Improved metabolomic data-based prediction of depressive symptoms using nonlinear machine learning with feature selection

This study developed a novel prediction model for depressive symptoms using the Hilbert-Schmidt independence criterion least absolute shrinkage and selection operator (HSIC Lasso) algorithm, a nonlinear feature selection machine learning method. The model was applied to a large metabolomic dataset (897 subjects) from a population affected by the Great East Japan Earthquake. The HSIC Lasso model demonstrated superior predictive power compared to Lasso, support vector machine, partial least squares, random forest, and neural network models. Key metabolites contributing to prediction included L-leucine, 3-hydroxyisobutyrate, and gamma-linolenyl carnitine. The findings suggest that HSIC Lasso improves the prediction of depressive symptoms using metabolomic data in a Japanese population.

Yuta Takahashi, Masao Ueki, Makoto Yamada, Gen Tamiya, Ikuko N. Motoike, Daisuke Saigusa, Miyuki Sakurai, Fuji Nagami, Soichi Ogishima, Seizo Koshiba, Kengo Kinoshita, Masayuki Yamamoto, Hiroaki Tomita