Assessment of modelling strategies for drug response prediction in cell lines and xenografts

An extensive review on lung cancer therapeutics using machine learning techniques: state-of-the-art and perspectives

There are over 100 types of human cancer, accounting for millions of deaths every year. Lung cancer alone claims over 1.8 million lives per year and is expected to surpass 3.2 million by 2050, which underscores the urgent need for rapid drug development and repurposing initiatives. The application of AI emerges as a pivotal solution to developing anti-cancer therapeutics. This state-of-the-art review aims to explore the various applications of AI in lung cancer therapeutics. Predictive models can analyse large datasets, including clinical data, genetic information, and treatment outcomes, for novel drug design and to generate personalised treatment recommendations, potentially optimising therapeutic strategies, enhancing treatment efficacy, and minimising adverse effects. A thorough literature review study was conducted based on articles indexed in PubMed and Scopus. We compiled the use of various machine learning approaches, including CNN, RNN, GAN, VAEs, and other AI techniques, enhancing efficiency with accuracy exceeding 95%, which is validated through a computer-aided drug design process. AI can revolutionise lung cancer therapeutics, streamlining processes and saving biological scientists' time and effort-however, further research is needed to overcome challenges and fully unlock AI's potential in Lung Cancer Therapeutics.

Integration of Pan-Cancer Cell Line and Single-Cell Transcriptomic Profiles Enables Inference of Therapeutic Vulnerabilities in Heterogeneous Tumors

Single-cell RNA sequencing (scRNA-seq) greatly advanced the understanding of intratumoral heterogeneity by identifying distinct cancer cell subpopulations. However, translating biological differences into treatment strategies is challenging due to a lack of tools to facilitate efficient drug discovery that tackles heterogeneous tumors. Developing such approaches requires accurate prediction of drug response at the single-cell level to offer therapeutic options to specific cell subpopulations. Here, we developed a transparent computational framework (nicknamed scIDUC) to predict therapeutic efficacies on an individual cell basis by integrating single-cell transcriptomic profiles with large, data-rich pan-cancer cell line screening data sets. This method achieved high accuracy in separating cells into their correct cellular drug response statuses. In three distinct prospective tests covering different diseases (rhabdomyosarcoma, pancreatic ductal adenocarcinoma, and castration-resistant prostate cancer), the predicted results using scIDUC were accurate and mirrored biological expectations. In the first two tests, the framework identified drugs for cell subpopulations that were resistant to standard-of-care (SOC) therapies due to intrinsic resistance or tumor microenvironmental effects, and the results showed high consistency with experimental findings from the original studies. In the third test using newly generated SOC therapy-resistant cell lines, scIDUC identified efficacious drugs for the resistant line, and the predictions were validated with in vitro experiments. Together, this study demonstrates the potential of scIDUC to quickly translate scRNA-seq data into drug responses for individual cells, displaying the potential as a tool to improve the treatment of heterogenous tumors.

SIGNIFICANCE

A versatile method that infers cell-level drug response in scRNA-seq data facilitates the development of therapeutic strategies to target heterogeneous subpopulations within a tumor and address issues such as treatment failure and resistance.

Improving drug response prediction via integrating gene relationships with deep learning

Predicting the drug response of cancer cell lines is crucial for advancing personalized cancer treatment, yet remains challenging due to tumor heterogeneity and individual diversity. In this study, we present a deep learning-based framework named Deep neural network Integrating Prior Knowledge (DIPK) (DIPK), which adopts self-supervised techniques to integrate multiple valuable information, including gene interaction relationships, gene expression profiles and molecular topologies, to enhance prediction accuracy and robustness. We demonstrated the superior performance of DIPK compared to existing methods on both known and novel cells and drugs, underscoring the importance of gene interaction relationships in drug response prediction. In addition, DIPK extends its applicability to single-cell RNA sequencing data, showcasing its capability for single-cell-level response prediction and cell identification. Further, we assess the applicability of DIPK on clinical data. DIPK accurately predicted a higher response to paclitaxel in the pathological complete response (pCR) group compared to the residual disease group, affirming the better response of the pCR group to the chemotherapy compound. We believe that the integration of DIPK into clinical decision-making processes has the potential to enhance individualized treatment strategies for cancer patients.

Integration of Computational Docking into Anti-Cancer Drug Response Prediction Models

Cancer is a heterogeneous disease in that tumors of the same histology type can respond differently to a treatment. Anti-cancer drug response prediction is of paramount importance for both drug development and patient treatment design. Although various computational methods and data have been used to develop drug response prediction models, it remains a challenging problem due to the complexities of cancer mechanisms and cancer-drug interactions. To better characterize the interaction between cancer and drugs, we investigate the feasibility of integrating computationally derived features of molecular mechanisms of action into prediction models. Specifically, we add docking scores of drug molecules and target proteins in combination with cancer gene expressions and molecular drug descriptors for building response models. The results demonstrate a marginal improvement in drug response prediction performance when adding docking scores as additional features, through tests on large drug screening data. We discuss the limitations of the current approach and provide the research community with a baseline dataset of the large-scale computational docking for anti-cancer drugs.

Inferring therapeutic vulnerability within tumors through integration of pan-cancer cell line and single-cell transcriptomic profiles

Single-cell RNA sequencing greatly advanced our understanding of intratumoral heterogeneity through identifying tumor subpopulations with distinct biologies. However, translating biological differences into treatment strategies is challenging, as we still lack tools to facilitate efficient drug discovery that tackles heterogeneous tumors. One key component of such approaches tackles accurate prediction of drug response at the single-cell level to offer therapeutic options to specific cell subpopulations. Here, we present a transparent computational framework (nicknamed scIDUC) to predict therapeutic efficacies on an individual-cell basis by integrating single-cell transcriptomic profiles with large, data-rich pan-cancer cell line screening datasets. Our method achieves high accuracy, with predicted sensitivities easily able to separate cells into their true cellular drug resistance status as measured by effect size (Cohen's d > 1.0). More importantly, we examine our method's utility with three distinct prospective tests covering different diseases (rhabdomyosarcoma, pancreatic ductal adenocarcinoma, and castration-resistant prostate cancer), and in each our predicted results are accurate and mirrored biological expectations. In the first two, we identified drugs for cell subpopulations that are resistant to standard-of-care (SOC) therapies due to intrinsic resistance or effects of tumor microenvironments. Our results showed high consistency with experimental findings from the original studies. In the third test, we generated SOC therapy resistant cell lines, used scIDUC to identify efficacious drugs for the resistant line, and validated the predictions with in-vitro experiments. Together, scIDUC quickly translates scRNA-seq data into drug response for individual cells, displaying the potential as a first-line tool for nuanced and heterogeneity-aware drug discovery.

Reassessing pharmacogenomic cell sensitivity with multilevel statistical models

Pharmacogenomic experiments allow for the systematic testing of drugs, at varying dosage concentrations, to study how genomic markers correlate with cell sensitivity to treatment. The first step in the analysis is to quantify the response of cell lines to variable dosage concentrations of the drugs being tested. The signal to noise in these measurements can be low due to biological and experimental variability. However, the increasing availability of pharmacogenomic studies provides replicated data sets that can be leveraged to gain power. To do this, we formulate a hierarchical mixture model to estimate the drug-specific mixture distributions for estimating cell sensitivity and for assessing drug effect type as either broad or targeted effect. We use this formulation to propose a unified approach that can yield posterior probability of a cell being susceptible to a drug conditional on being a targeted effect or relative effect sizes conditioned on the cell being broad. We demonstrate the usefulness of our approach via case studies. First, we assess pairwise agreements for cell lines/drugs within the intersection of two data sets and confirm the moderate pairwise agreement between many publicly available pharmacogenomic data sets. We then present an analysis that identifies sensitivity to the drug crizotinib for cells harboring EML4-ALK or NPM1-ALK gene fusions, as well as significantly down-regulated cell-matrix pathways associated with crizotinib sensitivity.

Pan-Cancer Drug Response Prediction Using Integrative Principal Component Regression

The pursuit of precision oncology heavily relies on large-scale genomic and pharmacological data garnered from preclinical cancer model systems such as cell lines. While cell lines are instrumental in understanding the interplay between genomic programs and drug response, it well-established that they are not fully representative of patient tumors. Development of integrative methods that can systematically assess the commonalities between patient tumors and cell-lines can help bridge this gap. To this end, we introduce the Integrative Principal Component Regression (iPCR) model which uncovers both joint and model-specific structured variations in the genomic data of cell lines and patient tumors through matrix decompositions. The extracted joint variation is then used to predict patient drug responses based on the pharmacological data from preclinical models. Moreover, the interpretability of our model allows for the identification of key driver genes and pathways associated with the treatment-specific response in patients across multiple cancers. We demonstrate that the outputs of the iPCR model can assist in inferring both model-specific and shared co-expression networks between cell lines and patients. We show that iPCR performs favorably compared to competing approaches in predicting patient drug responses, in both simulation studies and real-world applications, in addition to identifying key genomic drivers of cancer drug responses.

A performance evaluation of drug response prediction models for individual drugs

Drug response prediction is important to establish personalized medicine for cancer therapy. Model construction for predicting drug response (i.e., cell viability half-maximal inhibitory concentration [IC₅₀]) of an individual drug by inputting pharmacogenomics in disease models remains critical. Machine learning (ML) has been predominantly applied for prediction, despite the advent of deep learning (DL). Moreover, whether DL or traditional ML models are superior for predicting cell viability IC_50s has to be established. Herein, we constructed ML and DL drug response prediction models for 24 individual drugs and compared the performance of the models by employing gene expression and mutation profiles of cancer cell lines as input. We observed no significant difference in drug response prediction performance between DL and ML models for 24 drugs [root mean squared error (RMSE) ranging from 0.284 to 3.563 for DL and from 0.274 to 2.697 for ML; R² ranging from -7.405 to 0.331 for DL and from -8.113 to 0.470 for ML]. Among the 24 individual drugs, the ridge model of panobinostat exhibited the best performance (R² 0.470 and RMSE 0.623). Thus, we selected the ridge model of panobinostat for further application of explainable artificial intelligence (XAI). Using XAI, we further identified important genomic features for panobinostat response prediction in the ridge model, suggesting the genomic features of 22 genes. Based on our findings, results for an individual drug employing both DL and ML models were comparable. Our study confirms the applicability of drug response prediction models for individual drugs.

Translating transcriptomic findings from cancer model systems to humans through joint dimension reduction

Model systems are an essential resource in cancer research. They simulate effects that we can infer into humans, but come at a risk of inaccurately representing human biology. This inaccuracy can lead to inconclusive experiments or misleading results, urging the need for an improved process for translating model system findings into human-relevant data. We present a process for applying joint dimension reduction (jDR) to horizontally integrate gene expression data across model systems and human tumor cohorts. We then use this approach to combine human TCGA gene expression data with data from human cancer cell lines and mouse model tumors. By identifying the aspects of genomic variation joint-acting across cohorts, we demonstrate how predictive modeling and clinical biomarkers from model systems can be improved.

A systematic literature review for the prediction of anticancer drug response using various machine-learning and deep-learning techniques

Computational methods have gained prominence in healthcare research. The accessibility of healthcare data has greatly incited academicians and researchers to develop executions that help in prognosis of cancer drug response. Among various computational methods, machine-learning (ML) and deep-learning (DL) methods provide the most consistent and effectual approaches to handle the serious aftermaths of the deadly disease and drug administered to the patients. Hence, this systematic literature review has reviewed researches that have investigated drug discovery and prognosis of anticancer drug response using ML and DL algorithms. Fot this purpose, PRISMA guidelines have been followed to choose research papers from Google Scholar, PubMed, and Sciencedirect websites. A total count of 105 papers that align with the context of this review were chosen. Further, the review also presents accuracy of the existing ML and DL methods in the prediction of anticancer drug response. It has been found from the review that, amidst the availability of various studies, there are certain challenges associated with each method. Thus, future researchers can consider these limitations and challenges to develop a prominent anticancer drug response prediction method, and it would be greatly beneficial to the medical professionals in administering non-invasive treatment to the patients.

Utilization of Cancer Cell Line Screening to Elucidate the Anticancer Activity and Biological Pathways Related to the Ruthenium-Based Therapeutic BOLD-100

BOLD-100 (sodium trans-[tetrachlorobis(1H indazole)ruthenate(III)]) is a ruthenium-based anticancer compound currently in clinical development. The identification of cancer types that show increased sensitivity towards BOLD-100 can lead to improved developmental strategies. Sensitivity profiling can also identify mechanisms of action that are pertinent for the bioactivity of complex therapeutics. Sensitivity to BOLD-100 was measured in a 319-cancer-cell line panel spanning 24 tissues. BOLD-100's sensitivity profile showed variation across the tissue lineages, including increased response in esophageal, bladder, and hematologic cancers. Multiple cancers, including esophageal, bile duct and colon cancer, had higher relative response to BOLD-100 than to cisplatin. Response to BOLD-100 showed only moderate correlation to anticancer compounds in the Genomics of Drug Sensitivity in Cancer (GDSC) database, as well as no clear theme in bioactivity of correlated hits, suggesting that BOLD-100 may have a differentiated therapeutic profile. The genomic modalities of cancer cell lines were modeled against the BOLD-100 sensitivity profile, which revealed that genes related to ribosomal processes were associated with sensitivity to BOLD-100. Machine learning modeling of the sensitivity profile to BOLD-100 and gene expression data provided moderative predictive value. These findings provide further mechanistic understanding around BOLD-100 and support its development for additional cancer types.

Optimizing Drug Response Study Design in Patient-Derived Tumor Xenografts

Patient-derived tumor xenograft (PDX) models were used to evaluate the effectiveness of preclinical anticancer agents. A design using 1 mouse per patient per drug (1 × 1 × 1) was considered practical for large-scale drug efficacy studies. We evaluated modifiable parameters that could increase the statistical power of this design based on our consolidated PDX experiments. Real studies were used as a reference to investigate the relationship between statistical power with treatment effect size, inter-mouse variation, and tumor measurement frequencies. Our results showed that large effect sizes could be detected at a significance level of .2 or .05 under a 1 × 1 × 1 design. We found that the minimum number of mice required to achieve 80% power at an alpha level of .05 under all situations explored was 21 mice per group for a small effect size and 5 mice per group for a medium effect size.

Deep Morphology Learning Enhances Ex Vivo Drug Profiling-Based Precision Medicine

Drug testing in patient biopsy-derived cells can identify potent treatments for patients suffering from relapsed or refractory hematologic cancers. Here we investigate the use of weakly supervised deep learning on cell morphologies (DML) to complement diagnostic marker-based identification of malignant and nonmalignant cells in drug testing. Across 390 biopsies from 289 patients with diverse blood cancers, DML-based drug responses show improved reproducibility and clustering of drugs with the same mode of action. DML does so by adapting to batch effects and by autonomously recognizing disease-associated cell morphologies. In a post hoc analysis of 66 patients, DML-recommended treatments led to improved progression-free survival compared with marker-based recommendations and physician's choice-based treatments. Treatments recommended by both immunofluorescence and DML doubled the fraction of patients achieving exceptional clinical responses. Thus, DML-enhanced ex vivo drug screening is a promising tool in the identification of effective personalized treatments.

SIGNIFICANCE

We have recently demonstrated that image-based drug screening in patient samples identifies effective treatment options for patients with advanced blood cancers. Here we show that using deep learning to identify malignant and nonmalignant cells by morphology improves such screens. The presented workflow is robust, automatable, and compatible with clinical routine. This article is highlighted in the In This Issue feature, p. 476.

Interpretable Machine Learning Models to Predict the Resistance of Breast Cancer Patients to Doxorubicin from Their microRNA Profiles

Doxorubicin is a common treatment for breast cancer. However, not all patients respond to this drug, which sometimes causes life-threatening side effects. Accurately anticipating doxorubicin-resistant patients would therefore permit to spare them this risk while considering alternative treatments without delay. Stratifying patients based on molecular markers in their pretreatment tumors is a promising approach to advance toward this ambitious goal, but single-gene gene markers such as HER2 expression have not shown to be sufficiently predictive. The recent availability of matched doxorubicin-response and diverse molecular profiles across breast cancer patients permits now analysis at a much larger scale. 16 machine learning algorithms and 8 molecular profiles are systematically evaluated on the same cohort of patients. Only 2 of the 128 resulting models are substantially predictive, showing that they can be easily missed by a standard-scale analysis. The best model is classification and regression tree (CART) nonlinearly combining 4 selected miRNA isoforms to predict doxorubicin response (median Matthew correlation coefficient (MCC) and area under the curve (AUC) of 0.56 and 0.80, respectively). By contrast, HER2 expression is significantly less predictive (median MCC and AUC of 0.14 and 0.57, respectively). As the predictive accuracy of this CART model increases with larger training sets, its update with future data should result in even better accuracy.

A comprehensive evaluation of regression-based drug responsiveness prediction models, using cell viability inhibitory concentrations (IC50 values)

MOTIVATION

Predicting drug response is critical for precision medicine. Diverse methods have predicted drug responsiveness, as measured by the half-maximal drug inhibitory concentration (IC50), in cultured cells. Although IC50s are continuous, traditional prediction models have dealt mainly with binary classification of responsiveness. However, since there are few regression-based IC50 predictions, comprehensive evaluations of regression-based IC50 prediction models, including machine learning (ML) and deep learning (DL), for diverse data types and dataset sizes, have not been addressed.

RESULTS

Here, we constructed 11 input data settings, including multi-omics settings, with varying dataset sizes, then evaluated the performance of regression-based ML and DL models to predict IC50s. DL models considered two convolutional neural network architectures: CDRScan and residual neural network (ResNet). ResNet was introduced in regression-based DL models for predicting drug response for the first time. As a result, DL models performed better than ML models in all the settings. Also, ResNet performed better than or comparable to CDRScan and ML models in all settings.

AVAILABILITY AND IMPLEMENTATION

The data underlying this article are available in GitHub at https://github.com/labnams/IC50evaluation.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Integrating Molecular Graph Data of Drugs and Multiple -Omic Data of Cell Lines for Drug Response Prediction

Previous studies have either learned drug's features from their string or numeric representations, which are not natural forms of drugs, or only used genomic data of cell lines for the drug response prediction problem. Here, we proposed a deep learning model, GraOmicDRP, to learn drug's features from their graph representation and integrate multiple -omic data of cell lines. In GraOmicDRP, drugs are represented as graphs of bindings among atoms; meanwhile, cell lines are depicted by not only genomic but also transcriptomic and epigenomic data. Graph convolutional and convolutional neural networks were used to learn the representation of drugs and cell lines, respectively. A combination of the two representations was then used to be representative of each pair of drug-cell line. Finally, the response value of each pair was predicted by a fully connected network. Experimental results indicate that transcriptomic data shows the best among single -omic data; meanwhile, the combinations of transcriptomic and other -omic data achieved the best performance overall in terms of both Root Mean Square Error and Pearson correlation coefficient. In addition, we also show that GraOmicDRP outperforms some state-of-the-art methods, including ones integrating -omic data with drug information such as GraphDRP, and ones using -omic data without drug information such as DeepDR and MOLI.

An overview of machine learning methods for monotherapy drug response prediction

For an increasing number of preclinical samples, both detailed molecular profiles and their responses to various drugs are becoming available. Efforts to understand, and predict, drug responses in a data-driven manner have led to a proliferation of machine learning (ML) methods, with the longer term ambition of predicting clinical drug responses. Here, we provide a uniquely wide and deep systematic review of the rapidly evolving literature on monotherapy drug response prediction, with a systematic characterization and classification that comprises more than 70 ML methods in 13 subclasses, their input and output data types, modes of evaluation, and code and software availability. ML experts are provided with a fundamental understanding of the biological problem, and how ML methods are configured for it. Biologists and biomedical researchers are introduced to the basic principles of applicable ML methods, and their application to the problem of drug response prediction. We also provide systematic overviews of commonly used data sources used for training and evaluation methods.

Using mathematical modeling to estimate time-independent cancer chemotherapy efficacy parameters

One of the primary cancer treatment modalities is chemotherapy. Unfortunately, traditional anti-cancer drugs are often not selective and cause damage to healthy cells, leading to serious side effects for patients. For this reason more targeted therapeutics and drug delivery methods are being developed. The effectiveness of new treatments is initially determined via in vitro cell viability assays, which determine the IC 50  of the drug. However, these assays are known to result in estimates of IC 50  that depend on the measurement time, possibly resulting in over- or under-estimation of the IC 50 . Here, we test the possibility of using cell growth curves and fitting of mathematical models to determine the IC 50  as well as the maximum efficacy of a drug ( ε max ). We measured cell growth of MCF-7 and HeLa cells in the presence of different concentrations of doxorubicin and fit the data with a logistic growth model that incorporates the effect of the drug. This method leads to measurement time-independent estimates of IC 50  and ε max , but we find that ε max  is not identifiable. Further refinement of this methodology is needed to produce uniquely identifiable parameter estimates.

Predicting patient response with models trained on cell lines and patient-derived xenografts by nonlinear transfer learning

Preclinical models have been the workhorse of cancer research, producing massive amounts of drug response data. Unfortunately, translating response biomarkers derived from these datasets to human tumors has proven to be particularly challenging. To address this challenge, we developed TRANSACT, a computational framework that builds a consensus space to capture biological processes common to preclinical models and human tumors and exploits this space to construct drug response predictors that robustly transfer from preclinical models to human tumors. TRANSACT performs favorably compared to four competing approaches, including two deep learning approaches, on a set of 23 drug prediction challenges on The Cancer Genome Atlas and 226 metastatic tumors from the Hartwig Medical Foundation. We demonstrate that response predictions deliver a robust performance for a number of therapies of high clinical importance: platinum-based chemotherapies, gemcitabine, and paclitaxel. In contrast to other approaches, we demonstrate the interpretability of the TRANSACT predictors by correctly identifying known biomarkers of targeted therapies, and we propose potential mechanisms that mediate the resistance to two chemotherapeutic agents.

Ill-matched timescales in coupled systems can induce oscillation suppression

We explore the behavior of two coupled oscillators, considering combinations of similar and dissimilar oscillators, with their intrinsic dynamics ranging from periodic to chaotic. We first investigate the coupling of two different real-world systems, namely, the chemical mercury beating heart oscillator and the electronic Chua oscillator, with the disparity in the timescales of the constituent oscillators. Here, we are considering a physical situation that is not commonly addressed: the coupling of sub-systems whose characteristic timescales are very different. Our findings indicate that the oscillations in coupled systems are quenched to oscillation death (OD) state, at sufficiently high coupling strength, when there is a large timescale mismatch. In contrast, phase synchronization occurs when their timescales are comparable. In order to further strengthen the concept, we demonstrate this timescale-induced oscillation suppression and phase synchrony through numerical simulations, with the disparity in the timescales serving as a tuning or control parameter. Importantly, oscillation suppression (OD) occurs for a significantly smaller timescale mismatch when the coupled oscillators are chaotic. This suggests that the inherent broad spectrum of timescales underlying chaos aids oscillation suppression, as the temporal complexity of chaotic dynamics lends a natural heterogeneity to the timescales. The diversity of the experimental systems and numerical models we have chosen as a test-bed for the proposed concept lends support to the broad generality of our findings. Last, these results indicate the potential prevention of system failure by small changes in the timescales of the constituent dynamics, suggesting a potent control strategy to stabilize coupled systems to steady states.

A gentle introduction to understanding preclinical data for cancer pharmaco-omic modeling

A central goal of precision oncology is to administer an optimal drug treatment to each cancer patient. A common preclinical approach to tackle this problem has been to characterize the tumors of patients at the molecular and drug response levels, and employ the resulting datasets for predictive in silico modeling (mostly using machine learning). Understanding how and why the different variants of these datasets are generated is an important component of this process. This review focuses on providing such introduction aimed at scientists with little previous exposure to this research area.

Large-scale pharmacogenomic studies and drug response prediction for personalized cancer medicine

The response rate of most anti-cancer drugs is limited because of the high heterogeneity of cancer and the complex mechanism of drug action. Personalized treatment that stratifies patients into subgroups using molecular biomarkers is promising to improve clinical benefit. With the accumulation of preclinical models and advances in computational approaches of drug response prediction, pharmacogenomics has made great success over the last 20 years and is increasingly used in the clinical practice of personalized cancer medicine. In this article, we first summarize FDA-approved pharmacogenomic biomarkers and large-scale pharmacogenomic studies of preclinical cancer models such as patient-derived cell lines, organoids, and xenografts. Furthermore, we comprehensively review the recent developments of computational methods in drug response prediction, covering network, machine learning, and deep learning technologies and strategies to evaluate immunotherapy response. In the end, we discuss challenges and propose possible solutions for further improvement.

A machine learning-based gene signature of response to the novel alkylating agent LP-184 distinguishes its potential tumor indications

BACKGROUND

Non-targeted cytotoxics with anticancer activity are often developed through preclinical stages using response criteria observed in cell lines and xenografts. A panel of the NCI-60 cell lines is frequently the first line to define tumor types that are optimally responsive. Open data on the gene expression of the NCI-60 cell lines, provides a unique opportunity to add another dimension to the preclinical development of such drugs by interrogating correlations with gene expression patterns. Machine learning can be used to reduce the complexity of whole genome gene expression patterns to derive manageable signatures of response. Application of machine learning in early phases of preclinical development is likely to allow a better positioning and ultimate clinical success of molecules. LP-184 is a highly potent novel alkylating agent where the preclinical development is being guided by a dedicated machine learning-derived response signature. We show the feasibility and the accuracy of such a signature of response by accurately predicting the response to LP-184 validated using wet lab derived IC50s on a panel of cell lines.

RESULTS

We applied our proprietary RADR® platform to an NCI-60 discovery dataset encompassing LP-184 IC50s and publicly available gene expression data. We used multiple feature selection layers followed by the XGBoost regression model and reduced the complexity of 20,000 gene expression values to generate a 16-gene signature leading to the identification of a set of predictive candidate biomarkers which form an LP-184 response gene signature. We further validated this signature and predicted response to an additional panel of cell lines. Considering fold change differences and correlation between actual and predicted LP-184 IC50 values as validation performance measures, we obtained 86% accuracy at four-fold cut-off, and a strong (r = 0.70) and significant (p value 1.36e-06) correlation between actual and predicted LP-184 sensitivity. In agreement with the perceived mechanism of action of LP-184, PTGR1 emerged as the top weighted gene.

CONCLUSION

Integration of a machine learning-derived signature of response with in vitro assessment of LP-184 efficacy facilitated the derivation of manageable yet robust biomarkers which can be used to predict drug sensitivity with high accuracy and clinical value.

Concise Polygenic Models for Cancer-Specific Identification of Drug-Sensitive Tumors from Their Multi-Omics Profiles

In silico models to predict which tumors will respond to a given drug are necessary for Precision Oncology. However, predictive models are only available for a handful of cases (each case being a given drug acting on tumors of a specific cancer type). A way to generate predictive models for the remaining cases is with suitable machine learning algorithms that are yet to be applied to existing in vitro pharmacogenomics datasets. Here, we apply XGBoost integrated with a stringent feature selection approach, which is an algorithm that is advantageous for these high-dimensional problems. Thus, we identified and validated 118 predictive models for 62 drugs across five cancer types by exploiting four molecular profiles (sequence mutations, copy-number alterations, gene expression, and DNA methylation). Predictive models were found in each cancer type and with every molecular profile. On average, no omics profile or cancer type obtained models with higher predictive accuracy than the rest. However, within a given cancer type, some molecular profiles were overrepresented among predictive models. For instance, CNA profiles were predictive in breast invasive carcinoma (BRCA) cell lines, but not in small cell lung cancer (SCLC) cell lines where gene expression (GEX) and DNA methylation profiles were the most predictive. Lastly, we identified the best XGBoost model per cancer type and analyzed their selected features. For each model, some of the genes in the selected list had already been found to be individually linked to the response to that drug, providing additional evidence of the usefulness of these models and the merits of the feature selection scheme.