Russian researchers develop neural network model to detect depression with over 96% accuracy

Russian scientists have developed a graph-based neural network model that detects depression by jointly analyzing genetic and neurophysiological data, according to a report from the Siberian branch of the Russian Academy of Sciences. Aleksandr Savostyanov, one of the developers, said the system is built around a graph neural network architecture trained on genetic patterns and EEG readings from both healthy and depressed individuals before being tested on its ability to distinguish patients from healthy subjects. The study covered more than 3,000 participants drawn from across Siberia, from the Altai region to Buryatia. Researchers sequenced 164 genetic regions from blood and buccal mucosa samples, collected EEG recordings, and administered psychological questionnaires. Savostyanov noted that genetic predisposition to depression can be identified from birth, but prediction accuracy using genetic data alone remains low, while EEG readings reflect only momentary brain states and can be quickly distorted by factors such as hunger or recent dental procedures. Prior neural network approaches achieved roughly 86% accuracy in identifying depression, but the new graph-based method exceeded 96%, according to Savostyanov. He said the model also has a low false-negative rate, which the researchers consider critical for the early screening of psychiatric conditions. The team said combining genetic, neurophysiological and behavioral data substantially improved diagnostic reliability compared with any single data stream.