Individual Drug Treatment Prediction in Oncology Based on Machine Learning Using Cell Culture Gene Expression Data

Abstract

Development of individual predictors of clinical drug efficiency becomes the mainstream in modern oncology. According to this approach, for a given patient with known type of cancer and a chosen drug, we should be able to estimate the treatment effect caused by the drug. Almost all works in this field apply machine learning techniques, which perform deep statistical analysis of a set of clinical cases supported by gene expression data for every patient. This important approach, unfortunately, suffers from an essential obstacle: the total set of cases available for analysis is very limited (usually several tens, very seldom several hundreds). On the other hand, in biotech drug industry, there are thousands of cell line cultures, supported by the gene expression data, which are analyzed to measure drug scoring. In this paper, we show how the cell lines data can be incorporated into to machine learning analysis to improve the development of individual predictors.