6, ETV1, ANTXR2, RASSF1.

[0264] In other embodiments the first computational model 114a, is based on at least 1, at least 2, at least 3, at least 4, at least 5, at least 6, at least 7, at least 9, at least 10 genes selected from the group consisting of KRAS, ETV1, ANTXR2, GDE1, MPC2, TAOK2, AREG, PLAT, KLHL22, SPRY2. In a specific embodiment the first computational model 114a is based on KRAS, ETV1, ANTXR2, GDE1, MPC2, TAOK2, AREG, PLAT, KLHL22, SPRY2. In other embodiments the first computational model 114a, is based on at least 1, at least 2, at least 3, at least 4, at least 5, at least 6, at least 7, at least 9, at least 10 genes selected from the group consisting of KRAS, MAP3K5, TAOK2, AREG, PLAT, KLHL22, SPRY2, SPRED2, SSX2IP, RP1 A. In a specific embodiment the first computational model 114a is based on KRAS, MAP3K5, TAOK2, AREG, PLAT, KLHL22, SPRY2, SPRED2, SSX2IP, RP1A. In some embodiments the first computational model 114a, is based on at least 1, at least 2, at least 3, at least 4, at least 5, at least 6, at least 7, at least 9, at least 10, at least 11, at least 12, at least 13, genes selected from the group consisting of KRAS, DUSP6, MLPH, AREG, PLAT, KLHL22, SPRY2, MAP3K14, SLC6A6, ACTR2, MDFIC, MAPK9, EPH32.

[0265] Some embodiments refer to a computational model based on at least 1, at least 2, at least 3, at least 4, at least 5, at least 6, at least 7, at least 9, at least 10, at least 11, at least 12, at least 13, at least 14, at least 15 genes selected from the group consisting of KRAS, RASSF1, ETV1, TP53, SIKE1, ANTXR2, DUSP6, MLPH, MAP3K5, DUSP4, RAP1B, IFI44L, RAB5A, GDE1, MPC2, TAOK2, AREG, PLAT, KLHL22, SPRY2, SPRED2, SSX2IP, MAP3K14, SLC6A6, ACTR2, MDFIC, BARD1, ATG10, GI 2, MAPK9, EPH32, RNF167,MAP4K4, LPGAT1, VWA5A,CEACAM1, ERRFI1, MAP3K13, DUSP7, GPNMB, RP1A. Some embodiments refer to a computational model based on at least 1, at least 2, at least 3, at least 4, at least 5, at least 6, at least 7, at least 9, at least 10, at least 11, at least 12, at least 13, at least 14 genes, at least 15 genes selected from the group consisting of KRAS, RASSF1, ETV1, TP53,SIKE1, ANTXR2, DUSP6, MLPH, MAP3K5, DUSP4, RAP1B, IFI44L, RAB5A, GDE1, MPC2, TA0K2, AREG, PLAT, KLHL22, SPRY2. Other embodiments refer to a computational model based on at least 1, at least 2, at least 3, at least 4, at least 5, at least 6, at least 7, at least 9, at least 10, at least 11, at least 12, at least 13, genes selected from the group consisting of TP53, SIKE1, KLHL22, RAP1B, RAB5A, MAP3K5, DUSP4, MLPH, DUSP6, ETV1, ANTXR2, RASSF1. Further embodiments refer to a computational model based on at least 1, at least 2, at least 3, at least 4, at least 5, at least 6, at least 7, at least 9, at least 10 genes selected from the group consisting of TP53, SIKE1, DUSP4, MLPH, MAP3K5, DUSP6, ETV1, ANTXR2, RASSF1, and KRAS. Some embodiments refer to a computational model based on TP53, SIKE1, KLHL22, RAP1B, RAB5A, MAP3K5, DUSP4, MLPH, DUSP6, ETV1, ANTXR2, RASSF1. Other embodiments refer to a computational model based on TP53, SIKE1, KLHL22, RAP1B, RAB5A, MAP3K5, DUSP4, MLPH, DUSP6, ETV1, ANTXR2, RASSF1.

[0266] Some embodiments refer to a computational model based on at least 1, at least 2, at least 3, at least 4, at least 5, at least 6, at least 7, at least 9, at least 10 genes selected from the group consisting of KRAS, ETV1, ANTXR2, GDE1, MPC2, TAOK2, AREG, PLAT, KLHL22, SPRY2. Some embodiments refer to a computational model based on KRAS, ETV1, ANTXR2, GDE1, MPC2, TAOK2, AREG, PLAT, KLHL22, SPRY2. Other embodiments refer to a computational model based on at least 1, at least 2, at least 3, at least 4, at least 5, at least 6, at least 7, at least 9, at least 10 genes selected from the group consisting of KRAS, MAP3K5, TAOK2, AREG, PLAT, KLHL22, SPRY2, SPRED2, SSX2IP, RP1A. Further embodiments refer to a computational model based on KRAS, MAP3K5, TAOK2, AREG, PLAT, KLHL22, SPRY2, SPRED2, SSX2IP, RP1 A. Some embodiments refer to a computational model based on at least 1, at least 2, at least 3, at least 4, at least 5, at least 6, at least 7, at least 9, at least 10, at least 11, at least 12, at least 13, genes selected from the group consisting of KRAS, DUSP6, MLPH, AREG, PLAT, KLHL22, SPRY2, MAP3K14, SLC6A6, ACTR2, MDFIC, MAPK9, EPH32.

[0267] With continued reference to FIG. 3, an activity model or the second computational model 114b (which may be referred to herein as at least one additional computational model 114b) may be generated at block S16, which can include the determination of one or more activation state scores or pathway activation state scores for one or more biological or biological signaling pathways. The second computational model 114b is also referred to herein asadditional computational model 114b. Optionally, one or more further computational models 114c may be generated, such as e.g. a computational model 114c modelling one or more tumor extrinsic parameters, such as one or more of a microenvironment of the tumor, angiogenesis of the tumor, activity of the VEGF signaling pathway, and cellular composition of the tumor, or others as described.

[0268] Blocks S14, S16, and S18 may refer to phase 1 development of multiple biologic or computational models 114a-c.

[0269] Block S20 can be considered as phase 2 of the development, wherein one or more of the created computational models 114a-c may be combined or incorporated in a classifier 115 to determine a response probability and / or classification result, as indicated by the bold arrow in FIG. 3

[0270] As schematically illustrated by block S22, one or more foundation models may be incorporated at each of steps or blocks S16, S18, and S20, to aid or augment feature extraction, biologic module creation, and response classification.

[0271] In the following, various aspects, advantages and use cases of the present disclosure are exemplarily summarized. The present disclosure at least partly relates to an RNA-based biomarker designed to stratify responses including assessment of clinical benefit to KRAS therapeutic agents, such as e.g. G12C inhibitors in non-small cell lung cancer (NSCLC) patients using RNA sequencing. Therein, high-dimensional and multimodal RNA seq data can be utilized to analyze KRAS biology, tailoring patient stratification to their unique biological characteristics. At its core, two key KRAS biological aspects, dependency and activation, are combined to identify those most likely to derive clinical benefit using a multi-axis model. The overall modelling and approach was, inter aha, validated in predicting cell culture responses to KRAS inhibitors and in a real world dataset of non-small cell lung cancer patients treated with the KRAS G12C-inhibitors Sotorasib.

[0272] The present disclosure and computer-implemented method particularly address the challenge of accurately predicting those patients which may receive clinical benefit from KRAS therapy, such as e.g. KRAS G12C inhibitors in NSCLC patients, a task where existing biomarker solutions are limited. The method and approach described herein overcomes many of these limitations and provides the potential to impact healthcare in a variety of areas, includingimproved guidance for clinical trials, optimization of therapeutic outcomes, rationalization of effective combination strategies, and acceleration of approvals in various medical contexts.

[0273] Aspects of the present disclosure, the computer-implemented method and corresponding computing device described herein relate to an RNA-based sequencing classifier that can predict responses and / or clinical benefit to KRAS therapeutic agent or KRAS therapy, e.g. KRAS inhibitors, using a single RNA assay. It models KRAS biology, focusing on dependency and activation, to create a multi-axis or multi-dimensional model comprising at least the first and second computational models 114a, and 114b. The multi-axis model can define a phenotypic vector space for response prediction, and can allow to identify patients most likely to respond to targeted KRAS therapies effectively.

[0274] A major advantage of the method and computing device disclosed herein may be provided by the RNA-based computational models 114a-c that model core KRAS biology to predict KRAS therapeutic response. Therein, significant advantages can be provided over conventional DNA-based biomarkers, which are typically limited to assessing mutational states and fail to capture the complexity of diseases like cancer. Unlike DNA biomarkers, the method and computing device disclosed herein leverages RNA transcriptomic profiling to provide a quantitative analysis of cellular states, which may be relevant for understanding the biology of KRAS mutated tumors. This RNA-based approach, enhanced by machine learning and statistical methods, can allow for a more precise identification of biological features, including RNA expression and genetic variants, predictive of treatment response. The method and computing device disclosed herein focuses on fundamental KRAS biology, including KRAS dependency and activation, enabling it to capture both normal and aberrant KRAS biology across various cellular contexts, tumor types, and mutational states. This comprehensive modeling is applicable across multiple treatments, not just a single agent, presenting a more versatile and effective tool for guiding the utilization of this class of therapies.

[0275] Also, the method and computing device disclosed herein can provide the potential for dynamic insights in clinical development and treatment decisions. Unlike current point-of-care or conventional DNA-based biomarkers that provide static information, the method and computing device disclosed herein can offer the potential for real-time guidance throughout a patient’s treatment journey. For example, the method and computing device disclosed herein may be capable of discerning KRAS dependency and activation, which can serve as an initial pre-treatment assessment for those patients' tumors most reliant on KRAS signaling for survival. In addition, the assessment of KRAS dependency and activation of related biochemical signaling pathways can also provide a metric for assessing on-treatment efficacy as the perturbations of these features could be used to compare with baseline assessments to understand which aspects of KRAS biology and related biologies are being impacted by therapy. Such information could enable early assessment of efficacy, generation of hypotheses for resistance mechanisms, and / or guidance for the selection of follow-on therapies. By incorporating multiple biological axes representative of activation (also referred to as activity of the KRAS signaling pathway) and dependency on the KRAS signaling pathway, the method and computing device disclosed herein can enable dynamic patient stratification for clinical trials and informed treatment decisions from diagnosis to relapse.

[0276] Further, the method and computing device disclosed herein can offer versatility in predicting efficacy across multiple KRAS-specific therapies, e.g., KRAS G12C-mutation specific therapies. Therein, the method and computing device disclosed herein can stand out for its versatility in predicting treatment efficacy across an entire class of KRAS G12C-mutation specific therapies, rather than being limited to a single G12C agent. This broad applicability stems from its comprehensive analysis of KRAS biology, which captures a wide spectrum of genetic and molecular interactions relevant to various KRAS-targeted treatments. This ability to generalize across different KRAS inhibitors makes it an invaluable tool for clinicians in selecting the most appropriate treatment strategy from a range of available KRAS-G12C targeted therapies.

[0277] The method and computing device disclosed herein can be considered a first- generation RNA-based biomarker designed to stratify responses to KRAS therapy or therapeutic agents, such as e.g. KRAS G12C inhibitors. It operates by utilizing machine learning algorithms tailored to an RNA-based feature set anchored on KRAS biology. The biomarker models two primary tumor biology classes of parameters, dependency and activation, and incorporates additional axes or computational models for enhanced stratification across tissue types and cellular contexts.

[0278] For KRAS dependency modelling based on or using the first computational model 114a, a feature set may be selected reflecting KRAS biology, for example including genes involved in Ras, MAPK, PI3K, and EGFR signaling, as well as processes like metabolicreprogramming and immune dysregulation. This set can be refined through feature selection methods, leading to a robust geneset represented across datasets. Dependency can, for example, be modeled using Comprehensive Cancer Cell Line (CCLE) data, employing regression models and feature stability assessments to identify reliable KRAS dependency models.

[0279] In modeling KRAS activation, pathway activity can be inferred from gene expression data. Therein, activity can be assessed by analyzing the expression of pathway-responsive genes, providing a functional view of pathways like MAPK, PI3K, and EGFR. This can allow the assignment of a KRAS activation axis and identification of additional activated pathway signatures. In particular, pathway signatures can be utilized and in some cases the actual genes contained within those pathways, as a feature of the method and computing device disclosed herein, which can predict clinical benefit of KRAS therapy or therapeutic agents, e.g., KRAS inhibitor drugs.

[0280] The robust feature set, defined by these methods, can construct dependency and activation axes and identify other significant axes, which can be combined into a multi-axis model predictive of clinical benefit to KRAS therapy or KRAS therapeutic agents, e.g. KRAS G12C inhibitor drugs. The overall model can be validated a) in cell lines treated with KRAS inhibitors to predict cytotoxicity (e.g., IC50, EC50) and b) in real-world data of cancer patients, e.g., NSCLC patients treated with Sotorasib, a KRAS G12C inhibitor, as will be shown hereinbelow.

[0281] In the following, aspects related to methodological processes surrounding the implementation and the construction or development of the method and computing device disclosed herein are summarized.

[0282] As also exemplary described with reference to block S10 of FIG. 3 high-throughput next generation sequencing methodologies may be utilized to acquire or generate RNA seq data, which includes RNA gene expression information originating from tumor tissue. The method and computing device disclosed herein can make use of conventional methodological approaches for RNA sequencing including total-RNA, polyA enriched, and RNA-Exome-based sequencing modalities. Importantly, the method can be extended to include data derived from diverse sources including fresh tissue, Formalin-Fixed Paraffin-Embedded (FFPE), cell culture, and potentially cell free sources.

[0283] Further, the method and computing device disclosed herein can use the raw nextgeneration sequencing RNA gene expression data that is initially pre-processed for normalized counts, as described, e.g., with reference to block S12 of FIG. 3. Pre-processed RNA sequencing data can be fed into the first and second computational models 114a, 114b and optional further models 114c yielding (i) a KRAS dependency score or metric and (ii) an activation score, such as e.g. an activation score for the MAPK / RAS signaling pathway, and (iii) optionally additional pathway activation state scores for one or more signaling pathways at least partly related to KRAS. In some instances, these continuous variable metrics serve as input for a hyperplane, for example using a Support Vector Machine to derive two biological axes that stratify response (e.g. KRAS dependency and MAPK / RAS activation). In other instances, these continuous variable metrics are used in a decision tree-style architecture to incorporate multiple metrics to stratify responders.

[0284] The overall multi-axis model can be validated using in vitro KRAS inhibitor cytotoxicity data and real-world data, e.g., from NSCLC patients treated with Sotorasib. This dual approach allows to assess the model's accuracy in predicting drug efficacy and patient responses. Additionally, model explanations can be provided to elucidate the mechanics of the best-performing model, including model weights and SHAP values, to enhance interpretability.

[0285] For the development of the single computational models, e.g. the first and second computational models 114a, 114b, and the overall multi-axis or multi-dimensional model, biology-informed and data-driven feature identification and selection techniques can be utilized, as e.g. described with reference to block S14 and S16 of FIG. 3. Specifically, both biologically- informed and unbiased-approaches can be utilized to identify gene sets to serve as features for downstream machine learning applications, such as the classifier 115. For biologically-informed gene sets, KRAS-centric biology can be modeled by extracting publicly available gene sets in Kyoto Encyclopedia of Genes and Genomes (KEGG) and Molecular Signatures Database (MSigDB). These gene sets capture MAPK, RAS, EGFR, PI3K, and KRAS differentially expressed genes. In addition, genes involved in additional core KRAS-associated hallmark biologies as identified in the Catalogue Of Somatic Mutations In Cancer (COSMIC) can be included. Together, this biologically-focused gene set can capture those genes involved in KRAS-specific biologies, which are central to tumor biology, along with extrinsic sets that capture tumor microenvironment factors and cellular composition. This framework is adaptable,allowing exploration of both comprehensive and tailored gene sets to meet the specific needs of biomarker discovery and tumor biology.

[0286] In addition, features can be selected at block S16 using unbiased-approaches by employing unsupervised machine learning techniques, differential gene expression analysis, pathway analysis, and identification of synthetic lethal gene pairs or partners, which can aid in identifying synthetic lethal gene pairs and refining the selection of relevant genetic features, especially in the context of KRAS.

[0287] For feature filtering, e.g. at block S14, and further feature selection at block S16, multiple RNA-seq normalization techniques, such as CPM, CTF, and TMM can be applied, as described with reference to block S12 of FIG. 3. These techniques address diverse aspects of RNA-seq data quality and alignment. The selection of which technique is chosen to best fit the needs of the downstream classifier 115 can be based on evaluating their impact on classifier performance. Further, filtering steps can be applied at block S14 to ensure robustness, wherein features with stable annotations can be filtered, low-expressers or gene features with less than about 100 read counts at 80th percentile ca be removed, and genes with sufficient variation can be retained, e.g. using MAD score. After these steps, a curated feature set can be obtained, creating a solid foundation for downstream model applications.

[0288] For developing the dependency model and / or the first computational model 114a at block S18, a rigorous model evaluation process may be applied to ensure the robustness and reliability of the biomarker development framework. Initially, the distribution of the target variable or dependency may be assessed, preferably confirming its suitability for regression analysis. Next, a range of regression models can be explored, such as linear regression, Multilayer Perceptron (MLP) regression, and XGBoost regression, to identify the most appropriate models. To further enhance model performance and prevent overfitting, the application of L2 regularization (ridge) and a combination of L2 and LI regularization (elastic net) can be investigated. To robustly evaluate the selected regression models and regularization strategies, repeated k-fold cross-validation can be employed, which can provide a comprehensive assessment of model performance. Therein, key performance metrics can be reported to ensure transparency and reliability.

[0289] Additionally, a thorough evaluation of feature stability and robustness can be conducted. Therein, feature redundancy can be assessed, thereby determining whether certainfeatures can be safely eliminated without compromising model performance. Further, the stability of feature selection can be tested by varying parameters such as the random seed or sample subsets, examining whether different selections result in varying sets of features.

[0290] For developing the activation model and / or the second computational model 114b at block S18 (which may be referred to herein as at least one additional computational model 114b), pathway activity can be inferred from gene expression data. In particular, at least the activity of the MAPK and / or RAS signaling pathway may be quantified and utilized by the second computational model 114b. Optionally, pathway activity of one or more further signaling pathways can be considered, such as e.g. Androgen Receptor (AR) Pathway, Estrogen Receptor (ER) Pathway, Hypoxia, JAK-STAT Signaling, MAPK Signaling, NFkB Pathway, Notch Signaling, PI3K / AKT Signaling, p53 Pathway, RTK Signaling, TGFb Signaling, TNFa Signaling, Trail Signaling, and WNT / p-Catenin Signaling. This can allow identifying the key pathways that may be active in different disease or mutational states, such as in the setting of KRAS mutations. One or more activation scores of one or more of the aforementioned signaling pathways can be used to extrapolate KRAS activation state by examining MAPK activity state, which may be considered a core KRAS signaling pathway.

[0291] The multi-axis or multi-dimensional model can then be constructed by combining at least the models for dependency and activation, i.e. the first and second computational models 114a, 114b, optionally along with other biological axes or one or more further computational models 114c. The multi-axis model's methodology can be tailored to the complexity of the underlying data source, for example using linear analysis with hyperplanes in less complex cell line data or structured analysis with decision trees in more complex tissue samples.

[0292] Based on the method and computing device disclosed herein, preclinical drug candidate selection can be conducted. For instance, in a research setting, the method can assist in selecting drug candidates for a clinical study by analyzing cell line data to predict responses to KRAS inhibitors or therapeutic agents. This can guide the nomination of the most effective drug candidates for clinical development.

[0293] Moreover, the method and computing device described herein can be used to stratify patients in clinical trials based on their predicted response to KRAS therapeutic agents, thereby enhancing trial efficiency and the likelihood of successful outcomes.

[0294] Further, the method and computing device described herein can be used to predict therapy outcome and / or therapeutic success of a KRAS therapeutic agent treatment. For example, the method and computing device described herein can be used to predict response to a KRAS therapeutic agent treatment based on tumor volume change. Thus, the method and computing device described herein can be used to quantitatively predict response to a KRAS therapeutic agent treatment. Moreover, using tumor volume change in response to KRAS therapeutic agent treatment as indicator for therapeutic success may be advantageous as tumor volume change represents a continuous, quantitative measure of treatment response that is correlated with therapeutic efficacy.

[0295] Apart from that, the method and computing device described herein can be used to stratify patients based on their predicted time on KRAS therapeutic agent treatment. For example, the method and computing device described herein can be used to predict response to a treatment with a KRAS therapeutic agent with an associated predicted response duration. In particular, the method and computing device described herein can be used to classify whether patients treated with a KRAS therapeutic agent will (likely) be on therapy for a shorter or a longer duration. This may be indicative for a respective time on disease or time to disease progression. Thus, patient stratification based on a predicted time to disease progression may be advantageous for improving therapeutic decisions and patient management.

[0296] Further, the method and computing device disclosed herein can be used as companion diagnostic in clinical settings for patient selection. For instance, in a clinical setting, the method and computing device described herein can serve as a companion diagnostic tool to select patients who are more likely to respond to targeted KRAS therapy or inhibition. By analyzing individual patient profiles, those patients with genetic and molecular characteristics indicating a higher likelihood of positive response to KRAS therapy or inhibitors can be identified, thus allowing to personalize and optimize treatment plans. As another example, the method and optionally the computing device disclosed herein can be used as a companion diagnostic as follows. For example, disclosed herein is also a KRAS therapeutic agent for use in a method of treating a tumor in a subject in need thereof, wherein the method comprises performing the method disclosed herein and, when the tumor is classified (based on the evaluation of the RNA sequencing data with the first computational model and the one or more additional computational models) as being responsive to the KRAS therapeutic agent, administering the KRAS therapeutic agent to the subject. For example, in case of a subject or patients suffering from or being suspected of suffering from a pancreatic ductaladenocarcinoma, the KRAS therapeutic agent may be a pan-Ras inhibitor such as RMC-6236. Thus, in case the tumor is a pancreatic ductal adenocarcinoma, the KRAS therapeutic agent may be a pan-Ras inhibitor such as RMC-6236. As a further example, in case the tumor is a non-small cell lung cancer (NSCLC) the KRAS therapeutic agent may be sotorasib (also referred to herein as Sotorasib). Thus, in case of a subject or patients suffering from or being suspected of suffering from a non-small cell lung cancer, the KRAS therapeutic agent may be sotorasib. Accordingly, the present disclosure encompasses for example i) sotorasib for use in a method of treating a NSCLC in a subject in need thereof, wherein the method comprises performing the method disclosed herein and, when the NSCLC is classified (based on the evaluation of the RNA sequencing data with the first computational model and the one or more additional computational models) as being responsive to sotorasib, administering sotorasib to the subject and ii) sotorasib for use in the treatment of NSCLC (in a subject), wherein the NSCLC is classified as being responsive to sotorasib, preferably using the method and optionally the computing device disclosed herein.

[0297] Apart from that, adaptive predictions for KRAS-targeted therapies development and administration can be conducted. In drug development and clinical practice, the method and computing device disclosed herein may be used at discrete points during a patient’s treatment journey to infer responses to KRAS-targeted therapies by assigning biological metrics, dependency and activation. This can inform drug developers in trial design and real-time response monitoring, while in clinical practice, it can guide therapy selection, and monitor posttreatment tumor response.

[0298] Moreover, predictions for identification of co-therapeutic targets can be improved. This use case focuses on identifying additional biological axes or further computational models 114c which may also be targetable in combination with KRAS therapeutics. In some instances, these axes or models 114c may help define novel therapies for combination therapy, while in other instances, they may help stratify response to combinational therapy in clinical trials or practice. In particular, the multi-axis model may be utilized to stratify patient or cell line responses to a variety of KRAS inhibitor therapies, wherein the multi-axis model can be adapted to assess the effectiveness of different KRAS inhibitors which can be either mutation-specific or -agnostic.

[0299] Also, treatment evolution and resistance development may be monitored. The method and computing device disclosed herein can be used to monitor the evolution of tumor response during treatment with KRAS therapeutic agents, such as inhibitors. By continually analyzingRNA expression profiles either through additional tissue sampling or in some instances through the integration with other diagnostic modalities (e.g. liquid diagnostics), changes in gene expression can be detected that may indicate developing resistance to the current treatment. This can allow for timely adjustments in therapeutic strategies, potentially switching to alternative KRAS inhibitors or combination therapies to overcome resistance.

[0300] Further, the method and computing device disclosed herein can assist in personalized dosage optimization for KRAS therapies, e.g. using KRAS inhibitors. By understanding the specific biological dynamics of a patient's tumor, the multi-axis model can help determine the most effective dosage that maximizes therapeutic efficacy while minimizing potential side effects. This may also serve purposes in preclinical studies to aid in selecting optimal dosing strategies for future clinical development and employment.

[0301] Moreover, the method and computing device disclosed herein can be applied for early detection and intervention in individuals at high risk of developing KRAS-mutant cancers. At present, KRAS therapies are administered as third line agents. Using the method and computing device disclosed herein, however, RNA profiles can be analyzed in high-risk populations, and early molecular changes suggestive of KRAS-driven tumorigenesis or those patients which might benefit from early intervention and administration of KRAS therapies can be identified.

[0302] Yet another use case of the method and computing device described herein relates to a research tool for understanding KRAS biology in diverse cancer types. Specifically, beyond its clinical applications, the method and computing device disclosed herein can serve as a valuable research tool, which can be used to deepen the understanding of KRAS biology across various cancer types, contributing to the broader knowledge base and potentially uncovering novel insights into cancer subtyping, pathogenesis and treatment.

[0303] Also, the method and computing device disclosed herein can be employed to tailor neoadjuvant (pre-surgical) and adjuvant (post-surgical) therapies in cancer treatment in essence by aligning with and integrating into current clinical guideline frameworks. By analyzing the tumor's RNA profile, the method and computing device disclosed herein can inform the selection of the most effective therapies before and after surgery, potentially improving patient outcomes.

[0304] FIG. 5 illustrates a plurality of computational models, named model 1 through 15, to determine a tumor response to a KRAS therapeutic agent and / or to KRAS therapy according to an exemplary embodiment. In particular, FIG. 5 shows 15 computational models, eachmodelling and / or being configured to model at least one tumor biology parameter and / or tumor parameter to determine tumor response. The 15 different models are listed on the y-axis according to their number and the weight associated with each of the 15 models is shown in arbitrary units on the x-axis of FIG. 5. Therein, the weights shown in Figure 5 may be regarded as model weights, which can correlate with or be indicative of a relative importance of the respective model. The weights may be used to factor the 15 models or at least a subset thereof into a classifier 115 for classifying the tumor as being responsive or non-responsive to the KRAS therapeutic agent and / or to KRAS therapy in general. Accordingly, the classifier 115 may be a composite, ensemble and / or arrangement of at least a subset or all of the computational models 1 to 15 illustrated in Figure 5.

[0305] FIGs. 6A, 6B and 6C each illustrate steps of a computer-implemented method of determining a tumor response to a KRAS therapeutic agent and / or to KRAS therapy according to exemplary embodiments. Alternatively or additionally, each of FIGs. 6A, 6B and 6C schematically illustrates an output of a classifier 115 composed of or comprising a plurality of computational models 114a-c. Such output, illustration or representation of a multi-model or multi-module classifier 115 is also referred to as multi-axis model 200, which comprises at least the first computational model 114a and the second computational model 114b or at least one additional computational model 114b.

[0306] The ensemble of the first computational model 114a and at least one additional computational model 114b, which may be referred to as second computational model 114b, and optionally one or more further computational models 114c as in the case of FIGs. 6A and 6C, may be understood as or referred to as multi-axis or multi-dimensional computational model 200 and / or multi-model or multi-module classifier 115 to determine tumor response to a KRAS therapy and / or to a KRAS therapeutic agent.

[0307] In the examples of FIGs. 6A, 6B and 6C, the first computational model 114a is a dependency model 114a configured to model dependency on the KRAS signaling pathway. Alternatively or additionally, the first computational model 114a can be a sensitivity model modelling responsiveness to KRAS therapy. The first computational model 114a defines a first model axis 202 of the multi-axis model 200. The second computational model 114b in the examples of FIGs. 6A, 6B and 6C is an activation model 114b, which is configured to model the activation of one or more biological pathways at least partly related to KRAS, and which may bereferred to herein as at least one additional computational model 114b. The second computational model 114b defines a second axis 204 of the multi-axis model 200.

[0308] Generally, the example of FIG. 6A is a schematic example of multiple module or model classifier 115, respectively a multi-axis model 200, wherein three modules or models 114a-c are used. In this example, one axis describes the scoring of an individual module or model 114a-c, such as the scoring of dependency or sensitivity on the x-axis and the scoring of the activation of the KRAS signaling or related biological pathway on the y-axis. Response to KRAS therapy and / or the KRAS therapeutic agent can be delineated according to a function of these scores. As further illustrated in FIG. 6A, one or more additional or further axes 206 may be defined by one or more further computational models 114c modeling one or more further tumor extrinsic or tumor-intrinsic parameters. For instance, one or more cell or tumor intrinsic signaling pathways, one or more cell or tumor extrinsic signaling pathways, cellular composition of the tumor, cellular states of one or more tumor cells, biological activities of one or more tumor cells, molecular processes, genetic variants, synthetic lethal partners, metabolic signatures, modifications, protein-interaction networks, a tumor microenvironment, angiogenesis of the tumor, and activity of the VEGF signaling pathway. Alternatively or additionally, activity of one or more signaling pathways may be modeled by a further computational model 114c as further or additional axis 206 of the multi-axis model 200. For instance, activity or activation of one or more of the following signaling pathways may be modeled: Androgen Receptor signaling pathway, Estrogen Receptor signaling pathway, Hypoxia signaling pathway, JAK-STAT signaling pathway, MAPK signaling pathway, NFkB signaling pathway, Notch signaling pathway, PI3K / AKT signaling pathway, p53 signaling pathway, RTK signaling pathway, TGFb signaling pathway, TNFa signaling pathway, Trail signaling pathway, and WNT / p-Catenin signaling pathway.

[0309] In the example of FIGs. 6A and 6B, the model axes 202, 204, 206 are shown as orthogonal axes, which however, is not necessary, in particular in higher-dimensional spaces with more than three axes, as shown in the example of FIG. 6C.

[0310] Generally, the multi-axis model 200 defines a multi-dimensional phenotypic space 212, also referred to herein as vector space 212 or phenotypic vector space for the phenotype of the tumor. The phenotypic vector space 212 includes at least a first region 220 associated with afirst phenotype of the tumor responsive to the KRAS therapeutic agent and a second region 230 associated with a second phenotype of the tumor non-responsive to KRAS therapeutic agent.

[0311] To determine whether or not the tumor is responsive to the KRAS therapeutic agent, a dependency (or sensitivity) score may be computed by means of the first computational model 114a, and an activation state score may be computed by the second computational model 114b and / or by at least one additional computational model 114b, and optionally one or more scores may be computed with one or more further computational models 114c. The determined scores can then be assigned or mapped to the multi-axis computational model 200 to determine whether or not the computed dependence, activity and optional other scores are located in the first region 220 or the second region 230. Alternatively or additionally, the determined scores can be evaluated with a classifier 115.

[0312] FIG. 6B exemplary illustrates the results for a plurality of lung tumor cell lines, wherein each dot in FIG. 6B represents the results for one lung tumor cell line. Therein, solid dots represent the wild type KRAS, circles represent G12C KRAS mutational status, and dashed dots represent other KRAS mutations. In particular, the example shown in FIG. 6B illustrates multiple modules or models using two different computational models, namely the first computational model 114a termed “dependency” and the second computational model 114b termed “activation” in FIG. 6B, to identify hypothetical responders. Therein, the location of individual dots indicates a potential tumor that may respond to a given drug.

[0313] In particular, FIG. 6B shows a scatterplot stratifying dependency on the KRAS signaling pathway versus the activation of the KRAS signaling pathway, which may be computed based on the activation of the MAPK signaling pathway, across a collection of lung tumor cell lines.

[0314] As can be seen in FIG. 6B, some of the lung tumor cell lines illustrated in Figure 6B, in particular those associated with G12C mutations or other KRAS mutations, have dependency and activity or activation scores located in the first region 220, which corresponds to the top right quadrant in FIG. 6B. The remaining samples or dots, in particular those associated with the wild type, are located in the second region 230 of the phenotypic vector space 212 that is associated with the second phenotype of the tumor non-responsive to the KRAS therapeutic agent.

[0315] FIG. 6C is an example of a multi-model classifier 115 and / or multi-axis model 200 using five different modules or models to identify hypothetical responders for a plurality oftumor samples or cell lines (such as lung tumor cell lines as in the example of FIG. 6B) shown as individual data points or dots in FIG. 6C.

[0316] Specifically, the example of FIG. 6C is based on the first computational model 114a modelling dependency on the KRAS signaling pathway and defining axis 202, the second (or at least one additional) computational model 114b modelling activation of the KRAS signaling pathway and defining axis 204, e.g. based on activation of the MAPK signaling pathway, and three further or additional computational models 114c modeling three further tumor biology parameters and defining three further or additional axes 206, 208, 210. The three further or additional models 114c may, for example, model activation of the PI3K signaling pathway, tumor microenvironment, and activation of the Androgen Receptor signaling pathway. One or more other tumor biology parameters may be selected in addition or as an alternative. The five actual modules or models 114a-c define and / or are annotated on the five different axes 202, 204, 206, 208, 210. As in the case of FIG. 6B, the location of individual dots indicates a potential tumor that may respond to a given drug.

[0317] In particular, the phenotypic vector space 212 or vector space 212 expanded or defined by the five models 114a-c and the respective axes 202-210 comprises at least one first region 220 associated with the first phenotype of the tumor responsive to KRAS therapy or a given KRAS therapeutic agent, and at least one second region 230 associated with the second phenotype of the tumor non-responsive to KRAS therapy or a given KRAS therapeutic agent. It is emphasized that particularly in the multi-dimensional phenotypic vector space 212 having more than three dimensions, as shown in FIG. 6C, a plurality of first regions 220 and / or a plurality of second regions 230 may be present. Several first regions 220 can be connected to each other or can be separated from each other by one or more second regions 230. Alternatively or additionally, a plurality of second regions 230 and / or a plurality of first regions 220 may be present. Several second regions 230 can be connected to each other or can be separated from each other by one or more first regions 220.

[0318] As in the example in two dimensions of FIG. 6B, scores or values for the five tumor biology parameters or models 114a-c considered can be mapped to the phenotypic vector space 212 to determine whether the respective tumor sample or cell line is located within or can be assigned to at least one first region 220 (and hence represents a tumor responsive to the KRAS therapeutic agent), or whether the respective tumor sample or cell line is located within or can beassigned to at least one second region 230 (and hence represents a tumor non-responsive to the KRAS therapeutic agent). Tumor samples or cell lines classified as responders or being responsive to the KRAS therapeutic agent are shown as solid dots in FIG. 6C, whereas tumor samples or cell lines classified as non-responders are shown as circles in FIG. 6C.

[0319] FIGs. 7A to 7F each illustrate steps of a computer-implemented method of determining a tumor response to a KRAS therapeutic agent and / or to KRAS therapy according to exemplary embodiments. FIGs. 7A to 7F illustrate lung tumors, wherein circles represent decreased cytotoxicity (non-responders) and solid dots represent increased cytotoxicity (responders). Analogue to FIG. 6B, each of FIGs. 7A to 7F schematically illustrates an output of a classifier 115 and / or multi-axis model 200 comprising the first computational model 114a modelling dependency on the KRAS signaling pathway, which is annotated on the x-axis in FIGs. 7A to 7F, and the second computational model 114b modelling activation of the KRAS signaling pathway, which is annotated on the y-axis in FIGs. 7A to 7F. Therein, activation of the KRAS signaling pathway is modelled or computed based on activation of the MAPK signaling pathway in the examples of FIGs. 7A to 7F.

[0320] Specifically, FIGs. 7A to 7F represent the ability to use the method according to the present disclosure to stratify response to multiple different drugs having different molecular mechanisms of actions and across multiple different model systems, such as e.g. 2D-cell lines, 3D cell lines, and xenograft, using a combination of the first and second computational models 114a, 114b to model dependency and activation. All data shown in FIGs. 7A to 7F correspond to data sourced from peer-reviewed, published experimental studies data using available KRAS G12C Inhibitors’ Cytotoxicity Data in Preclinical Models. Therein, FIG. 7A shows the results for all available preclinical models, including 2D cell cultures, 3D cell cultures, and Xenograft treated with either Sotorasib, RMC-6291, or Adagrasib as KRAS therapeutic agent. FIG. 7B shows the results for Sotorasib treated 2D-cell culture models, FIG. 7C shows the results for RMC-6291 treated 2D-cell culture models, FIG. 7D shows the results for Adagrasib treated 2D- cell culture, FIG. 7E shows the results for Adagrasib treated 3D-cell models, and FIG. 7F shows the results for Adagrasib treated Xenograft models.

[0321] As in the example of FIG. 6B, cell lines or models responsive to the KRAS therapeutic agent are located in first region 220 of the phenotypic vector space 212, which corresponds to the top right quadrant in each of FIGs. 7A to 7F and is associated with high- 13 -dependency and high activation scores. As is evident from FIGs. 7A to 7F the method according to the present disclosure can be applied across multiple different therapeutic agents, such as Sotorasib, RMC-6291, or Adagrasib, and across multiple different cell model systems, including 2D cell culture models, 3D cell culture models and xenograft models.

[0322] FIGs. 8A to 8C show Kaplan-Meier survival curves for clinical data and predicted data using a computer-implemented method of determining a tumor response to a KRAS therapeutic agent according to an exemplary embodiment. Therein, each of the Kaplan-Meier survival curves shown in FIGs. 8A to 8C provide a graphical representation to estimate the probability of time on Sotorasib therapy as a surrogate measure for progression-free survival, shown on the y-axis of FIGs. 8A to 8C, over time for non-small cell lung cancer (NSCLC) patients treated with Sotorasib, shown as number of days on Sotorasib on the x-axis of FIGs. 8A to 8C.

[0323] FIG. 8A shows Kaplan-Meier survival curves for real world clinical data. It shows a total of 64 cancer patients treated with Sotorasib, illustrated as dashed line 300 or baseline 300. The black thick solid line 310 in FIG. 8A represents patients having clinical benefit (investigator determined objective tumor response or stable disease) from and / or being responsive to the Sotorasib treatment. The grey thin solid line 320 in FIG. 8A represents the remaining patients having no clinical benefit from and / or being non-responsive to the Sotorasib treatment.

[0324] FIG. 8B predicted data using a computer-implemented method of determining a tumor response to a KRAS therapeutic agent according to an exemplary embodiment. It again shows the total of 64 cancer patients treated with Sotorasib as dashed line 300 or baseline 300 in comparison with the patients stratified as being responsive (thick solid black line 300 in FIG. 8B) and as being non-responsive (grey thin solid line 320 in FIG. 8B) to the KRAS therapeutic agent Sotorasib by applying the computer-implemented method of determining tumor response to the KRAS therapeutic agent according to the present disclosure.

[0325] FIG. 8C is a combination of FIG. 8A and 8B and contains the dotted baseline 300, the actual probability of survival of patients with clinical benefit (line 310) compared to the predicted patients with clinical benefit (line 312), and the actual probability of survival of patients without clinical benefit (line 320) compared to the predicted patients without clinical benefit (line 322).

[0326] As is evident from a comparison of FIGs. 8A and 8B, which is shown in consolidated form in FIG. 8C, curve 312 of patients being responsive and curve 322 of patients being non- responsive to the KRAS therapeutic agent as predicted by the method disclosed herein closely follow the corresponding curves 310, 320 of the clinical data. For instance, applying the computer-implemented method disclosed herein provides an estimate of 32 patients having clinical benefit from the treatment and / or being responsive to the KRAS therapeutic agent Sotorasib with a median of 338 days on treatment, which compares to 35 patients in fact being responsive to the KRAS treatment with a median of 469 days on treatment. Likewise, applying the computer-implemented method disclosed herein provides an estimate of 32 patients having no clinical benefit from the treatment and / or being responsive to the KRAS therapeutic agent Sotorasib with a median of 158 days on treatment, which compares to 29 patients in fact being non-responsive to the KRAS treatment and / or having no clinical benefit with a median of 135 days on treatment.

[0327] Therefore, FIGs. 8A to 8C impressively illustrate the ability to use the computer- implemented method described herein to delineate drug response in relation to clinical benefit and survival. The illustration in FIGs. 8A to 8C shows examples using real world data from a cohort of patients that closely resembles the demographics of clinical trial cohorts that were utilized for Sotorasib Phase II and III development, demonstrating its utility for stratifying response in both a real world clinical context and in a clinical trial context.

[0328] Fig. 9 schematically shows outputs of multi-axes models 200, thereby illustrating a further use case or application of the method of the present disclosure. Therein, pan-RAS inhibitor sensitivity is clustered by biologies modeled by one or more computational models 114a, 114b, and 114c in preclinical models for non-small cell lung cancer (NSCLC), for colorectal cancer (CRC), and for pancreatic cancer (PDAC). Specifically, response to the RMC- 7977 Pan-RAS inhibitor, more precisely the median effective concentration (EC50), was experimentally derived across multiple histologies, including CRC, PDAC, and NSCLC cell lines (sample number N= 23, 17, 49, respectively, total N=89). For this purpose, dependency was computed by means of a first computational model 114a, and activation and survival were computed by means of a second and tertiary computational models 114b and 114c, respectively, configured to model and / or simulate activation of one or more biological signaling pathways, including one or more of the TNFa signaling pathway, the EGFR signaling pathway, theEstrogen signaling pathway, the peptidylprolyl isomerase A (PPIA) signaling pathway, the MAPK signaling pathway, the Androgen signaling pathway, the PI3K signaling pathway, and the p53 signaling pathway. Optionally, gene-specific synthetic lethal signatures, and / or peptidylprolyl isomerase A (PPIA) gene expression can be used.

[0329] Two different corresponding multi-axis models 200 were developed, one was based on a linear regression model with ElasticNet and recursive feature elimination to predict EC50 as continuous variable, while another multi-axes model 200 was based on a classification model using logistic regression with ElasticNet and recursive feature elimination to predict low or high EC50. Model performance for these two multi-axes models 200 is summarized in the table below as area under the receiver operating characteristic curve (ROC AUC).

[0330] Evidently, both multi-axes models 200 show good performance, thereby further demonstrating that the present disclosure can be used to predict responses across different cancer or tumor types. In particular, these results showcase the ability of the present disclosure to predict response to pan-RAS inhibitors in a mutation- and tissue-agnostic manner.

[0331] The output of the multi-axes models 200 is schematically shown in FIG. 9, which illustrates the principal components of the models 200 as determined based on principal component analysis (PCA). Therein, a first model axis 202 models the dependency, a second model axis 204 models the activation the KRAS signaling pathway and a third model axis 206 models sensitivity and / or tumor survival signaling pathway of the tumor. The selected model axes 202, 204, 206 cluster tissue- and mutation-specific pan-RAS inhibitor responses for NSCLC, CRC and PDAC tumors in corresponding regions 220a, 220b, 230. Circles represent G12C mutations, triangles represent G12D mutations, and crosses represent other mutations. Thedatapoint size in FIG. 9 corresponds to the RMC7977 sensitivity. Therein, region 220a (also referred to as response clade 1 in FIG. 9) clusters the best CRC, the best PDAC and the best G12D responders. Region 220b (also referred to as response clade 2 in FIG. 9) clusters the best NSCLC, the best G12C and intermediate PDAC responders. Region 230 (also referred to as nonresponders clade in FIG. 9) clusters the worst / non-responding CRC, the worst / non-responding PDAC, and the worst / non-responding NSCLC patients.

[0332] FIGs. 10, 11 and 12 are graphs illustrating exemplary use cases or applications of the method according to the present disclosure. Therein, responders to KRAS inhibitor monotherapy and immune checkpoint combinations in NSCLC and PDAC patients are identified. A combination of biological modules (e.g., MAPK, PI3K, and Trail activity or activation of the respective pathway) was modelled in one or more computational models 114a, b, c to predict KRAS monotherapy (KRAS inhibitor monotherapy) responders, and an immune signature of the tumor was modelled in a further computational model 114c to compute an immune score and to stratify immune checkpoint inhibitor (I CI). The intersection of these response groups can classify patients into those likely to respond to KRAS inhibitors, ICIs, or combination therapy, as schematically illustrated in FIG. 10.

[0333] FIG. 10 schematically illustrates an output of a multi-axes model 200 composed of or comprising the aforementioned computational models to stratify a real world cohort of sotorasib- treated NSCLC patients. The different colors of the dots illustrate progressive disease (PD), stable disease (SD), partial responders (PR), and complete responders (CR). Therein, activation scores are plotted on the x-axis and the immune score is plotted on the y-axis. The phenotypic space 212 includes a first region 220a (region A in FIG. 10) associated with patients likely to respond to ICI monotherapy, a second region 220b (region B in FIG. 10) associated with patients likely to respond to a combination therapy of KRAS inhibitor and immune checkpoint inhibitor, a third region 220c (region C in FIG. 10) associated with patients likely to respond to KRAS inhibitor monotherapy, and a fourth region 230 (region D in FIG. 10) associated with non-responders or patients to be treated with standard or conventional chemotherapy. Evidently, these results illustrate the ability to inform either monotherapy or combination therapy selection in multiple histologic settings including NSCLC.

[0334] FIG. 11 shows a Kaplan-Meier plot illustrating treatment durability (or time on sotorasib), with the patients from FIG. 10, grouped by response cohort. Specifically, curve 420-n-shows the Kaplan-Meier plot for patients predicted to likely respond to KRAS inhibitor therapy (KRAS high in FIG. 11), curve 430 shows the Kaplan-Meier plot for patients predicted to likely respond to ICI therapy and likely to not respond to KRAS inhibitor therapy (KRAS low, immune high in FIG. 11), and curve 440 shows the Kaplan-Meier plot for patients predicted to likely not respond to KRAS inhibitor therapy and not respond to ICI therapy (KRAS low, immune low in FIG. 11). Also, curve 440 corresponds to a shorter period of time on sotorasib and / or time on therapy, when compared to the other curves 420, 430, where the curve 420 for KRAS high shows the best median time on treatment and the curve 430 for KRAS low, immune high shows intermediate time on treatment. The table below the Kaplan-Meier plot ("At risk” table) summarizes information about the number of subjects at risk on being on sotorasib treatment at different time points of the analysis per response cohort. Rows of the table correspond to the respective response cohort as indicated and as shown in the Kaplan-Meier plot, and columns represent discrete time points in line with the x axis of the above displayed Kaplan-Meier plot and thus refer to T=0, 200, 400, 600, and 800 days on sotorasib, respectively.

[0335] FIG. 12 shows the output of a multi-axes model 200 illustrating clustering by biologies and hypothesized relationships to KRAS inhibitor therapy, ICI therapy or combination therapy in a CPTAC-3 (Clinical Proteomic Tumor Analysis Consortium, CPTAC) PDAC patient cohort. In particular, the principal components of the model 200 as determined based on principal component analysis (PCA) is shown. Therein, a first model axis 202 models the dependency, a second model axis 204 models the activation of the PI3K signaling pathway, a third model axis 206 models the activity of the MAPK signaling pathway, and a fourth model axis 208 models immune activity. Red dots correspond to the wild type, purple dots correspond to G12D mutations, brown dots to G12D mutations and blue dots correspond to activating mutations. A first region 220a of the multi-dimensional phenotypic space 212 corresponds to, is indicative of, or is associated with patients responding to ICI therapy. A second region 220b of the multi-dimensional phenotypic space 212 corresponds to, is indicative of, or is associated with patients responding to combination therapy, namely a combination of KRAS inhibitor therapy and ICI therapy. A third region 220c of the multi-dimensional phenotypic space 212 corresponds to, is indicative of, or is associated with patients responding to KRAS inhibitor therapy. A fourth region 220d of the multi-dimensional phenotypic space 212 corresponds to, is indicative of, or isassociated with patients not responding to KRAS inhibitor and / or ICI therapy, which thus should be treated with conventional or standard chemotherapy.

[0336] FIG. 13 shows the output of a multi-axes model 200 for a plurality of computational models 114a-c. In particular, FIG. 13 illustrates the use or application of the present disclosure in evaluating pharmacodynamic responses by comparing pre-treatment to post-treatment biologies. Therein, pre-treatment and post-treatment samples or biopsies of an 81 -year-old male with stage 4 non-small lung cancer, who developed resistance to sotorasib (KRAS G12C inhibitor), were sequenced and the corresponding RNA sequencing data were analysed using the method described herein. Specifically, a computational model 114a for modelling dependency of the tumor on the KRAS signaling pathway was utilized. Further computational models 114b, c were used to model activation of one or more biological signaling pathways at least partly related to KRAS and / or to model one or more further tumor parameters. Exemplary illustrated in FIG. 13 as corresponding axes is the output of computational models 114b, c modelling EGFR, MAPK, PI3K, p53, NFkB, Trail, WNT, JAK-STAT, TNFa, TGFb, VEGF, Hypoxia, Estrogen, Androgen, Immune, and Angiogenesis. Line 400 represents results obtained for the pre-treatment RNA sequencing data, and line 410 represents results obtained for the post-treatment RNA sequencing data.

[0337] When applied in the relapse setting, the approach of evaluating the RNA sequencing data with a plurality of computational models 114a-c can permit identification of potential resistance mechanisms and therapeutically actionable biological responses. This demonstrates a potential for use in molecular residual disease (MRD) monitoring and evaluating pharmacodynamic responses. For example, the upregulation of WNT, angiogenesis, and estrogen receptor signaling pathways (with over 150% increase in WNT / p-catenin activity, angiogenesis, and hormone activity) for the post-treatment data (line 410 in FIG. 13), relative to the pretreatment data (line 400), indicate this patient may benefit from receiving inhibitors towards these respective pathways. Hence, the method of the present disclosure can allow to stratify sotorasib relapsed patients to nominate follow-on therapeutic selection.

[0338] In the context of embodiments of the present disclosure, any numerical value indicated is typically associated with an interval of accuracy that the person skilled in the art will understand to still ensure the technical effect of the feature in question. As used herein, the deviation from the indicated numerical value is in the range of ± 10%, and preferably of ± 5%.The aforementioned deviation from the indicated numerical interval of ± 10%, and preferably of ± 5% is also indicated by the terms “about” and “approximately” used herein with respect to a numerical value.

[0339] FIG. 14 is a graph illustrating with a Kaplan-Meier plot predicted and observed times on sotorasib treatment at 180 days as a further exemplary use case or application of the method according to the present disclosure. Therein, the method according to the present disclosure was applied with the aim of predicting whether patients stay on a KRAS therapeutic agent treatment beyond a given number of days. RNA sequencing data from in total 72 non-small cell lung cancer (NSCLC) patients with known durations of treatment with sotorasib as an exemplary KRAS G12C inhibitor was used to calculate the modules, which served as inputs to a classifier 115 being in this example a decision tree-based classifier. As regards the modules, dependency was computed by means of a first computational model 114a, and in total 16 computational models were configured to model, e.g., activation of one or more biological signaling pathways at least partly related to KRAS activation, including MAPK signaling pathway, Androgen signaling pathway, JAK-STAT signaling pathway, Trail signaling pathway and PI3K signaling pathway. The classification model was then used to predict time on treatment at 180 days in NSCLC patients treated with sotorasib. Thus, the model obtained by the method of the present disclosure was used to predict whether patients treated with sotorasib were likely to remain on therapy for shorter (<180 days) or longer (>180 days) durations. Additional tuning and model refinement was performed with nested CV, in particular 10-fold 3x repeated nested CV.

[0340] FIG. 14 shows a Kaplan-Meier plot illustrating treatment durability (or time on sotorasib) of the 72 NSCLC patients, grouped by cohort. Specifically, the dark gray solid line indicates the curve showing the Kaplan-Meier plot for patients predicted to be likely on sotorasib at day 180 (cohort labelled “On_sotorasib_at_180_days”) with the median time on sotorasib treatment being 472 days for the 40 patients of this group. The light gray solid line indicates the curve showing the Kaplan-Meier plot for patients predicted likely not to be on sotorasib at day 180 (cohort labelled “No On_sotorasib_at_180_days”) with the median time on sotorasib treatment being 98 days for the 32 patients of this group. For comparison, non-predicted data for all 77 patients are shown as black dotted line (cohort labelled “Baseline (all)”) with the median time on sotorasib treatment being 213 days. Thus, observed time to disease progression was comparable to results obtained in the CodeBreaK 200 phase III trial of KRAS G12C NSCLCpatients, wherein the median progression-free survival (PFS) for sotorasib was with 5.6 months about 170 days. The table below the Kaplan-Meier plot ("At risk” table) summarizes information about the number of subjects at risk of being on sotorasib treatment at different time points of the analysis per response cohort. Rows of the table correspond to the respective response cohort as indicated and as shown in the Kaplan-Meier plot, and columns represent discrete time points in line with the x axis of the above displayed Kaplan-Meier plot and thus refer to T=0, 200, 400, 600, and 800 days on sotorasib, respectively.

[0341] The model was further investigated by assessing accuracy, precision, recall, specificity, Fl score and AUROC. Accuracy refers to the number of correct predictions divided by the total number of predictions. Accuracy focuses on correct predictions, treating all classes equally, but requires balanced data in contrast to precision and recall. Precision (also referred to as Positive Predictive Value; PPV) refers to the number of true biomarker responses divided by the total number of predicted biomarker responses. Recall (also referred to as true positive rate or sensitivity) focusses on the number of true biomarker responses divided by the total number of actual responses. An Fl Score represents the harmonic mean of precision and recall. Specificity (also referred to as True Negative Rate) refers to the number of true biomarker non-responses divided by the total number of actual non-responses. AUROC (herein also referred to as ROC AU) refers to the Area Under the Receiver Operating Curve and indicates the degree to which a model is capable of distinguishing between classes. The model used to predict time on sotorasib treatment at day 180 day in NSCUC patients achieved an accuracy of 0.78, a precision of 0.78, a recall of 0.69, an Fl score of 0.73, a specificity of 0.85, and an AUROC of 0.8. Thus, the model achieved an AUROC of at least 0.8, demonstrating robust performance in predicting response duration.

[0342] Hence, this example demonstrated the potential of the method according to the present disclosure to stratify patients based on their predicted time on treatment as a surrogate measure for time to disease progression, which represents a critical clinical parameter.

[0343] As can be seen also in the example disclosed herein, the method according to the present disclosure has the advantage of being efficient and flexible and thus suitable for a variety of applications and / or use cases. Based on the evaluation of RNA sequencing data with the first computational model 114a and the one or more additional computational models 114b, optionally also at least one further computational model 114c, responsiveness or non-responsiveness of a tumor to a KRAS therapeutic agent can be obtained as classification information. Received classification information may aid determining tumor response to a KRAS therapeutic agent, determining efficacy of a KRAS therapeutic agent against a tumor, predicting time on a KRAS therapeutic agent treatment, stratifying patients based on their predicted response to a KRAS therapeutic agent, selecting and / or identifying patients likely to respond to a KRAS therapeutic agent treatment, and / or monitoring tumor response during treatment with a KRAS therapeutic agent. Further flexibility of the method according to the present disclosure can be provided and / or increased, e.g., in view of the choice of the classifier 115 and / or weights the classifier 115 can ascribe to the modules, wherein the weights preferably correlate with or are indicative of a respective model's relative importance. Hence, based on evaluated modules, comprising the first computational model 114a, the second computational model 114b (or additional computational model 114b), and optionally one or more further or additional computational models 114c, the method according to embodiments of the present disclosure can flexibly adapt to various different applications and / or use cases by identifying models or modules of particular relevance based on provided RNA sequencing data.

[0344] The present disclosure may be embodied as a system, a method, and / or a computer program product. The computer program product may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present disclosure.

[0345] The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non- exhaustive list of more specific examples of the computer readable storage medium includes the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or raised structures in a groove having instructions recorded thereon, and anysuitable combination of the foregoing. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.

[0346] Computer readable program instructions described herein can be downloaded to respective computing / processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and / or a wireless network. The network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and / or edge servers. A network adapter card or network interface in each computing / processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing / processing device.

[0347] Computer readable program instructions for carrying out operations of the present disclosure may be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, 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 conventional procedural programming languages, such as the “C” programming language or similar programming languages. The computer readable program instructions may execute entirely on the 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. In the latter scenario, the remote computer may be connected to the user’s computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate arrays (FPGA), or programmable logic arrays (PLA) may execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present disclosure.

[0348] 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 by computer readable program instructions.

[0349] 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 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.

[0350] The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process, such that the instructions which execute on the computer, other programmable apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0351] The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in thereverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustration, and combinations of blocks in the block diagrams and / or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts or carry out combinations of special purpose hardware and computer instructions.

[0352] The descriptions of the various embodiments of the present disclosure have been presented for purposes of illustration, but are not intended to be exhaustive or limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein was chosen to best explain the principles of the embodiments, the practical application or technical improvement over technologies found in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

CLAIMSWhat is claimed:

1. A computer-implemented method of determining a tumor response to a KRAS therapeutic agent and / or of determining an efficacy of a KRAS therapeutic agent against a tumor, the method comprising: providing, at a computing device including one or more processors for data processing, RNA sequencing data associated with a tumor, the RNA sequencing data being indicative of gene expression data of a plurality of genes involved in one or more biological signaling pathways at least partly related to KRAS; evaluating the RNA sequencing data with a first computational model modelling at least one of dependency of the tumor on the KRAS signaling pathway, and sensitivity of the tumor to KRAS inhibitor therapy; evaluating the RNA sequencing data with one or more additional computational models modelling activation of one or more biological signaling pathways at least partly related to KRAS; and classifying, based on the evaluation of the RNA sequencing data with the first computational model and the one or more additional computational models, the tumor as being responsive or non-responsive to the KRAS therapeutic agent.

2. The method according to the preceding claim, wherein the first computational model is configured to model one or both of the dependency of the tumor on the KRAS signaling pathway and the sensitivity of the tumor toKRAS inhibitor therapy as continuous tumor biology parameter and / or with a continuous metric; and / or wherein the one or more additional computational models are configured to model the activation of said one or more biological signaling pathways at least partly related to KRAS as continuous tumor biology parameter and / or with a continuous metric.

3. The method according to any one of the preceding claims, wherein the first computational model and the one or more additional computational models each define a model axis of a computational multi-axis model modelling at least two different tumor biology parameters selected from the group consisting of dependency of the tumor on the KRAS signaling pathway, sensitivity of the tumor to KRAS inhibitor therapy, and activation of the one or more biochemical signaling pathways at least partly related to KRAS.

4. The method according to the preceding claim, wherein the multi-axis model includes two or more model axes, the two or more model axes comprising at least a first model axis representative of at least one of the dependency of the tumor on the KRAS signaling pathway and the sensitivity of the tumor to KRAS inhibitor therapy, and a second model axis representative of the activation of the one or more biological signaling pathways at least partly related to KRAS.

5. The method according to the preceding claim, wherein the multi-axis model includes at least one further model axes representative of one or more of: one or more cell or tumor intrinsic signaling pathways, one or more cell or tumor extrinsic signaling pathways, cellular compositionof the tumor, cellular states of one or more tumor cells, biological activities of one or more tumor cells, molecular processes, genetic variants, synthetic lethal partners, metabolic signatures, modifications, and protein-interaction networks.

6. The method according to any one of claims 3 to 5, wherein the multi-axis model defines a multi-dimensional phenotypic vector space for the phenotype of the tumor; and wherein the multi-dimensional phenotypic vector space includes at least a first region associated with a first phenotype of the tumor responsive to the KRAS therapeutic agent and a second region associated with a second phenotype of the tumor non-responsive to KRAS therapeutic agent.

7. The method according to any one of claims 3 to 6, further comprising: determining, based on the evaluation of the RNA sequencing data with the first computational model, one or both of a dependency score indicative of a degree of dependency of the tumor on the KRAS signaling pathway and a sensitivity score indicative of a degree of responsiveness of the tumor to KRAS inhibitor therapy; determining, based on the evaluation of the RNA sequencing data with the one or more additional computational models, at least one activation state score indicative of a degree of activation of the one or more biological signaling pathways at least partly related to KRAS; and mapping the determined at least one activation state score and at least one of the dependency score and the sensitivity score into the multi-dimensional phenotypic vector space defined by the multi-axis model.

8. The method according to claims 6 and 7, further comprising: determining that the at least one activation state score and at least one of the dependency score and the sensitivity score are located within the first region of the multi-dimensional phenotypic vector space associated with the first phenotype of the tumor responsive to the KRAS therapeutic agent, thereby classifying the tumor as being responsive to the KRAS therapeutic agent; and / or determining that the at least one activation state score and at least one of the dependency score and the sensitivity score are located within the second region of the multi-dimensional phenotypic vector space associated with the second phenotype of the tumor non-responsive to the KRAS therapeutic agent, thereby classifying the tumor as being non-responsive to the KRAS therapeutic agent.

9. The method according to any one of the preceding claims, wherein evaluating the RNA sequencing data with the first computational model includes determining one or both of a dependency score indicative of a degree of dependency of the tumor on the KRAS signaling pathway and a sensitivity score indicative of a degree of responsiveness of the tumor to KRAS inhibitor therapy; and / or wherein evaluating the RNA sequencing data with the one or more additional computational models includes determining one or more activation state scores indicative of a degree of activation of the one or more biological signaling pathways at least partly related to KRAS.

10. The method according to the preceding claim, further comprising:providing the at least one activation state score and at least one of the dependency score and the sensitivity score as inputs to a classifier for classifying the tumor as being responsive or non-responsive to the KRAS therapeutic agent.

11. The method according to the preceding claim, wherein the classifier is based on logistic regression.

12. The method according to any one of the preceding claims, wherein the first computational model and the one or more additional computational models are independent computational models.

13. The method according to any one of the preceding claims, wherein the first computational model is a trained machine learning model; and / or wherein at least one of the one or more additional computational models is a statistical model.

14. The method according to any one of the preceding claims, wherein the first computational model is a trained machine learning model configured to model dependency of the tumor on the KRAS signaling pathway and / or sensitivity of the tumor to KRAS inhibitor therapy based on model features associated with a plurality genes involved in one or more of the RAS signaling, the MAPK signaling, the PI3K signaling, and the EGFR signaling pathways.

15. The method according to the preceding claim, wherein the first computational model is trained based on LI regularization and L2 regularization.

16. The method according to any one of the preceding claims, wherein at least one of the one or more additional computational models is a statistical model configured to model activation of one or more biological signaling pathways at least partly related to KRAS based on quantifying gene expression of a plurality of genes involved in one or more biological signaling pathways at least partly related to KRAS.

17. The method according to the preceding claim, wherein the one or more biological signaling pathways at least partly related to KRAS include Androgen Receptor signaling pathway, Estrogen Receptor signaling pathway, Hypoxia signaling pathway, JAK-STAT signaling pathway, MAPK signaling pathway, NFkB signaling pathway, Notch signaling pathway, PI3K / AKT signaling pathway, p53 signaling pathway, RTK signaling pathway, TGFb signaling pathway, TNFa signaling pathway, Trail signaling pathway, and WNT / p-Catenin signaling pathway.

18. The method according to any one of the preceding claims, further comprising: evaluating the RNA sequencing data with at least one further computational model configured to model at least one further tumor parameter, wherein the at least one further tumor parameter is indicative of one or more of a microenvironment of the tumor, RNA sequencing metadata, demographic data of the patient, angiogenesis of the tumor, activity of the VEGF signaling pathway, and cellular composition ofthe tumor, preferably wherein the at least one further computational model is a trained machine learning model.

19. The method according to any one of the preceding claims, wherein providing the RNA sequencing data at the computing device includes one or more of accessing the RNA sequencing data with the computing device, obtaining the RNA sequencing data at the computing device, receiving the RNA sequencing data at the computing device, and retrieving the RNA sequencing data with the computing device.

20. The method according to any one of the preceding claims, wherein providing the RNA sequencing data at the computing device includes measuring the RNA sequencing data based on one or more RNA sequencing modalities including total-RNA sequencing, polyA enriched sequencing, and RNA-Exome-based sequencing.

21. The method according to any one of the preceding claims, wherein the RNA sequencing data include data associated with human tumor tissue, with human-derived tumor tissue, or preclinical tumor tissue.

22. The method according to any one of the preceding claims, wherein the RNA sequencing data include data from one or more sources including fresh tissue, biofluid, Formalin-Fixed Paraffin-Embedded, cell culture, and a cell-free source.

23. The method according to any one of the preceding claims, further comprising:normalizing the RNA sequencing data based on one or more of a counts per million normalization, a fragments per kilobase million normalization, a transcripts per million normalization, an upper quartile normalization, a normalization with respect to counts adjusted with upper quartile factors, a normalization with respect to trimmed mean of M- values, and a normalization with respect to counts adjusted with trimmed mean of M- values factors.

24. The method according to any one of the preceding claims, wherein the RNA sequencing data is acquired prior to administration of the KRAS therapeutic agent, during treatment with the KRAS therapeutic agent or after administration of the KRAS therapeutic agent.

25. The method according to any one of the preceding claims, wherein the RNA sequencing data is acquired prior to administration of the KRAS therapeutic agent, and wherein the method further comprises: obtaining further RNA sequencing data acquired after administration of the KRAS therapeutic agent; and determining, based on evaluating the further RNA sequencing data with the first computational model and the one or more additional computational models, an evolution and / or alteration of the tumor response to the KRAS therapeutic agent.

26. The method according to any one of the preceding claims, wherein the KRAS therapeutic agent is a KRAS inhibitor.

27. The method according to any one of the preceding claims, wherein the KRAS therapeutic agent is a mutation specific inhibitor or a pan-KRAS inhibitor.

28. The method according to any one of the preceding claims, wherein the KRAS therapeutic agent is a G12C inhibitor or a KRAS G12D inhibitor.

29. The method according to any one of the preceding claims, wherein the tumor is related to lung cancer, non-small cell lung cancer, pancreatic cancer, or a colorectal cancer.

30. The method of any one of the preceding claims, wherein the first computational model is based on one or more genes selected from the group consisting of TP53, SIKE1, DUSP4, MLPH, MAP3K5, DUSP6, ETV1, ANTXR2, RASSF1, KRAS, KLHL22, RAP1B, and RAB5A.

31. A computer program, which when executed on a computing device, instructs the computing device to perform steps of the method of any one of the preceding claims.

32. A non-transitory computer-readable medium storing a computer program according to the preceding claim.

33. A computing device with one or more processors, the computing device being configured to perform steps of the method according to any one of claims 1 to 30.

34. Use of RNA sequencing data for determining i) a tumor response to a KRAS therapeutic agent, ii) an efficacy of a KRAS therapeutic agent against a tumor and / or iii) a time on a KRAS therapeutic agent treatment, using the method according to any one of claims 1 to 30.

Images