Integrated analysis of multi-source data in drug discovery experiments using structural equation models

Abstract

The drug discovery and development processes are typically costly and time consuming. Hence, it is crucial to identify early failure of candidate compounds and thereby save time and investment in a later stage. We propose structural equation modeling (SEM) based approach for an integrated analysis which combines information from three data sources: (1) bioactivity variables, (2) variables representing the chemical structure of the compounds, and (3) gene expression data. The proposed model allows to estimate the effects of the gene expression on the biological activity variable and furthermore, it allows to decompose the effect of the chemical structure on the biological activity into direct and indirect (i.e. the effect via the gene expression) effects.