Modern anti-cancer treatments such as tyrosine kinase and checkpoint inhibitors are typically efficient in a minority of patients. Due to statistical challenges, attempts to augment this knowledge by using omics profiling of patient samples have been mostly fruitless. We create a novel generation of biomarker panels using network-level analysis and fuzzy set theory. When individual gene/protein profiles fail to explain the differential response, a deep data mining framework integrates disparate, pathway-like patterns into cumulative scores. The marker sets arranged into diagnostic panels are compact, statistically and biologically significant predictors of treatment outcome which can be validated with independent clinical datasets.
Known determinants of treatment response (specific mutations, total mutation burden, immunoscores etc.) just partially explain the patient response. Our method, NeaMarker [1] maps patient-specific (mutated, differentially expressed etc.) genes to characteristic pathways or otherwise defined proprietary gene sets. The pathway-level markers should then be arranged into diagnostic panels.
Beyond the routine NGS and proteomics data processing, we apply various data integration, machine learning, and data mining methods. The method of network enrichment analysis [2] employs omics data using the “global network vs. pathway vs. patient” analytic paradigm. It gathers information from scarce, rare molecular driver events and builds decision support by predicting patients’ response in a robust and reproducible manner.
Yet another application of the methodology is NEAdriver [3], a toolbox for discovering novel cancer driver mutations which have not been spotted previously because of being individually rare events [3].