Computational methods for coherent analysis of large-scale transcriptional, metabolic and histomorphometric profiles of a biological system: application to biomarker discovery

Summary form only given. This paper describes computational inference methods for coherently analyzing three disparate and large-scale data streams for biomarker discovery. High-throughput biological approaches can assess the global molecular states of a living system by measuring the levels of thousands of transcribed genes (transcriptional profiles) and metabolites (metabolic profiles). However, effectively analyzing this information for understanding the disease pathways is a challenging problem. To address this problem a unique approach based on the concurrent analysis of gene expression, metabolic and histomorphometric data is developed. A novel machine vision based method was developed to assay the cellular state using quantitative tissue features. In this work we explore the formalization of biomarker discovery in a machine learning paradigm. In our approach the biomolecular and quantitative tissue profiles correspond to large and noisy descriptor space, and the classification attributes are defined by phenotypes. The relatively small number of samples compared to very high-dimensional and noisy descriptor space makes biomarker discovery an extremely challenging classification problem. The response of acetaminophen (APAP) in the rat liver was studied through transcriptional profiling using Affymetrix gene chips and metabolic profiling using liquid chromatography coupled with mass spectrometry (LC/MS). This paper describes the computational approach for the integrated metabolic, transcriptional, and quantitative tissue profiles elucidating the response of the rat liver to APAP exposure.