Personalized Disease Phenotypes from Massive OMICs Data

Introduction

Application of new high-throughput technologies in molecular medicine such as microarrays and next generation sequencing generates massive amounts of data for each individual patient studied. These methods enable to characterize the genotype and/or molecular phenotype on a personalized basis with the aim to increase our understanding of disease genesis and progression and, in final consequence, to improve diagnosis and treatment options. New methods are needed to handle such ‘big data’ sets collected for hundreds to thousands of persons in large epidemiological cohort studies, e.g. to accomplish data mining and classification tasks with impact for diagnosis and therapy. From the perspective of bioinformatics and systems biomedicine, ‘big data’ challenge objectives such as data integration, dimension reduction, data compression and visual perception. To finally achieve a personalized therapy it is necessary to link genetic variations to molecular disease phenotypes, to associate molecular with clinical data, to extract, to filter and to interpret bio-medical information and finally, to translate these discoveries into medical practice.

Machine learning represents one interesting option to tackle these tasks. Particularly, neural network algorithms such as self-organizing maps (SOMs) combine effective data processing and dimension reduction with strong visualization capabilities. These methods provide a suited basis to analyze large and complex data generated by modern bioanalytics.

The present contribution shortly describes the method of ‘SOM portraying’. We demonstrate data compression capabilities which reduce the dimension of the relevant (in terms of functional information) data by several orders of magnitude. The strong visualization capabilities of the SOM approach are illustrated. They enable the comprehensive, intuitive and detailed analysis of ‘big data’ in molecular medicine by mapping them into phenotype and feature space. To illustrate the performance of the method we present a series of representative case studies from different disease entities and OMICs realms related to the human colon.