Bioinformatic analysis of ACAD9 genetic variability

Mitochondrial diseases (MD) are a heterogeneous group of disorders with a diagnostic rate often below 50%, mostly due to the difficulty of interpreting Variants of Unknown Significance (VUS). ACAD9 is a nuclear gene that encodes a mitochondrial protein involved in fatty acid β-oxidation and the assembly of respiratory complex I. This study aimed to refine the classification of ACAD9 missense VUS through an integrative computational approach combining in silico, evolutionary and structural analysis. A gene-specific dataset of missense pathogenic, benign and uncertain significance variants was curated from public databases (HGMD, ClinVar, GnomAD, UniProt or LOVD) and in-house sequencing data. Thirty-five in silico pathogenicity predictors were retrieved from the dbNSFP v5.3 and benchmarked using the datasets. Predictor performance was evaluated using 10 metrics and the top predictors were integrated with evolutionary conservation (ConSurf), structural modelling (PyMOL, HOPE) and thermodynamic stability analysis (FoldX) to re-evaluate 17 VUS. Predictor performance for ACAD9 variants varied significantly. Because no statistically significant differences were detected among the ten highest-ranked predictors, ClinPred, MetaRNN, PHACTboost, MutationTaster and REVEL were selected as the top stability-informed predictors. The integrative predictive and structural analysis reclassified 8 variants as "likely pathogenic" and 5 as "likely benign", while 4 remained classified as VUS due to conflicting evidence. A gene-specific, multi-layered computational strategy substantially improves VUS classification in ACAD9. These in silico findings provide strong, albeit non-definitive, evidence to guide diagnostic efforts. It should be complemented with functional validation, which remains essential to confirm pathogenicity.