Carcinogenicity prediction by in vitro human liver cell gene expression and chemical structure

Cancer is one of the leading causes of mortality in the world while the carcinogenicity prediction remains challenging. In this work, carcinogenicity prediction is researched in two ways, by chemical structures and sub-structures and by gene expression exposed to drugs. Probabilistic support vector machine is used to train model by the gene expression data. The results show that the gene expression gives clues about the carcinogenicity. A comprehensive literature mining for top-ranked microarray probes provided by feature selection showed strong relevance between corresponding genes and cancer. This work illustrated that our method is an effective computational strategy for carcinogenicity prediction and cancer biomarker discovery.