Abstract

Biomarkers identification is difficult for cancer and other polygenic traits because such complicated diseases occur due to an intricate interplay of various genetic materials. Although high-throughput data from recent technolo-gies provide access to a tremendous amount of information still there is a huge gap in harnessing knowledge from the generated multi-omics data. It is evident from the availability of subject-specific multi-omics data from large consortiums that there is a growing need for appropriate tools to analyze such data. Traditional single-omics association tests more often identify strong signals but fail to explore the between-omics relationship and find moderately weak signals due to multiple testing burdens. Multi-omics data integration intuitively provides a clear advantage in understanding the genetic architecture of disease a little better by imparting complementary information. But the construction of such methods is challenging because of the diversity in the nature of multiple omics and the sample size which is much less than the number of omics variables. It is important to consider factors such as data diversity and prior biological knowledge to make meaningful and better predictions. Dimension reduction techniques such as feature selection are used to circumvent the sample size issue in general but treating all the omics variables similarly might be an oversimplification of the complex biological interactions. The lack of appropriate approaches for biomarker identification from complex multi-omics data led us to develop this method. ioSearch is a tool for integrating two omics assays with continuous measurements. Based on a two-step model, ioSearch explores the inter-relationship of the omics in a principal regression framework and selects features using sparse principal component analysis to provide easily interpretable inference in terms of p-values. Also, it uses prior biological information to reduce multiple testing burdens. Extensive simulation results show that our method is statistically powerful with a controlled type I error rate. Application of ioSearch to two publicly available breast cancer datasets identified relevant genes and proteins in important pathways.