Matrix factorization-based data fusion for drug-induced liver injury prediction

Marinka Zitnik and Blaz Zupan Matrix factorization-based data fusion for drug-induced liver injury prediction. In: CAMDA, July 2013, Berlin.

Preview

PDF Download (384Kb)

Abstract

We report on a data fusion approach for prediction of outcome of drug-induced liver injury (DILI) in humans from gene expression studies as provided by the CAMDA 2013 Challenge. Our aim was to investigate if the data from all four toxicogenomics studies can be fused together to boost prediction accuracy. We show that recently proposed matrix factorization-based fusion provides an elegant framework for integration of CAMDA and related data sets. Our data fusion approach yields a high cross-validated AUC of 0.819 (in vivo assays), which is above the accuracy of standard machine learning procedures (stacked classification with feature selection). Achieved accuracy is also a substantial improvement of the highest scores on the same data sets reported in CAMDA 2012. Our data analysis shows that animal studies can be replaced with in vitro assays (AUC = 0.799) and that we can predict liver injury in humans from animal data (AUC = 0.811).

Actions (login required)