In Silico Assessment of SYNGAP1 Missense Variant Pathogenicity 

ABSTRACT

SynGAP, a critical regulator at the postsynaptic density, is associated with Alzheimer's disease, tumorigenesis, and most notably, SYNGAP1-related disorders (SRD), which manifests as autism, epilepsy, and cognitive impairment. SRD is likely primarily caused by SynGAP missense (SGM) variants, whose pathogenicity is difficult to confirm experimentally. To address this issue, a wide range of structure- and/or sequence-based as well as machine learning- and deep learning-based in silico predictors were benchmarked against clinically annotated SRD data. Folding stability calculations (e.g., Foldetta) and deep learning methods ESM1b and AlphaMissense showed the highest prediction accuracy under default settings. Applying optimized nondefault thresholds, especially with AlphaMissense, could further improve predictor performance. The predictions exhibited a strong structural pattern: Pathogenic variants were highly enriched within defined structural domains, particularly the GAP and PH domains, whereas variants located in nondomain and nonmotif regions were predominantly predicted to be benign. The verified phosphosites and predicted phase separation impacting variants occupy matching sections, including the protein's N-terminal region and the vicinity of putative SH3-binding motif in the disordered regions. Thus, these regions collectively could act as phase separation-promoting segments, complementing the protein's trimerization-coupled coiled-coil domain. Finally, the SynGAP Missense (SGM) server was presented to support SynGAP variant assessment. The SGM server integrates a Data Table with top-performing predictors, a 3D Viewer for structure-level assessment and AI-generated lay summaries for all variants arising from single-base substitutions, including those not yet observed in public databases.

PMID:43005510 | PMC:PMC13745803 | DOI:10.1155/humu/8234670

 Read More