News Releases & Research Results Prediction of the risks of depressive symptoms using machine learning - For personalized medicine for mental disorders -

The results of research and development project conducted by medical staff Yuta Takahashi, Assistant Professor Masao Ueki (currently, professor of Nagasaki University), Professor Gen Tamiya, and Professor Hiroaki Tomita of the Tohoku Medical Megabank Organization, Tohoku University.

• A machine learning method for suppressing overfitting* was demonstrated to be useful in predicting the risks of mental disorders, such as depression, through comparison of multiple mathematical models.
  *Low prediction accuracy at actual testing despite high prediction accuracy, i.e., apparent high performance, at learning.
• Specifically, the “STMGP method*,” a machine learning method developed by the research group, was demonstrated to be useful in predicting the risks of mental disorders, such as depression, in which various DNA polymorphisms are assumed to be intricately involved in their pathological conditions.
  *Abbreviation of Smooth-Threshold Multivariate Genetic Prediction.
• The results of this R&D project should facilitate the personalized medicine and prevention of depression and the pathophysiological elucidation of genetic predispositions.

This R&D project was conducted with the support of the Strategic Research Program for Brain Sciences by AMED.

The results were published in the American scientific journal Translational Psychiatry on August 17.

Takahashi Y., et al. Improved metabolomic data-based prediction of depressive symptoms using nonlinear machine learning with feature selection Translational Psychiatry
DOI：10.1038/s41398-020-0831-9