Integrating heterogeneous drug sensitivity data from cancer pharmacogenomic studies

In vitro effects and mathematical modelling of CTCE-9908 (a chemokine receptor 4 antagonist) on melanoma cell survival

CTCE-9908, a CXC chemokine receptor 4 (CXCR4) antagonist, prevents CXCR4 phosphorylation and inhibits the interaction with chemokine ligand 12 (CXCL12) and downstream signalling pathways associated with metastasis. This study evaluated the in vitro effects of CTCE-9908 on B16 F10 melanoma cells with the use of mathematical modelling. Crystal violet staining was used to construct a mathematical model of CTCE-9908 B16 F10 (melanoma) and RAW 264.7 (non-cancerous macrophage) cell lines on cell viability to predict the half-maximal inhibitory concentration (IC50). Morphological changes were assessed using transmission electron microscopy. Flow cytometry was used to assess changes in cell cycle distribution, apoptosis via caspase-3, cell survival via extracellular signal-regulated kinase1/2 activation, CXCR4 activation and CXCL12 expression. Mathematical modelling predicted IC50 values from 0 to 100 h. At IC50, similar cytotoxicity between the two cell lines and ultrastructural morphological changes indicative of cell death were observed. At a concentration 10 times lower than IC50, CTCE-9908 induced inhibition of cell survival (p = 0.0133) in B16 F10 cells but did not affect caspase-3 or cell cycle distribution in either cell line. This study predicts CTCE-9908 IC50 values at various time points using mathematical modelling, revealing cytotoxicity in melanoma and non-cancerous cells. CTCE-9908 significantly inhibited melanoma cell survival at a concentration 10 times lower than the IC50 in B16 F10 cells but not RAW 264.7 cells. However, CTCE-9908 did not affect CXCR4 phosphorylation, apoptosis,\ or cell cycle distribution in either cell line.

Using birth-death processes to infer tumor subpopulation structure from live-cell imaging drug screening data

Tumor heterogeneity is a complex and widely recognized trait that poses significant challenges in developing effective cancer therapies. In particular, many tumors harbor a variety of subpopulations with distinct therapeutic response characteristics. Characterizing this heterogeneity by determining the subpopulation structure within a tumor enables more precise and successful treatment strategies. In our prior work, we developed PhenoPop, a computational framework for unravelling the drug-response subpopulation structure within a tumor from bulk high-throughput drug screening data. However, the deterministic nature of the underlying models driving PhenoPop restricts the model fit and the information it can extract from the data. As an advancement, we propose a stochastic model based on the linear birth-death process to address this limitation. Our model can formulate a dynamic variance along the horizon of the experiment so that the model uses more information from the data to provide a more robust estimation. In addition, the newly proposed model can be readily adapted to situations where the experimental data exhibits a positive time correlation. We test our model on simulated data (in silico) and experimental data (in vitro), which supports our argument about its advantages.

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.

High Mesothelin expression in pancreatic adenocarcinoma is associated with aggressive tumor features but not prognosis

Mesothelin is a cell surface marker expressed on most pancreatic cancers and has been associated with aggressive biology. Despite its popularity as a drug target, clinical relevance of Mesothelin expression in pancreatic cancer is unclear. We set out to define transcriptomic signatures associated with high Mesothelin expression and identify its role in tumor biology and its clinical relevance. We analyzed pancreatic adenocarcinomas in the cancer genome atlas (TCGA), (n = 145) and the results were validated using GSE62452 cohort (n = 69). We divided the cohorts into high and low Mesothelin expression by the median. High Mesothelin was not associated with progression-free, disease-free, disease specific, nor overall survival in TCGA cohort. Despite this, high Mesothelin expression was associated with high Ki67 expression and enriched all five cell proliferation-related Hallmark gene sets, but not with previously investigated pathways: TNF-alpha, PI3K, nor angiogenesis. Mesothelin expression did not correlate with MUC16 expression. The high Mesothelin pancreatic cancers demonstrated higher homologous recombination deficiency, fraction altered, and silent and non-silent mutation rates (all P < 0.001) that indicate aggressive cancer biology. However, lymphocyte infiltration score, TIL regional fraction, TCR richness, infiltration of CD8 T-cells, and cytolytic activity were all significantly lower in Mesothelin high tumors (all P < 0.015). Finally, we found that Mesothelin expression significantly correlated with sensitivity to cytotoxic chemotherapy in pancreatic cancer cell lines. In conclusion, high Mesothelin expression is associated with enhanced proliferation, depressed immune response, and sensitivity to cytotoxic chemotherapy, which may explain there was no difference in survival in pancreatic cancer patients.

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.

Refinement of safety and efficacy of anti-cancer chemotherapeutics by tailoring their site-specific intracellular bioavailability through transporter modulation

Low intracellular bioavailability, off-site toxicities, and multi drug resistance (MDR) are the major constraints involved in cancer chemotherapy. Many anticancer molecules fail to become a good lead in drug discovery because of their poor site-specific bioavailability. Concentration of a molecule at target sites is largely varied because of the wavering expression of transporters. Recent anticancer drug discovery strategies are paying high attention to enhance target site bioavailability by modulating drug transporters. The level of genetic expression of transporters is an important determinant to understand their ability to facilitate drug transport across the cellular membrane. Solid carrier (SLC) transporters are the major influx transporters involved in the transportation of most anti-cancer drugs. In contrast, ATP-binding cassette (ABC) superfamily is the most studied class of efflux transporters concerning cancer and is significantly involved in efflux of chemotherapeutics resulting in MDR. Balancing SLC and ABC transporters is essential to avoid therapeutic failure and minimize MDR in chemotherapy. Unfortunately, comprehensive literature on the possible approaches of tailoring site-specific bioavailability of anticancer drugs through transporter modulation is not available till date. This review critically discussed the role of different specific transporter proteins in deciding the intracellular bioavailability of anticancer molecules. Different strategies for reversal of MDR in chemotherapy by incorporation of chemosensitizers have been proposed in this review. Targeted strategies for administration of the chemotherapeutics to the intracellular site of action through clinically relevant transporters employing newer nanotechnology-based formulation platforms have been explained. The discussion embedded in this review is timely considering the current need of addressing the ambiguity observed in pharmacokinetic and clinical outcomes of the chemotherapeutics in anti-cancer treatment regimens.

Phenotypic deconvolution in heterogeneous cancer cell populations using drug-screening data

Tumor heterogeneity is an important driver of treatment failure in cancer since therapies often select for drug-tolerant or drug-resistant cellular subpopulations that drive tumor growth and recurrence. Profiling the drug-response heterogeneity of tumor samples using traditional genomic deconvolution methods has yielded limited results, due in part to the imperfect mapping between genomic variation and functional characteristics. Here, we leverage mechanistic population modeling to develop a statistical framework for profiling phenotypic heterogeneity from standard drug-screen data on bulk tumor samples. This method, called PhenoPop, reliably identifies tumor subpopulations exhibiting differential drug responses and estimates their drug sensitivities and frequencies within the bulk population. We apply PhenoPop to synthetically generated cell populations, mixed cell-line experiments, and multiple myeloma patient samples and demonstrate how it can provide individualized predictions of tumor growth under candidate therapies. This methodology can also be applied to deconvolution problems in a variety of biological settings beyond cancer drug response.

Computational Methods for the Analysis and Prediction of EGFR-Mutated Lung Cancer Drug Resistance: Recent Advances in Drug Design, Challenges and Future Prospects

Lung cancer is a major cause of cancer deaths worldwide, and has a very low survival rate. Non-small cell lung cancer (NSCLC) is the largest subset of lung cancers, which accounts for about 85% of all cases. It has been well established that a mutation in the epidermal growth factor receptor (EGFR) can lead to lung cancer. EGFR Tyrosine Kinase Inhibitors (TKIs) are developed to target the kinase domain of EGFR. These TKIs produce promising results at the initial stage of therapy, but the efficacy becomes limited due to the development of drug resistance. In this paper, we provide a comprehensive overview of computational methods, for understanding drug resistance mechanisms. The important EGFR mutants and the different generations of EGFR-TKIs, with the survival and response rates are discussed. Next, we evaluate the role of important EGFR parameters in drug resistance mechanism, including structural dynamics, hydrogen bonds, stability, dimerization, binding free energies, and signaling pathways. Personalized drug resistance prediction models, drug response curve, drug synergy, and other data-driven methods are also discussed. Recent advancements in deep learning; such as AlphaFold2, deep generative models, big data analytics, and the applications of statistics and permutation are also highlighted. We explore limitations in the current methodologies, and discuss strategies to overcome them. We believe this review will serve as a reference for researchers; to apply computational techniques for precision medicine, analyzing structures of protein-drug complexes, drug discovery, and understanding the drug response and resistance mechanisms in lung cancer patients.

Multiplexed and reproducible high content screening of live and fixed cells using Dye Drop

High-throughput measurement of cells perturbed using libraries of small molecules, gene knockouts, or different microenvironmental factors is a key step in functional genomics and pre-clinical drug discovery. However, it remains difficult to perform accurate single-cell assays in 384-well plates, limiting many studies to well-average measurements (e.g., CellTiter-Glo®). Here we describe a public domain Dye Drop method that uses sequential density displacement and microscopy to perform multi-step assays on living cells. We use Dye Drop cell viability and DNA replication assays followed by immunofluorescence imaging to collect single-cell dose-response data for 67 investigational and clinical-grade small molecules in 58 breast cancer cell lines. By separating the cytostatic and cytotoxic effects of drugs computationally, we uncover unexpected relationships between the two. Dye Drop is rapid, reproducible, customizable, and compatible with manual or automated laboratory equipment. Dye Drop improves the tradeoff between data content and cost, enabling the collection of information-rich perturbagen-response datasets.

CREAMMIST: an integrative probabilistic database for cancer drug response prediction

Extensive in vitro cancer drug screening datasets have enabled scientists to identify biomarkers and develop machine learning models for predicting drug sensitivity. While most advancements have focused on omics profiles, cancer drug sensitivity scores precalculated by the original sources are often used as-is, without consideration for variabilities between studies. It is well-known that significant inconsistencies exist between the drug sensitivity scores across datasets due to differences in experimental setups and preprocessing methods used to obtain the sensitivity scores. As a result, many studies opt to focus only on a single dataset, leading to underutilization of available data and a limited interpretation of cancer pharmacogenomics analysis. To overcome these caveats, we have developed CREAMMIST (https://creammist.mtms.dev), an integrative database that enables users to obtain an integrative dose-response curve, to capture uncertainty (or high certainty when multiple datasets well align) across five widely used cancer cell-line drug-response datasets. We utilized the Bayesian framework to systematically integrate all available dose-response values across datasets (>14 millions dose-response data points). CREAMMIST provides easy-to-use statistics derived from the integrative dose-response curves for various downstream analyses such as identifying biomarkers, selecting drug concentrations for experiments, and training robust machine learning models.

Chemical optimization of siRNA for safe and efficient silencing of placental sFLT1

Preeclampsia (PE) is a rising, potentially lethal complication of pregnancy. PE is driven primarily by the overexpression of placental soluble fms-like tyrosine kinase 1 (sFLT1), a validated diagnostic and prognostic marker of the disease when normalized to placental growth factor (PlGF) levels. Injecting cholesterol-conjugated, fully modified, small interfering RNAs (siRNAs) targeting sFLT1 mRNA into pregnant mice or baboons reduces placental sFLT1 and ameliorates clinical signs of PE, providing a strong foundation for the development of a PE therapeutic. siRNA delivery, potency, and safety are dictated by conjugate chemistry, siRNA duplex structure, and chemical modification pattern. Here, we systematically evaluate these parameters and demonstrate that increasing 2'-O-methyl modifications and 5' chemical stabilization and using sequence-specific duplex asymmetry and a phosphocholine-docosanoic acid conjugate enhance placental accumulation, silencing efficiency and safety of sFLT1-targeting siRNAs. The optimization strategy here provides a framework for the chemical optimization of siRNAs for PE as well as other targets and clinical indications.

Biomarker LEPRE1 induces pelitinib-specific drug responsiveness by regulating ABCG2 expression and tumor transition states in human leukemia and lung cancer

Biomarkers for treatment sensitivity or drug resistance used in precision medicine include prognostic and predictive molecules, critical factors in selecting appropriate treatment protocols and improving survival rates. However, identification of accurate biomarkers remain challenging due to the high risk of false-positive findings and lack of functional validation results for each biomarker. Here, we discovered a mechanical correlation between leucine proline-enriched proteoglycan 1 (LEPRE1) and pelitinib drug sensitivity using in silico statistical methods and confirmed the correlation in acute myeloid leukemia (AML) and A549 lung cancer cells. We determined that high LEPRE1 levels induce protein kinase B activation, overexpression of ATP-binding cassette superfamily G member 2 (ABCG2) and E-cadherin, and cell colonization, resulting in a cancer stem cell-like phenotype. Sensitivity to pelitinib increases in LEPRE1-overexpressing cells due to the reversing effect of ABCG2 upregulation. LEPRE1 silencing induces pelitinib resistance and promotes epithelial-to-mesenchymal transition through actin rearrangement via a series of Src/ERK/cofilin cascades. The in silico results identified a mechanistic relationship between LEPRE1 and pelitinib drug sensitivity, confirmed in two cancer types. This study demonstrates the potential of LEPRE1 as a biomarker in cancer through in-silico prediction and in vitro experiments supporting the clinical development of personalized medicine strategies based on bioinformatics findings.

Drug sensitivity prediction from cell line-based pharmacogenomics data: guidelines for developing machine learning models

The goal of precision oncology is to tailor treatment for patients individually using the genomic profile of their tumors. Pharmacogenomics datasets such as cancer cell lines are among the most valuable resources for drug sensitivity prediction, a crucial task of precision oncology. Machine learning methods have been employed to predict drug sensitivity based on the multiple omics data available for large panels of cancer cell lines. However, there are no comprehensive guidelines on how to properly train and validate such machine learning models for drug sensitivity prediction. In this paper, we introduce a set of guidelines for different aspects of training gene expression-based predictors using cell line datasets. These guidelines provide extensive analysis of the generalization of drug sensitivity predictors and challenge many current practices in the community including the choice of training dataset and measure of drug sensitivity. The application of these guidelines in future studies will enable the development of more robust preclinical biomarkers.

Multi-project and Multi-profile joint Non-negative Matrix Factorization for cancer omic datasets

MOTIVATION

The integration of multi-omic data using machine learning methods has been focused on solving relevant tasks such as predicting sensitivity to a drug or subtyping patients. Recent integration methods, such as joint Non-negative Matrix Factorization (jNMF), have allowed researchers to exploit the information in the data to unravel the biological processes of multi-omic datasets.

RESULTS

We present a novel method called Multi-project and Multi-profile joint Non-negative Matrix Factorization (M&M-jNMF) capable of integrating data from different sources, such as experimental and observational multi-omic data. The method can generate co-clusters between observations, predict profiles and relate latent variables. We applied the method to integrate low-grade glioma omic profiles from The Cancer Genome Atlas (TCGA) and Cell Line Encyclopedia (CCLE) projects. The method allowed us to find gene clusters mainly enriched in cancer-associated terms. We identified groups of patients and cell lines similar to each other by comparing biological processes. We predicted the drug profile for patients, and we identified genetic signatures for resistant and sensitive tumors to a specific drug.

AVAILABILITY AND IMPLEMENTATION

Source code repository is publicly available at https://bitbucket.org/dsalazarb/mmjnmf/ - Zenodo DOI: 10.5281/zenodo.5150920.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Annexin A1 Expression Is Associated with Epithelial-Mesenchymal Transition (EMT), Cell Proliferation, Prognosis, and Drug Response in Pancreatic Cancer

Annexin A1 (ANXA1) is a calcium-dependent phospholipid-binding protein overexpressed in pancreatic cancer (PC). ANXA1 expression has been shown to take part in a wide variety of cancer biology, including carcinogenesis, cell proliferation, invasion, apoptosis, and metastasis, in addition to the initially identified anti-inflammatory effect in experimental settings. We hypothesized that ANXA1 expression is associated with cell proliferation and survival in PC patients. To test this hypothesis, we analyzed 239 PC patients in The Cancer Genome Atlas (TCGA) and GSE57495 cohorts. ANXA1 expression correlated with epithelial-mesenchymal transition (EMT) but weakly with angiogenesis in PC patients. ANXA1-high PC was significantly associated with a high fraction of fibroblasts and keratinocytes in the tumor microenvironment. ANXA1 high PC enriched multiple malignant gene sets, including hypoxia, tumor necrosis factor (TNF)-α signaling via nuclear factor-kappa B (NF-kB), and MTORC1, as well as apoptosis, protein secretion, glycolysis, and the androgen response gene sets consistently in both cohorts. ANXA1 expression was associated with TP53 mutation alone but associated with all KRAS, p53, E2F, and transforming growth factor (TGF)-β signaling pathways and also associated with homologous recombination deficiency in the TCGA cohort. ANXA1 high PC was associated with a high infiltration of T-helper type 2 cells in the TME, with advanced histological grade and MKI67 expression, as well as with a worse prognosis regardless of the grade. ANXA1 expression correlated with a sensitivity to gemcitabine, doxorubicin, and 5-fluorouracil in PC cell lines. In conclusion, ANXA1 expression is associated with EMT, cell proliferation, survival, and the drug response in PC.

Differential Splicing of Skipped-exons Predicts Drug Response in Cancer Cell Lines

Alternative splicing of pre-mRNA transcripts is an important regulatory mechanism that increases the diversity of gene products in eukaryotes. Various studies have linked specific transcript isoforms to altered drug response in cancer; however, few algorithms have incorporated splicing information into drug response prediction. In this study, we evaluated whether basal-level splicing information could be used to predict drug sensitivity by constructing doxorubicin-sensitivity classification models with splicing and expression data. We detailed splicing differences between sensitive and resistant cell lines by implementing quasi-binomial generalized linear modeling (QBGLM) and found altered inclusion of 277 skipped exons. We additionally conducted RNA-binding protein (RBP) binding motif enrichment and differential expression analysis to characterize cis- and trans-acting elements that potentially influence doxorubicin response-mediating splicing alterations. Our results showed that a classification model built with skipped exon data exhibited strong predictive power. We discovered an association between differentially spliced events and epithelial-mesenchymal transition (EMT) and observed motif enrichment, as well as differential expression of RBFOX and ELAVL RBP family members. Our work demonstrates the potential of incorporating splicing data into drug response algorithms and the utility of a QBGLM approach for fast, scalable identification of relevant splicing differences between large groups of samples.

Explainable drug sensitivity prediction through cancer pathway enrichment

Computational approaches to predict drug sensitivity can promote precision anticancer therapeutics. Generalizable and explainable models are of critical importance for translation to guide personalized treatment and are often overlooked in favor of prediction performance. Here, we propose PathDSP: a pathway-based model for drug sensitivity prediction that integrates chemical structure information with enrichment of cancer signaling pathways across drug-associated genes, gene expression, mutation and copy number variation data to predict drug response on the Genomics of Drug Sensitivity in Cancer dataset. Using a deep neural network, we outperform state-of-the-art deep learning models, while demonstrating good generalizability a separate dataset of the Cancer Cell Line Encyclopedia as well as provide explainable results, demonstrated through case studies that are in line with current knowledge. Additionally, our pathway-based model achieved a good performance when predicting unseen drugs and cells, with potential utility for drug development and for guiding individualized medicine.

Organotypic Models in Drug Development "Tumor Models and Cancer Systems Biology for the Investigation of Anticancer Drugs and Resistance Development"

The landscape of cancer treatment has improved over the past decades, aiming to reduce systemic toxicity and enhance compatibility with the quality of life of the patient. However, at the therapeutic level, metastatic cancer remains hugely challenging, based on the almost inevitable emergence of therapy resistance. A small subpopulation of cells able to survive drug treatment termed the minimal residual disease may either harbor resistance-associated mutations or be phenotypically resistant, allowing them to regrow and become the dominant population in the therapy-resistant tumor. Characterization of the profile of minimal residual disease represents the key to the identification of resistance drivers that underpin cancer evolution. Therapeutic regimens must, therefore, be dynamic and tailored to take into account the emergence of resistance as tumors evolve within a complex microenvironment in vivo. This requires the adoption of new technologies based on the culture of cancer cells in ways that more accurately reflect the intratumor microenvironment, and their analysis using omics and system-based technologies to enable a new era in the diagnostics, classification, and treatment of many cancer types by applying the concept "from the cell plate to the patient." In this chapter, we will present and discuss 3D model building and use, and provide comprehensive information on new genomic techniques that are increasing our understanding of drug action and the emergence of resistance.

Evaluation of publicly available in vitro drug sensitivity models for ovarian and uterine cancer

OBJECTIVE

Publicly available data on drug sensitivity for cancer cell lines have been curated into a single, integrated database, PharmacoDB. The contributing datasets report modeled estimates of drug effect from high throughput assays. These databases have been informative for developing new broad insights, but the reliability of these data specifically for drugs used to treat ovarian and uterine cancers in related cell lines has not been reported.

METHODS

In vitro viability assays were performed on A2780, OVCAR-3, TOV-21G, and RL95-2 cells with nine drugs to produce high resolution exposure-response curves. Lab generated data were compared to publicly available datasets by IC₂₀, IC₅₀, and IC₈₀ values, and the area between the logarithmic logistic regression curves.

RESULTS

For exposure-response curve comparisons with clinically indicated drugs between lab generated and publicly available data, the majority had area-between-curves less than 20%, indicating similarity. However, 15 out of 40 of these dataset curves were incomplete as indicated by the lack of, or extrapolated, IC₅₀ value. The common ovarian and uterine cancer drug, carboplatin, exemplified this incomplete status as all of the available dataset curves were incomplete and therefore non-informative.

CONCLUSIONS

For gynecologic malignancy cell line models, experimental drug sensitivity data is comparable to the available data in PharmacoDB when exposure-response curves are complete. Incomplete exposure-response curves due to incomplete concentration ranges tested and related extrapolation of IC values can mislead individual drug/cell line pair data for downstream applications.

Predicting Drug Response and Synergy Using a Deep Learning Model of Human Cancer Cells

Most drugs entering clinical trials fail, often related to an incomplete understanding of the mechanisms governing drug response. Machine learning techniques hold immense promise for better drug response predictions, but most have not reached clinical practice due to their lack of interpretability and their focus on monotherapies. We address these challenges by developing DrugCell, an interpretable deep learning model of human cancer cells trained on the responses of 1,235 tumor cell lines to 684 drugs. Tumor genotypes induce states in cellular subsystems that are integrated with drug structure to predict response to therapy and, simultaneously, learn biological mechanisms underlying the drug response. DrugCell predictions are accurate in cell lines and also stratify clinical outcomes. Analysis of DrugCell mechanisms leads directly to the design of synergistic drug combinations, which we validate systematically by combinatorial CRISPR, drug-drug screening in vitro, and patient-derived xenografts. DrugCell provides a blueprint for constructing interpretable models for predictive medicine.

A guide for bioinformaticians: 'omics-based drug discovery for precision oncology

Bioinformatics-centric drug development is inevitable in the era of precision medicine. Clinical 'omics information, including genomics, epigenomics, transcriptomics, and proteomics, provides the most comprehensive molecular landscape in which each patient's pathological history is delineated. Hence, the capability of bioinformaticians to manage integrative 'omics data is crucial to current drug development. Bioinformatics can accelerate drug development from initial time-consuming discoveries to the clinical stage by providing information-guided solutions. However, many bioinformaticians do not have opportunities to participate in drug discovery programs. As a starting point for bioinformaticians with no prior drug development experience, here we discuss bioinformatics applications during drug development with a focus on working-level omics-based methodologies.

Evaluating the consistency of large-scale pharmacogenomic studies

Recent years have seen an increase in the availability of pharmacogenomic databases such as Genomics of Drug Sensitivity in Cancer (GDSC) and Cancer Cell Line Encyclopedia (CCLE) that provide genomic and functional characterization information for multiple cell lines. Studies have alluded to the fact that specific characterizations may be inconsistent between different databases. Analysis of the potential discrepancies in the different databases is highly significant, as these sources are frequently used to analyze and validate methodologies for personalized cancer therapies. In this article, we review the recent developments in investigating the correspondence between different pharmacogenomics databases and discuss the potential factors that require attention when incorporating these sources in any modeling analysis. Furthermore, we explored the consistency among these databases using copulas that can capture nonlinear dependencies between two sets of data.

Optimum concentration-response curve metrics for supervised selection of discriminative cellular phenotypic endpoints for chemical hazard assessment

High-content imaging (HCI) provides quantitative and information-rich measurements of chemical effects on human in vitro cell models. Identification of discriminative phenotypic endpoints from cellular features obtained from HCI is required for accurate assessments of potential chemical hazards. However, the use of suboptimal metrics to quantify the concentration-response curves (CRC) of chemicals based on these features may obscure discriminative features, and lead to non-predictive endpoints and poor chemical classifications or hazard assessments. Here, we present a systematic and data-driven study on the performances of different CRC metrics in identifying image-based phenotypic features that can accurately classify the effects of reference chemicals with known in vivo toxicities. We studied four previous HCI in vitro nephro- or pulmono-toxicity datasets, which contain phenotypic feature measurements from different cell and feature types. Within a feature type, we found that efficacy metrics at higher chemical concentrations tend to give higher classification accuracy, whereas potency metrics do not have obvious trends across different response levels. Across different cell and feature types, efficacy metrics generally gave higher classification accuracy than potency metrics and area under the curve (AUC). Our results suggest that efficacy metrics, especially at higher concentrations, are more likely to help us to identify discriminative phenotypic endpoints. Therefore, HCI experiments for toxicological applications should include measurements at sufficiently high chemical concentrations, and efficacy metrics should always be analyzed. The identified features may be used as specific toxicity endpoints for further chemical hazard assessment.

Gene regulatory network analysis with drug sensitivity reveals synergistic effects of combinatory chemotherapy in gastric cancer

The combination of docetaxel, cisplatin, and fluorouracil (DCF) is highly synergistic in advanced gastric cancer. We aimed to explain these synergistic effects at the molecular level. Thus, we constructed a weighted correlation network using the differentially expressed genes between Stage I and IV gastric cancer based on The Cancer Genome Atlas (TCGA), and three modules were derived. Next, we investigated the correlation between the eigengene of the expression of the gene network modules and the chemotherapeutic drug response to DCF from the Genomics of Drug Sensitivity in Cancer (GDSC) database. The three modules were associated with functions related to cell migration, angiogenesis, and the immune response. The eigengenes of the three modules had a high correlation with DCF (−0.41, −0.40, and −0.15). The eigengenes of the three modules tended to increase as the stage increased. Advanced gastric cancer was affected by the interaction the among modules with three functions, namely cell migration, angiogenesis, and the immune response, all of which are related to metastasis. The weighted correlation network analysis model proved the complementary effects of DCF at the molecular level and thus, could be used as a unique methodology to determine the optimal combination of chemotherapy drugs for patients with gastric cancer.

Tumor cell sensitivity to vemurafenib can be predicted from protein expression in a BRAF-V600E basket trial setting

BACKGROUND

Genetics-based basket trials have emerged to test targeted therapeutics across multiple cancer types. However, while vemurafenib is FDA-approved for BRAF-V600E melanomas, the non-melanoma basket trial was unsuccessful, suggesting mutation status is insufficient to predict response. We hypothesized that proteomic data would complement mutation status to identify vemurafenib-sensitive tumors and effective co-treatments for BRAF-V600E tumors with inherent resistance.

METHODS

Reverse Phase Proteomic Array (RPPA, MD Anderson Cell Lines Project), RNAseq (Cancer Cell Line Encyclopedia) and vemurafenib sensitivity (Cancer Therapeutic Response Portal) data for BRAF-V600E cancer cell lines were curated. Linear and nonlinear regression models using RPPA protein or RNAseq were evaluated and compared based on their ability to predict BRAF-V600E cell line sensitivity (area under the dose response curve). Accuracies of all models were evaluated using hold-out testing. CausalPath software was used to identify protein-protein interaction networks that could explain differential protein expression in resistant cells. Human examination of features employed by the model, the identified protein interaction networks, and model simulation suggested anti-ErbB co-therapy would counter intrinsic resistance to vemurafenib. To validate this potential co-therapy, cell lines were treated with vemurafenib and dacomitinib (a pan-ErbB inhibitor) and the number of viable cells was measured.

RESULTS

Orthogonal partial least squares (O-PLS) predicted vemurafenib sensitivity with greater accuracy in both melanoma and non-melanoma BRAF-V600E cell lines than other leading machine learning methods, specifically Random Forests, Support Vector Regression (linear and quadratic kernels) and LASSO-penalized regression. Additionally, use of transcriptomic in place of proteomic data weakened model performance. Model analysis revealed that resistant lines had elevated expression and activation of ErbB receptors, suggesting ErbB inhibition could improve vemurafenib response. As predicted, experimental evaluation of vemurafenib plus dacomitinb demonstrated improved efficacy relative to monotherapies.

CONCLUSIONS

Combined, our results support that inclusion of proteomics can predict drug response and identify co-therapies in a basket setting.

Expression Levels of Therapeutic Targets as Indicators of Sensitivity to Targeted Therapeutics

Cancer precision medicine aims to predict the drug likely to yield the best response for a patient. Genomic sequencing of tumors is currently being used to better inform treatment options; however, this approach has had a limited clinical impact due to the paucity of actionable mutations. An alternative to mutation status is the use of gene expression signatures to predict response. Using data from two large-scale studies, The Genomics of Drug Sensitivity of Cancer (GDSC) and The Cancer Therapeutics Response Portal (CTRP), we investigated the relationship between the sensitivity of hundreds of cell lines to hundreds of drugs, and the relative expression levels of the targets these drugs are directed against. For approximately one third of the drugs considered (73/222 in GDSC and 131/360 in CTRP), sensitivity was significantly correlated with the expression of at least one of the known targets. Surprisingly, for 8% of the annotated targets, there was a significant anticorrelation between target expression and sensitivity. For several cases, this corresponded to drugs targeting multiple genes in the same family, with the expression of one target significantly correlated with sensitivity and another significantly anticorrelated suggesting a possible role in resistance. Furthermore, we identified nontarget genes that are significantly correlated or anticorrelated with drug sensitivity, and find literature linking several to sensitization and resistance. Our analyses provide novel and important insights into both potential mechanisms of resistance and relative efficacy of drugs against the same target.

More than fishing for a cure: The promises and pitfalls of high throughput cancer cell line screens

High-throughput screens in cancer cell lines (CCLs) have been used for decades to help researchers identify compounds with the potential to improve the treatment of cancer and, more recently, to identify genomic susceptibilities in cancer via genome-wide shRNA and CRISPR/Cas9 screens. Additionally, rich genomic and transcriptomic data of these CCLs has allowed researchers to pair this screening data with biological features, enabling efforts to identify biomarkers of treatment response and gene dependencies. In this paper, we review the major CCL screening efforts and the large datasets these screens have made available. We also assess the CCL screens collectively and include a resource with harmonized CCL and compound identifiers to facilitate comparisons across screens. The CCLs in these screens were found to represent a wide range of cancer types, with a strong correlation between the representation of a cancer type and its associated mortality. Patient ages and gender distributions of CCLs were generally as expected, with some notable exceptions of female underrepresentation in certain disease types. Also, ethnicity information, while largely incomplete, suggests that African American and Hispanic patients may be severely underrepresented in these screens. Nearly all genes were targeted in the genetic perturbations screens, but the compounds used for the drug screens target less than half of known cancer drivers, likely reflecting known limitations in our drug design capabilities. Finally, we discuss recent developments in the field and the promise they hold for enabling future screens to overcome previous limitations and lead to new breakthroughs in cancer treatment.

Dissecting RAF Inhibitor Resistance by Structure-based Modeling Reveals Ways to Overcome Oncogenic RAS Signaling

Clinically used RAF inhibitors are ineffective in RAS mutant tumors because they enhance homo- and heterodimerization of RAF kinases, leading to paradoxical activation of ERK signaling. Overcoming enhanced RAF dimerization and the resulting resistance is a challenge for drug design. Combining multiple inhibitors could be more effective, but it is unclear how the best combinations can be chosen. We built a next-generation mechanistic dynamic model to analyze combinations of structurally different RAF inhibitors, which can efficiently suppress MEK/ERK signaling. This rule-based model of the RAS/ERK pathway integrates thermodynamics and kinetics of drug-protein interactions, structural elements, posttranslational modifications, and cell mutational status as model rules to predict RAF inhibitor combinations for inhibiting ERK activity in oncogenic RAS and/or BRAFV600E backgrounds. Predicted synergistic inhibition of ERK signaling was corroborated by experiments in mutant NRAS, HRAS, and BRAFV600E cells, and inhibition of oncogenic RAS signaling was associated with reduced cell proliferation and colony formation.

Genetic load makes cancer cells more sensitive to common drugs: evidence from Cancer Cell Line Encyclopedia

Genetic alterations initiate tumors and enable the evolution of drug resistance. The pro-cancer view of mutations is however incomplete, and several studies show that mutational load can reduce tumor fitness. Given its negative effect, genetic load should make tumors more sensitive to anticancer drugs. Here, we test this hypothesis across all major types of cancer from the Cancer Cell Line Encyclopedia, which provides genetic and expression data of 496 cell lines together with their response to 24 common anticancer drugs. We found that the efficacy of 9 out of 24 drugs showed significant association with genetic load in a pan-cancer analysis. The associations for some tissue-drug combinations were remarkably strong, with genetic load explaining up to 83% of the variance in the drug response. Overall, the role of genetic load depended on both the drug and the tissue type with 10 tissues being particularly vulnerable to genetic load. We also identified changes in gene expression associated with increased genetic load, which included cell-cycle checkpoints, DNA damage and apoptosis. Our results show that genetic load is an important component of tumor fitness and can predict drug sensitivity. Beyond being a biomarker, genetic load might be a new, unexplored vulnerability of cancer.

Revisiting inconsistency in large pharmacogenomic studies

In 2013, we published a comparative analysis of mutation and gene expression profiles and drug sensitivity measurements for 15 drugs characterized in the 471 cancer cell lines screened in the Genomics of Drug Sensitivity in Cancer (GDSC) and Cancer Cell Line Encyclopedia (CCLE). While we found good concordance in gene expression profiles, there was substantial inconsistency in the drug responses reported by the GDSC and CCLE projects. We received extensive feedback on the comparisons that we performed. This feedback, along with the release of new data, prompted us to revisit our initial analysis. We present a new analysis using these expanded data, where we address the most significant suggestions for improvements on our published analysis - that targeted therapies and broad cytotoxic drugs should have been treated differently in assessing consistency, that consistency of both molecular profiles and drug sensitivity measurements should be compared across cell lines, and that the software analysis tools provided should have been easier to run, particularly as the GDSC and CCLE released additional data. Our re-analysis supports our previous finding that gene expression data are significantly more consistent than drug sensitivity measurements. Using new statistics to assess data consistency allowed identification of two broad effect drugs and three targeted drugs with moderate to good consistency in drug sensitivity data between GDSC and CCLE. For three other targeted drugs, there were not enough sensitive cell lines to assess the consistency of the pharmacological profiles. We found evidence of inconsistencies in pharmacological phenotypes for the remaining eight drugs. Overall, our findings suggest that the drug sensitivity data in GDSC and CCLE continue to present challenges for robust biomarker discovery. This re-analysis provides additional support for the argument that experimental standardization and validation of pharmacogenomic response will be necessary to advance the broad use of large pharmacogenomic screens.

Revisiting inconsistency in large pharmacogenomic studies

In 2013, we published a comparative analysis of mutation and gene expression profiles and drug sensitivity measurements for 15 drugs characterized in the 471 cancer cell lines screened in the Genomics of Drug Sensitivity in Cancer (GDSC) and Cancer Cell Line Encyclopedia (CCLE). While we found good concordance in gene expression profiles, there was substantial inconsistency in the drug responses reported by the GDSC and CCLE projects. We received extensive feedback on the comparisons that we performed. This feedback, along with the release of new data, prompted us to revisit our initial analysis. We present a new analysis using these expanded data, where we address the most significant suggestions for improvements on our published analysis - that targeted therapies and broad cytotoxic drugs should have been treated differently in assessing consistency, that consistency of both molecular profiles and drug sensitivity measurements should be compared across cell lines, and that the software analysis tools provided should have been easier to run, particularly as the GDSC and CCLE released additional data. Our re-analysis supports our previous finding that gene expression data are significantly more consistent than drug sensitivity measurements. Using new statistics to assess data consistency allowed identification of two broad effect drugs and three targeted drugs with moderate to good consistency in drug sensitivity data between GDSC and CCLE. For three other targeted drugs, there were not enough sensitive cell lines to assess the consistency of the pharmacological profiles. We found evidence of inconsistencies in pharmacological phenotypes for the remaining eight drugs. Overall, our findings suggest that the drug sensitivity data in GDSC and CCLE continue to present challenges for robust biomarker discovery. This re-analysis provides additional support for the argument that experimental standardization and validation of pharmacogenomic response will be necessary to advance the broad use of large pharmacogenomic screens.

Revisiting inconsistency in large pharmacogenomic studies

In 2013, we published a comparative analysis mutation and gene expression profiles and drug sensitivity measurements for 15 drugs characterized in the 471 cancer cell lines screened in the Genomics of Drug Sensitivity in Cancer (GDSC) and Cancer Cell Line Encyclopedia (CCLE). While we found good concordance in gene expression profiles, there was substantial inconsistency in the drug responses reported by the GDSC and CCLE projects. We received extensive feedback on the comparisons that we performed. This feedback, along with the release of new data, prompted us to revisit our initial analysis. Here we present a new analysis using these expanded data in which we address the most significant suggestions for improvements on our published analysis - that targeted therapies and broad cytotoxic drugs should have been treated differently in assessing consistency, that consistency of both molecular profiles and drug sensitivity measurements should both be compared across cell lines, and that the software analysis tools we provided should have been easier to run, particularly as the GDSC and CCLE released additional data.             Our re-analysis supports our previous finding that gene expression data are significantly more consistent than drug sensitivity measurements. The use of new statistics to assess data consistency allowed us to identify two broad effect drugs and three targeted drugs with moderate to good consistency in drug sensitivity data between GDSC and CCLE. For three other targeted drugs, there were not enough sensitive cell lines to assess the consistency of the pharmacological profiles. We found evidence of inconsistencies in pharmacological phenotypes for the remaining eight drugs.             Overall, our findings suggest that the drug sensitivity data in GDSC and CCLE continue to present challenges for robust biomarker discovery. This re-analysis provides additional support for the argument that experimental standardization and validation of pharmacogenomic response will be necessary to advance the broad use of large pharmacogenomic screens.