Prog Neuropsychopharmacol Biol Psychiatry. 2025 May 2:111390. doi: 10.1016/j.pnpbp.2025.111390. Online ahead of print.
ABSTRACT
BACKGROUND: For suicide in major depression disorder, it is urgent to seek for a reliable neuroimaging biomarker with interpretable links to molecular tissue signatures. Accordingly, we developed an ensemble learning scheme over transcriptome-defined parcellations (TDP) to explore homogeneously parcellated brain patterns and their interactions.
METHODS: 96 depressed patients without suicide attempt (SA), 86 with SA and 102 healthy controls were recruited for resting state fMRI scanning. Six genetic dimensions were created by homogenous transcriptomic delineations from Allen Human Brain Atlas. Spatially-continuous TDPs were generated according to expression-levels of each brain region along diverse dimensions. Subsequently, TDPs were integrated with a three-layer ensemble learning scheme, where brain dysfunction of each TDP related to suicide was quantified with a resting-state functional abnormality (RSFA) score. Then, personalized index of brain dysfunction was produced according to the interactive pattern across TDPs.
RESULTS: Ensemble learning over TDPs displayed higher suicide predictive performance, relative to that over the regions level, and over null model (95 % CI of accuracy: 73.23 ± 1.07 %; 64.59 ± 3.00 %; 65.41 ± 3.97 %, respectively). Empowered by specific parieto-occipital TDP (PO-TDP) pattern quantified with RSFA score in suicide risk prediction, its alternations of SA effects were spatially associated with transcriptional profiles of GRIN2A and GABRG2. Moreover, glutamatergic and GABAergic synapse were overrepresented in enrichment analysis.
CONCLUSION: Glutamatergic and GABAergic dysfunction in the visual cortex was suggested via the PO-TDP specific interaction pattern. The inherent excitatory/inhibitory imbalance could contribute to aberrant emotional processing and neurocognitive impairment, ultimately leading to suicide.
PMID:40320231 | DOI:10.1016/j.pnpbp.2025.111390
AI-Assisted Evidence Search
Share Evidence Blueprint
Search Google Scholar