Text-mining-based feature selection for anticancer drug response prediction

Motivation

Predicting anticancer treatment response from baseline genomic data is a critical obstacle in personalized medicine. Machine learning methods are commonly used for predicting drug response from gene expression data. In the process of constructing these machine learning models, one of the most significant challenges is identifying appropriate features among a massive number of genes.

Results

In this study, we utilize features (genes) extracted using the text-mining of scientific literatures. Using two independent cancer pharmacogenomic datasets, we demonstrate that text-mining-based features outperform traditional feature selection techniques in machine learning tasks. In addition, our analysis reveals that text-mining feature-based machine learning models trained on in vitro data also perform well when predicting the response of in vivo cancer models. Our results demonstrate that text-mining-based feature selection is an easy to implement approach that is suitable for building machine learning models for anticancer drug response prediction.

Availability and implementation

https://github.com/merlab/text_features.

Spatiotemporal modeling of chemoresistance evolution in breast tumors uncovers dependencies on SLC38A7 and SLC46A1

In solid tumors, drug concentrations decrease with distance from blood vessels. However, cellular adaptations accompanying the gradated exposure of cancer cells to drugs are largely unknown. Here, we modeled the spatiotemporal changes promoting chemotherapy resistance in breast cancer. Using pairwise cell competition assays at each step during the acquisition of chemoresistance, we reveal an important priming phase that renders cancer cells previously exposed to sublethal drug concentrations refractory to dose escalation. Therapy-resistant cells throughout the concentration gradient display higher expression of the solute carriers SLC38A7 and SLC46A1 and elevated intracellular concentrations of their associated metabolites. Reduced levels of SLC38A7 and SLC46A1 diminish the proliferative potential of cancer cells, and elevated expression of these SLCs in breast tumors from patients correlates with reduced survival. Our work provides mechanistic evidence to support dose-intensive treatment modalities for patients with solid tumors and reveals two members of the SLC family as potential actionable targets.

Bimodal gene expression in cancer patients provides interpretable biomarkers for drug sensitivity

Identifying biomarkers predictive of cancer cell response to drug treatment constitutes one of the main challenges in precision oncology. Recent large-scale cancer pharmacogenomic studies have opened new avenues of research to develop predictive biomarkers by profiling thousands of human cancer cell lines at the molecular level and screening them with hundreds of approved drugs and experimental chemical compounds. Many studies have leveraged these data to build predictive models of response using various statistical and machine learning methods. However, a common pitfall to these methods is the lack of interpretability as to how they make predictions, hindering the clinical translation of these models. To alleviate this issue, we used the recent logic modeling approach to develop a new machine learning pipeline that explores the space of bimodally expressed genes in multiple large in vitro pharmacogenomic studies and builds multivariate, nonlinear, yet interpretable logic-based models predictive of drug response. The performance of this approach was showcased in a compendium of the three largest in vitro pharmacogenomic data sets to build robust and interpretable models for 101 drugs that span 17 drug classes with high validation rates in independent datasets. These results along with in vivo and clinical validation, support a better translation of gene expression biomarkers between model systems using bimodal gene expression.

Ontogeny and Vulnerabilities of Drug-Tolerant Persisters in HER2+ Breast Cancer

Resistance to targeted therapies is an important clinical problem in HER2-positive (HER2+) breast cancer. "Drug-tolerant persisters" (DTP), a subpopulation of cancer cells that survive via reversible, nongenetic mechanisms, are implicated in resistance to tyrosine kinase inhibitors (TKI) in other malignancies, but DTPs following HER2 TKI exposure have not been well characterized. We found that HER2 TKIs evoke DTPs with a luminal-like or a mesenchymal-like transcriptome. Lentiviral barcoding/single-cell RNA sequencing reveals that HER2+ breast cancer cells cycle stochastically through a "pre-DTP" state, characterized by a G0-like expression signature and enriched for diapause and/or senescence genes. Trajectory analysis/cell sorting shows that pre-DTPs preferentially yield DTPs upon HER2 TKI exposure. Cells with similar transcriptomes are present in HER2+ breast tumors and are associated with poor TKI response. Finally, biochemical experiments indicate that luminal-like DTPs survive via estrogen receptor-dependent induction of SGK3, leading to rewiring of the PI3K/AKT/mTORC1 pathway to enable AKT-independent mTORC1 activation.

SIGNIFICANCE

DTPs are implicated in resistance to anticancer therapies, but their ontogeny and vulnerabilities remain unclear. We find that HER2 TKI-DTPs emerge from stochastically arising primed cells ("pre-DTPs") that engage either of two distinct transcriptional programs upon TKI exposure. Our results provide new insights into DTP ontogeny and potential therapeutic vulnerabilities. This article is highlighted in the In This Issue feature, p. 873.

PharmacoDB 2.0: improving scalability and transparency of in vitro pharmacogenomics analysis

Cancer pharmacogenomics studies provide valuable insights into disease progression and associations between genomic features and drug response. PharmacoDB integrates multiple cancer pharmacogenomics datasets profiling approved and investigational drugs across cell lines from diverse tissue types. The web-application enables users to efficiently navigate across datasets, view and compare drug dose-response data for a specific drug-cell line pair. In the new version of PharmacoDB (version 2.0, https://pharmacodb.ca/), we present (i) new datasets such as NCI-60, the Profiling Relative Inhibition Simultaneously in Mixtures (PRISM) dataset, as well as updated data from the Genomics of Drug Sensitivity in Cancer (GDSC) and the Genentech Cell Line Screening Initiative (gCSI); (ii) implementation of FAIR data pipelines using ORCESTRA and PharmacoDI; (iii) enhancements to drug-response analysis such as tissue distribution of dose-response metrics and biomarker analysis; and (iv) improved connectivity to drug and cell line databases in the community. The web interface has been rewritten using a modern technology stack to ensure scalability and standardization to accommodate growing pharmacogenomics datasets. PharmacoDB 2.0 is a valuable tool for mining pharmacogenomics datasets, comparing and assessing drug-response phenotypes of cancer models.

Assessing therapy response in patient-derived xenografts

[Figure: see text].

Novel subtypes of NPM1-mutated AML with distinct outcome

Acute myeloid leukemia (AML) is heterogeneous with one common subtype recognized by the presence of recurrent mutation of nucleophosmin-1 (NPM1). Emerging evidence indicates that within NPM1 mutated AML there is variation in outcome which challenges how best to characterize and treat the individual patient. Our recent findings show that there are two distinct (primitive and committed) subtypes within NPM1 mutated AML patients. These subtypes exhibit specific molecular characteristics, disease differentiation states, patient survival, and differential drug responses.

Proteogenomic characterization of pancreatic ductal adenocarcinoma

Pancreatic ductal adenocarcinoma (PDAC) is a highly aggressive cancer with poor patient survival. Toward understanding the underlying molecular alterations that drive PDAC oncogenesis, we conducted comprehensive proteogenomic analysis of 140 pancreatic cancers, 67 normal adjacent tissues, and 9 normal pancreatic ductal tissues. Proteomic, phosphoproteomic, and glycoproteomic analyses were used to characterize proteins and their modifications. In addition, whole-genome sequencing, whole-exome sequencing, methylation, RNA sequencing (RNA-seq), and microRNA sequencing (miRNA-seq) were performed on the same tissues to facilitate an integrated proteogenomic analysis and determine the impact of genomic alterations on protein expression, signaling pathways, and post-translational modifications. To ensure robust downstream analyses, tumor neoplastic cellularity was assessed via multiple orthogonal strategies using molecular features and verified via pathological estimation of tumor cellularity based on histological review. This integrated proteogenomic characterization of PDAC will serve as a valuable resource for the community, paving the way for early detection and identification of novel therapeutic targets.

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.

Highlights from the 16th International Society for Computational Biology Student Council Symposium 2020

In this meeting overview, we summarise the scientific program and organisation of the 16th International Society for Computational Biology Student Council Symposium in 2020 (ISCB SCS2020). This symposium was the first virtual edition in an uninterrupted series of symposia that has been going on for 15 years, aiming to unite computational biology students and early career researchers across the globe.

Abstract PO-070: Bimodality of gene expression in cancer patient tumors as interpretable biomarkers for drug sensitivity

Identifying biomarkers predictive of cancer cells’ response to drug treatment constitutes one of the main challenges in precision oncology. Recent large-scale cancer pharmacogenomic studies have boosted the research for finding predictive biomarkers by profiling thousands of human cancer cell lines at the molecular level and screening them with hundreds of approved drugs and experimental chemical compounds. Many studies have leveraged these data to build predictive models of response using various statistical and machine learning methods. However, a common challenge in these methods is the lack of interpretability as to how they make the predictions and which features were the most associated with response, hindering the clinical translation of these models. To alleviate this issue, we develop a new machine learning pipeline based on the recent LOBICO approach that explores the space of bimodally expressed genes in multiple large in vitro pharmacogenomic studies and builds multivariate, nonlinear, yet interpretable logic-based models predictive of drug response. Using our method, we used a compendium of three of the largest pharmacogenomic data sets to build robust and interpretable models for 101 drugs that span 17 drug classes with high validation rate in independent datasets.

Citation Format: Wail Ba-alawi, Sisira Kadambat Nair, Bo Li, Anthony Mammoliti, Petr Smirnov, Arvind Singh Mer, Linda Penn, Benjamin Haibe-Kains. Bimodality of gene expression in cancer patient tumors as interpretable biomarkers for drug sensitivity [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-070.

Abstract PO-052: Exploring patient derived xenografts based pharmacogenomic data for precision oncology

Preclinical cancer models play a vital role in oncology research and precision medicine. Patient-derived tumor xenografts (PDXs) are used as reliable preclinical models for studying tumor biology and for testing anti-cancer therapies that are tailored according to genomic characteristics of tumors. Several academic groups, research institutes, and commercial organizations are generating and distributing PDX models. However the distributed nature of PDX model generation and lack of central repository make it challenging to find PDX models with specific characteristics. Furthermore this also hinders meta-analysis (across datasets) of PDX pharmacogenomic data. International consortia and catalogs of PDX models such as PDXNet, EurOPDX and PDXFinder are being developed to standardize PDX associated metadata and facilitate material sharing. Recently we have developed Xenograft Visualization & Analysis (Xeva), an open-source software package in R programming language. Xeva allows PDX growth curve visualization, different response metrics computation and biomarker discovery. Extending to this we have developed XevaDB, a database of PDX drug response and genomic profiles. XevaDB is the first resource to allow concurrent visualization of drug response and associated molecular data such as mutation and copy number alterations. Furthermore XevaDB enables exploration of the tumor growth curve of a PDX model, along with corresponding control. XevaDB contains PDXs from >600 individual patients, spanning across nine different tissue types and >70 drugs. Using XevaDB, we have performed meta-analysis of PDX pharmacogenomic data and have identified 90 pathways significantly associated with response to 53 drugs (FDR < 5%). Our results show that activity of the EGFR signaling pathway is significantly associated with erlotinib response in lung cancer PDXs. We have also found that in PDXs, response to binimetinib is associated with the MAP kinase activation pathway. XevaDB provides a comprehensive resource to search and explore PDX pharmacogenomic data. By combining drug response with genomic data of PDXs, XevaDB allows researchers to quickly find the model of interest and access the data to answer their biological questions. As PDXs based pharmacogenomic datasets continue to expand, XevaDB will facilitate easy access and analysis of this valuable data by the scientific community.

Citation Format: Arvind Singh Mer, Benjamin Haibe-Kains. Exploring patient derived xenografts based pharmacogenomic data for precision oncology [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-052.

Colorectal Cancer Cells Enter a Diapause-like DTP State to Survive Chemotherapy

Cancer cells enter a reversible drug-tolerant persister (DTP) state to evade death from chemotherapy and targeted agents. It is increasingly appreciated that DTPs are important drivers of therapy failure and tumor relapse. We combined cellular barcoding and mathematical modeling in patient-derived colorectal cancer models to identify and characterize DTPs in response to chemotherapy. Barcode analysis revealed no loss of clonal complexity of tumors that entered the DTP state and recurred following treatment cessation. Our data fit a mathematical model where all cancer cells, and not a small subpopulation, possess an equipotent capacity to become DTPs. Mechanistically, we determined that DTPs display remarkable transcriptional and functional similarities to diapause, a reversible state of suspended embryonic development triggered by unfavorable environmental conditions. Our study provides insight into how cancer cells use a developmentally conserved mechanism to drive the DTP state, pointing to novel therapeutic opportunities to target DTPs.

Organoid Cultures as Preclinical Models of Non-Small Cell Lung Cancer

PURPOSE

Non-small cell lung cancer (NSCLC) is the most common cause of cancer-related deaths worldwide. There is an unmet need to develop novel clinically relevant models of NSCLC to accelerate identification of drug targets and our understanding of the disease.

EXPERIMENTAL DESIGN

Thirty surgically resected NSCLC primary patient tissue and 35 previously established patient-derived xenograft (PDX) models were processed for organoid culture establishment. Organoids were histologically and molecularly characterized by cytology and histology, exome sequencing, and RNA-sequencing analysis. Tumorigenicity was assessed through subcutaneous injection of organoids in NOD/SCID mice. Organoids were subjected to drug testing using EGFR, FGFR, and MEK-targeted therapies.

RESULTS

We have identified cell culture conditions favoring the establishment of short-term and long-term expansion of NSCLC organoids derived from primary lung patient and PDX tumor tissue. The NSCLC organoids recapitulated the histology of the patient and PDX tumor. They also retained tumorigenicity, as evidenced by cytologic features of malignancy, xenograft formation, preservation of mutations, copy number aberrations, and gene expression profiles between the organoid and matched parental tumor tissue by whole-exome and RNA sequencing. NSCLC organoid models also preserved the sensitivity of the matched parental tumor to targeted therapeutics, and could be used to validate or discover biomarker-drug combinations.

CONCLUSIONS

Our panel of NSCLC organoids closely recapitulates the genomics and biology of patient tumors, and is a potential platform for drug testing and biomarker validation.

Abstract 3378: Systematic pharmacogenomic analysis of large patient derived xenografts data

One of the key challenges in cancer precision medicine is finding robust biomarkers of drug response. Patient-derived tumor xenografts (PDXs) have emerged as reliable preclinical models since they better recapitulate tumor response to chemo- and targeted therapies. However, the lack of standard tools poses a challenge in the analysis of PDXs with molecular and pharmacological profiles. Efficient storage, access and analysis is key to the realization of the full potential of PDX pharmacogenomic data. To address this, we have developed Xeva (XEnograft Visualization & Analysis), an open-source software package for processing, visualization and integrative analysis of a compendium of in vivo pharmacogenomic datasets. The Xeva package follows the PDX minimum information (PDX-MI) standards and can handle both replicate-based and 1x1x1 experimental designs. We used Xeva to characterize the variability of gene expression and pathway activity across passages. We found that only a few genes and pathways have passage specific alterations (median intraclass correlation of 0.53 for genes and positive enrichment score for 92.5% pathways). For example, activity of the mRNA 3'-end processing and elongation arrest and recovery pathways were strongly affected by model passaging (gene set enrichment analysis false discovery rate [FDR] <5%). We then leveraged our platform to link the drug response and the pathways whose activity is consistent across passages by mining the Novartis PDX Encyclopedia (PDXE) data containing 1,075 PDXs spanning 5 tissue types and 62 anticancer drugs. We identified 87 pathways significantly associated with response to 51 drugs (FDR < 5%), including associations such as erlotinib response and signaling by EGFR in cancer pathways and MAP kinase activation in TLR cascade and binimetinib response. Among the significant pathway-drug associations, we found novel biomarkers based on gene expressions, Copy Number Aberrations (CNAs) and mutations predictive of drug response (concordance index > 0.60; FDR < 0.05). Xeva provides a flexible platform for integrative analysis of preclinical in vivo pharmacogenomics data to identify biomarkers predictive of drug response, a major step toward precision oncology.

Citation Format: Arvind Singh Mer, Wail Ba-alawi, Petr Smirnov, Yi Xiao Wang, Ben Brew, Ming-Sound Tsao, David Cescon, Anna Goldenberg, Benjamin Haibe-Kains. Systematic pharmacogenomic analysis of large patient derived xenografts data [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 3378.

Expression levels of long non-coding RNAs are prognostic for AML outcome

Background

Long non-coding RNA (lncRNA) expression has been implicated in a range of molecular mechanisms that are central in cancer. However, lncRNA expression has not yet been comprehensively characterized in acute myeloid leukemia (AML). Here, we assess to what extent lncRNA expression is prognostic of AML patient overall survival (OS) and determine if there are indications of lncRNA-based molecular subtypes of AML.

Methods

We performed RNA sequencing of 274 intensively treated AML patients in a Swedish cohort and quantified lncRNA expression. Univariate and multivariate time-to-event analysis was applied to determine association between individual lncRNAs with OS. Unsupervised statistical learning was applied to ascertain if lncRNA-based molecular subtypes exist and are prognostic.

Results

Thirty-three individual lncRNAs were found to be associated with OS (adjusted p value < 0.05). We established four distinct molecular subtypes based on lncRNA expression using a consensus clustering approach. LncRNA-based subtypes were found to stratify patients into groups with prognostic information (p value < 0.05). Subsequently, lncRNA expression-based subtypes were validated in an independent patient cohort (TCGA-AML). LncRNA subtypes could not be directly explained by any of the recurrent cytogenetic or mutational aberrations, although associations with some of the established genetic and clinical factors were found, including mutations in NPM1, TP53, and FLT3.

Conclusion

LncRNA expression-based four subtypes, discovered in this study, are reproducible and can effectively stratify AML patients. LncRNA expression profiling can provide valuable information for improved risk stratification of AML patients.

Electronic supplementary material

The online version of this article (10.1186/s13045-018-0596-2) contains supplementary material, which is available to authorized users.

Study design requirements for RNA sequencing-based breast cancer diagnostics

Sequencing-based molecular characterization of tumors provides information required for individualized cancer treatment. There are well-defined molecular subtypes of breast cancer that provide improved prognostication compared to routine biomarkers. However, molecular subtyping is not yet implemented in routine breast cancer care. Clinical translation is dependent on subtype prediction models providing high sensitivity and specificity. In this study we evaluate sample size and RNA-sequencing read requirements for breast cancer subtyping to facilitate rational design of translational studies. We applied subsampling to ascertain the effect of training sample size and the number of RNA sequencing reads on classification accuracy of molecular subtype and routine biomarker prediction models (unsupervised and supervised). Subtype classification accuracy improved with increasing sample size up to N = 750 (accuracy = 0.93), although with a modest improvement beyond N = 350 (accuracy = 0.92). Prediction of routine biomarkers achieved accuracy of 0.94 (ER) and 0.92 (Her2) at N = 200. Subtype classification improved with RNA-sequencing library size up to 5 million reads. Development of molecular subtyping models for cancer diagnostics requires well-designed studies. Sample size and the number of RNA sequencing reads directly influence accuracy of molecular subtyping. Results in this study provide key information for rational design of translational studies aiming to bring sequencing-based diagnostics to the clinic.

mBISON: Finding miRNA target over-representation in gene lists from ChIP-sequencing data

Background

Over-representation of predicted miRNA targets in sets of genes regulated by a given transcription factor (e.g. as defined by ChIP-sequencing experiments) helps to identify biologically relevant miRNA targets and is useful to get insight into post-transcriptional regulation.

Findings

To facilitate the application of this approach we have created the mBISON web-application. mBISON calculates the significance of over-representation of miRNA targets in a given non-ranked gene set. The gene set can be specified either by a list of genes or by one or more ChIP-seq datasets followed by a user-defined peak-gene association procedure. mBISON is based on predictions from TargetScan and uses a randomization step to calculate False-Discovery-Rates for each miRNA, including a correction for gene set specific properties such as 3’UTR length. The tool can be accessed from the following web-resource: http://cbdm.mdc-berlin.de/~mgebhardt/cgi-bin/mbison/home.

Conclusion

mBISON is a web-application that helps to extract functional information about miRNAs from gene lists, which is in contrast to comparable applications easy to use by everyone and can be applied on ChIP-seq data directly.

MicroRNAs modulate the dynamics of the NF-κB signaling pathway

BACKGROUND

NF-κB, a major transcription factor involved in mammalian inflammatory signaling, is primarily involved in regulation of response to inflammatory cytokines and pathogens. Its levels are tightly regulated since uncontrolled inflammatory response can cause serious diseases. Mathematical models have been useful in revealing the underlying mechanisms, the dynamics, and other aspects of regulation in NF-κB signaling. The recognition that miRNAs are important regulators of gene expression, and that a number of miRNAs target different components of the NF-κB network, motivate the incorporation of miRNA regulated steps in existing mathematical models to help understand the quantitative aspects of miRNA mediated regulation.

METHODOLOGY/PRINCIPAL FINDINGS

In this study, two separate scenarios of miRNA regulation within an existing model are considered. In the first, miRNAs target adaptor proteins involved in the synthesis of IKK that serves as the NF-κB activator. In the second, miRNAs target different isoforms of IκB that act as NF-κB inhibitors. Simulations are carried out under two different conditions: when all three isoforms of IκB are present (wild type), and when only one isoform (IκBα) is present (knockout type). In both scenarios, oscillations in the NF-κB levels are observed and are found to be dependent on the levels of miRNAs.

CONCLUSIONS/SIGNIFICANCE

Computational modeling can provide fresh insights into intricate regulatory processes. The introduction of miRNAs affects the dynamics of the NF-κB signaling pathway in a manner that depends on the role of the target. This "fine-tuning" property of miRNAs helps to keep the system in check and prevents it from becoming uncontrolled. The results are consistent with earlier experimental findings.