Linking Protein Stability to Pathogenicity: Predicting Clinical Significance of Single-Missense Mutations in Ocular Proteins Using Machine Learning

Int J Mol Sci. 2024 Oct 30;25(21):11649. doi: 10.3390/ijms252111649.

Abstract

Understanding the effect of single-missense mutations on protein stability is crucial for clinical decision-making and therapeutic development. The impact of these mutations on protein stability and 3D structure remains underexplored. Here, we developed a program to investigate the relationship between pathogenic mutations with protein unfolding and compared seven machine learning (ML) models to predict the clinical significance of single-missense mutations with unknown impacts, based on protein stability parameters. We analyzed seven proteins associated with ocular disease-causing genes. The program revealed an R-squared value of 0.846 using Decision Tree Regression between pathogenic mutations and decreased protein stability, with 96.20% of pathogenic mutations in RPE65 leading to protein instability. Among the ML models, Random Forest achieved the highest AUC (0.922) and PR AUC (0.879) in predicting the clinical significance of mutations with unknown effects. Our findings indicate that most pathogenic mutations affecting protein stability occur in alpha-helices, beta-pleated sheets, and active sites. This study suggests that protein stability can serve as a valuable parameter for interpreting the clinical significance of single-missense mutations in ocular proteins.

Keywords: computational biology; genetic mutations; inherited eye disease; machine learning; pathogenicity prediction; protein stability.

MeSH terms

  • Clinical Relevance
  • Eye Diseases / genetics
  • Eye Proteins / chemistry
  • Eye Proteins / genetics
  • Eye Proteins / metabolism
  • Humans
  • Machine Learning*
  • Mutation, Missense*
  • Protein Stability*
  • cis-trans-Isomerases / chemistry
  • cis-trans-Isomerases / genetics
  • cis-trans-Isomerases / metabolism

Substances

  • Eye Proteins
  • cis-trans-Isomerases
  • retinoid isomerohydrolase