On the combination of omics data for prediction of binary outcomes

TL;DR

This paper reviews methods for integrating multiple omic data sources to improve binary outcome prediction, evaluating their performance through real datasets on breast cancer detection and obesity prediction.

Contribution

It introduces and compares approaches based on double cross-validation and regularized regression for combining diverse omic biomarkers in binary outcome models.

Findings

Combined omic models improve prediction accuracy over single-omic models.

Methods effectively calibrate and discriminate binary health outcomes.

Real data applications demonstrate practical utility of the approaches.

Abstract

Enrichment of predictive models with new biomolecular markers is an important task in high-dimensional omic applications. Increasingly, clinical studies include several sets of such omics markers available for each patient, measuring different levels of biological variation. As a result, one of the main challenges in predictive research is the integration of different sources of omic biomarkers for the prediction of health traits. We review several approaches for the combination of omic markers in the context of binary outcome prediction, all based on double cross-validation and regularized regression models. We evaluate their performance in terms of calibration and discrimination and we compare their performance with respect to single-omic source predictions. We illustrate the methods through the analysis of two real datasets. On the one hand, we consider the combination of two fractions of proteomic mass spectrometry for the calibration of a diagnostic rule for the detection of early-stage breast cancer. On the other hand, we consider transcriptomics and metabolomics as predictors of obesity using data from the Dietary, Lifestyle, and Genetic determinants of Obesity and Metabolic syndrome (DILGOM) study, a population-based cohort, from Finland.