CERENKOV3: Clustering and molecular network-derived features improve computational prediction of functional noncoding...

Area of Science:

• Computational Biology
• Genetics
• Bioinformatics

Background:

• Identifying causal noncoding single nucleotide polymorphisms (SNPs) is crucial for human genome-wide association studies (GWAS).
• Functional SNP identification remains a significant challenge in computational biology.
• Existing machine learning methods show promise but require further improvement.

Purpose of the Study:

• To develop an advanced machine learning pipeline, CERENKOV3, for improved prediction of regulatory SNPs (rSNPs).
• To enhance post-GWAS analysis by increasing the accuracy of identifying functional SNPs.
• To integrate novel features derived from clustering and molecular networks.

Main Methods:

• Developed CERENKOV3, a machine learning pipeline incorporating clustering-derived and molecular network-derived features.
• Introduced 'locus size' as a clustering-derived feature based on SNP location clusters.
• Generated molecular network features using representation learning on SNP-gene and gene-gene networks.

Main Results:

• CERENKOV3 demonstrated significantly improved rSNP recognition performance.
• Performance gains were observed across multiple metrics: AUPRC, AUROC, and AVGRANK.
• The pipeline effectively leverages novel features for enhanced predictive accuracy.

Conclusions:

• CERENKOV3 represents a significant advancement in computational methods for identifying regulatory SNPs.
• The integration of locus size and molecular network features boosts prediction accuracy in post-GWAS analysis.
• This pipeline offers a powerful tool for maximizing genetic insights from large-scale human genetic studies.