ClinPred: Prediction Tool to Identify Disease-Relevant Nonsynonymous Single-Nucleotide Variants

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-throughput DNA sequencing has accelerated variant discovery but interpreting their phenotypic effects remains challenging.
• Existing computational tools for predicting variant impact have limited accuracy, necessitating improved methods.

Purpose of the Study:

• To introduce ClinPred, an efficient computational tool for identifying disease-relevant nonsynonymous variants.
• To enhance the accuracy of predicting variant pathogenicity by incorporating novel features and datasets.

Main Methods:

• ClinPred utilizes two machine learning algorithms, integrating existing pathogenicity scores.
• It incorporates normal population allele frequency from the gnomAD database as a key input feature.
• The tool is trained on confidently annotated disease-causing variants from the ClinVar database.

Main Results:

• ClinPred demonstrated superior accuracy in predicting variant pathogenicity compared to existing methods, achieving the highest area under the curve (AUC).
• The predictor showed increased specificity and sensitivity across various test datasets and remained robust across different disease types and mechanisms.
• Including allele frequency as a predictive feature, rather than fixed cutoffs, significantly boosted prediction performance.

Conclusions:

• ClinPred offers a significant advancement in predicting the pathogenicity of genetic variants.
• The tool's high accuracy and robustness make it valuable for clinical genetics and research.
• Pre-computed ClinPred scores for human missense variants are provided to facilitate community use.