Machine learning algorithms assisted identification of post-stroke depression associated biological features

Area of Science:

• Genomics and Bioinformatics
• Neuroscience
• Metabolomics

Background:

• Post-stroke depression (PSD) is a common complication impacting stroke patient recovery.
• Current diagnostic and therapeutic strategies for PSD lack effective, metabolism-related biomarkers.
• Stroke involves dynamic metabolic and hemodynamic changes, highlighting the need for metabolism-focused research.

Purpose of the Study:

• To discover metabolism-related diagnostic and therapeutic biomarkers for post-stroke depression (PSD).
• To identify reliable molecular targets for improving PSD diagnosis and treatment.

Main Methods:

• Utilized gene expression datasets (GSE140275, GSE122709, GSE180470) from the GEO database.
• Applied differential gene expression (DEG) analysis, Weighted Gene Co-expression Network Analysis (WGCNA), and machine learning algorithms (LASSO, random forest).
• Validated candidate genes in independent datasets and constructed a nomogram model for PSD diagnosis.

Main Results:

• Identified 557 metabolism-associated candidate hub genes.
• Selected two signature genes, SDHD and FERMT3, with significant roles in depression.
• The nomogram model demonstrated good diagnostic accuracy (AUC for SDHD: 0.896, FERMT3: 0.964) and correlated with depression severity (HAMD scores).

Conclusions:

• SDHD and FERMT3 are identified as valuable diagnostic and therapeutic biomarkers for post-stroke depression (PSD).
• These biomarkers offer potential for earlier diagnosis and prevention of PSD.
• The findings contribute to understanding the metabolic underpinnings of PSD and offer new avenues for clinical intervention.