Machine learning combined with WGCNA reveals potential calcium uptake-associated biomarkers in Parkinson's disease

Background: Parkinson's disease (PD) is the second most common neurodegenerative disease, characterised by progressive loss of dopaminergic neurons in the substantia nigra pars compacta. Although previous studies have suggested a correlation between calcium signal disruption and the development of PD, the role of genes associated with store-operated calcium entry (SOCE) in PD remains unclear.

Aim: To combine machine learning methods with Weighted Gene Co-expression Network Analysis (WGCNA) to systematically screen and identify SOCE-associated candidate genes and evaluate their potential for assisting in discrimination.

Subjects and methods: This study is based on the GSE6613 dataset, using ssGSEA and Weighted Gene Co-expression Network Analysis (WGCNA) to screen for co-expression modules related to SOCE, and combining LASSO, SVM-RFE, and random forest algorithm to identify feature genes. Subsequently, validation was conducted in GSE20163 and GSE22491, and in vitro experiments were carried out in 6-OHDA and MPP⁺ models of dopaminergic neurons derived from LUHMES.

Results: A total of 5 SOCE-associated feature genes (LPCAT3, CLCNKB, TXLNA, etc.) were identified, and their combined model showed moderate discriminative ability in external validation. Immune analysis showed immune dysfunction in PD patients, and in vitro experiments also observed consistent expression trends with patient samples.

Conclusions: In summary, this study identified potential PD feature genes associated with SOCE, providing clues for further exploration of the molecular mechanisms of SOCE-associated calcium signalling in PD.