Opportunities and challenges of machine learning approaches for biomarker signature identification in psychiatry

The identification of reproducible biomarkers is an important step toward personalized medicine of psychiatric disorders. A large repertoire of machine learning tools is available that can aid in identifying such biomarker patterns from high-dimensional biological data. However, in psychiatry, the identification of clinically useful patterns has been challenging, due to the biological complexity and heterogeneity of the disorders, and the low effect sizes of individual biological markers. The incorporation of additional biological knowledge, such as information on biological network structure, or data from diverse modalities, is a promising route to make high-dimensional data more accessible for machine learning, and to identify more meaningful biological illness signatures. Here, we describe opportunities of such integrative analytics approaches and discuss unresolved challenges.