Accurate Sequence-Based Prediction of Deleterious nsSNPs with Multiple Sequence Profiles and Putative Binding...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Non-synonymous single nucleotide polymorphisms (nsSNPs) can cause pathogenic changes linked to human diseases.
• Accurate prediction of deleterious nsSNPs is essential for understanding disease mechanisms.
• Current nsSNP prediction tools offer modest performance, necessitating improved methods.

Purpose of the Study:

• To develop a novel, highly accurate sequence-based predictor for deleterious nsSNPs.
• To enhance the prediction of pathogenic nsSNPs by leveraging protein sequence conservation and functional site information.
• To provide a tool that can effectively guide experimental validation in high-throughput settings.

Main Methods:

• Developed DMBS, a sequence-based predictor utilizing improved conservation estimates from multiple sequence profiles (two databases, two alignment algorithms).
• Integrated putative functional/binding residue annotations from state-of-the-art sequence-based methods.
• Employed a random forests model for classification, empirically compared against five other machine-learning algorithms.

Main Results:

• DMBS achieved an Area Under the Curve (AUC) > 0.94 across four benchmark datasets, outperforming existing methods.
• DMBS demonstrated superior performance on specific datasets, achieving AUC = 0.97 for SNPdbe and AUC = 0.97 for ExoVar, compared to 0.70 and 0.88 for prior methods.
• On the independent HumVar dataset, DMBS significantly outperformed the state-of-the-art SNPdryad method.

Conclusions:

• DMBS offers a significant advancement in predicting deleterious nsSNPs with high accuracy.
• The method's reliance on enhanced conservation estimates and functional site predictions contributes to its superior performance.
• DMBS provides a valuable tool for efficiently prioritizing nsSNPs for experimental validation in disease research.