Pathway-extended gene expression signatures integrate novel biomarkers that improve predictions of patient responses to kinase inhibitors

Abstract

Cancer chemotherapy responses have been related to multiple pharmacogenetic biomarkers, often for the same drug. This study utilizes machine learning to derive multi-gene expression signatures that predict individual patient responses to specific tyrosine kinase inhibitors, including erlotinib, gefitinib, sorafenib, sunitinib, lapatinib and imatinib. Support Vector Machine learning was used to train mathematical models that distinguished sensitivity from resistance to these drugs using a novel systems biology-based approach. This began with expression of genes previously implicated in specific drug responses, then expanded to evaluate genes whose products were related through biochemical pathways and interactions. Optimal pathway-extended support vector machines predicted responses in patients at accuracies of 70% (imatinib), 71% (lapatinib), 83% (sunitinib), 83% (erlotinib), 88% (sorafenib) and 91% (gefitinib). These best performing pathway-extended models demonstrated improved balance predicting both sensitive and resistant patient categories, with many of these genes having a known role in cancer etiology. Ensemble machine learning-based averaging of multiple pathway-extended models derived for an individual drug increased accuracy to >70% for erlotinib, gefitinib, lapatinib, and sorafenib. Through incorporation of novel cancer biomarkers, machine learning-based pathway-extended signatures display strong efficacy predicting both sensitive and resistant patient responses to chemotherapy.

Improved radiation expression profiling in blood by sequential application of sensitive and specific gene signatures

“…Both FSFS and BSFS models were derived from the top 50 ranked mRMR genes, in addition to other published radiation responsive genes: AEN , BAX , BCL2 , DDB2 , FDXR , PCNA , POU2AF1 , and WNT3 . SVMs were derived with a Gaussian radial basis function kernel by iterating over box-constraint (C) and kernel-scale (σ) parameters and gene features, minimizing to either misclassification or log loss by cross- validation (Zhao et al 2018a; Bagchee-Clark et al 2020). Gene signatures were then assessed with a validation dataset and re-evaluated (by misclassification rates, log loss, Matthews correlation coefficient, or goodness of fit).…”

“…Several groups (Boldt et al 2012;Budworth et al 2012;Knops et al 2012;Ghandhi et al 2017) have also used signatures to determine radiation exposures. Biochemically-inspired machine learning is a robust approach to derive diagnostic gene signatures for chemotherapy (Dorman et al 2016;Mucaki et al 2016;Mucaki et al 2019;Bagchee-Clark et al 2020) as well as radiation (Zhao et al 2018a). Given the limited sample sizes of typical datasets, appropriate ML methods for deriving gene signatures have included Support Vector Machines, Random Forest classifiers, Decision Trees, Simulated Annealing, and Artificial Neural Networks (Rogan, 2019;Boldrini et al 2019).…”

“…Our approach uses supervised machine learning (ML) with genes previously implicated or established from genetic evidence and biochemical pathways that are altered in response to these exposures (Zhao et al 2018a). Biochemically-inspired ML is a robust approach to derive diagnostic gene signatures for radiation and chemotherapy (Dorman et al 2016; Mucaki et al 2016; Mucaki et al 2019; Bagchee-Clark et al 2020). Given the limited sample sizes of typical datasets, appropriate ML methods for deriving gene signatures have included Support Vector Machines, Random Forest classifiers, Decision Trees, Simulated Annealing, and Artificial Neural Networks (Rogan, 2019; Boldrini et al 2019).…”

“…A Support Vector Model (SVM) or signature is generated from the top ranked gene set by adding and removing genes that minimize either model misclassification or log-loss (Zhao et al 2018a). ML alone is not sufficient to derive useful signatures, since underlying biochemical pathways and disease mechanisms also contribute to selecting relevant and non-redundant gene features (Bagchee-Clark et al 2020). Signatures were evaluated either by a traditional validation approach or by stratified k-fold validation (which splits the same dataset into k groups reserved for testing and training).…”

Decoding ‘Unnecessary Complexity’: A Law of Complexity and a Concept of Hidden Variation Behind “Missing Heritability” in Precision Medicine

“…Both genomic complexity- and physiology-based approaches are necessary to understand the relationship between genotypes and phenotypes. Finally, Bagchee-Clark et al ( 2020 ) have shown that machine learning-based, pathway-extended gene expressions measurements can be successfully used to identify novel biomarkers that improve predictions of patient response to drugs in cancer therapy.…”