Construction of depression risk prediction model using genetic markers and machine learning

Depression is a leading cause of global disease burden, with diagnosis often relying on subjective measures. Genetic factors significantly contribute to its onset, supporting the use of machine learning for objective risk prediction. This study aims to develop a genetic-based risk prediction model by machine learning to distinguish between patients with depression and healthy people.

This study included 791 patients with depression and 413 healthy controls. Based on the Kyoto Encyclopedia of Genes and Genomes pathways, 1309 candidate genes were selected. Genotyping used the Illumina MiSeq platform. Single nucleotide polymorphisms (SNPs) were filtered for missing rate > 5% and minor allele frequency < 5%. The data were split into training and testing sets. In training set, feature selection combined Boruta and Least Absolute Shrinkage and Selection Operator Regression (LASSO) methods. In testing set, model performance was evaluated.

The final model incorporated 12 features (only 12 SNPs). It achieved an area under the curve (AUC) of 0.66 in the testing set, demonstrating good discrimination and stability.

This risk prediction model shows promise for objective depression risk assessment and could aid in early detection.