FARMS: a generative framework for microarray data processing

Description

Cost-effective oligonucleotide arrays are the predominant technique to measure high-dimensional genomic data like expression levels of genes or DNA copy number variations (CNVs). We present a latent variable model for summarizing microarray data called "FARMS: Factor Analysis for Robust Microarray Summarization". FARMS is based on a multiplicative latent variable model, which accounts for linear dependencies in the data, and also captures heavy-tailed distributions as observed in real-world transcriptomic data. The generative framework allows to utilize well-founded model selection methods and to apply Bayesian techniques. In contrast to previous methods FARMS supplies model-based signal intensity values and a novel criterion for unsupervised feature selection named “I/NI-Calls”. In our feature selection we propose to exclude all probe sets where a variation of the latent variable cannot reliably be detected by a maximum a posteriori optimization that combines and trades-off noise and signal likelihood. In this session we will present: (a) a generative model for summarizing microarray data which additionally allows to filter out genes according to their information content; (b) a rigorous assessment with 130 competitors and (c) results of detecting copy number variations (CNVs) using genotyping microarrays and an assessment with the most prevalent methods.

Period	31 Aug 2012
Event title	International Biometric Conference 2012
Event type	Conference
Location	JapanShow on map