E-SNPs and GO

E-SNPs and GO predicts the pathogenicity of single amino acid variants in human proteins to distinguish disease-associated from benign variants for precision medicine applications.

Key Features:

• Machine-learning framework: Employs machine-learning models that use advanced artificial intelligence and protein language model embeddings to encode protein sequences.
• GO annotation encoding: Encodes Gene Ontology (GO) functional annotations alongside protein sequence embeddings.
• No reliance on evolutionary searches: Does not rely on database searches for evolutionary information to generate input features.
• Training dataset: Trained on 101,146 human single amino acid variants across 13,661 proteins derived from public resources.
• Predictive output: Produces binary predictions of disease association (pathogenic vs benign) for single amino acid variants.
• Performance: Validated on a blind test set of 10,266 variants, achieving a Matthews Correlation Coefficient (MCC) of 0.72.
• Large-scale annotation: Designed for large-scale annotation of protein variant datasets.

Scientific Applications:

• Variant annotation and prioritization: Annotates and prioritizes protein single amino acid variants for precision medicine and disease association studies.
• Genomic interpretation: Interprets single-nucleotide polymorphisms in protein-coding regions for potential functional impact.
• Functional effect prediction: Integrates sequence-derived embeddings with Gene Ontology annotations to inform functional effect predictions of variants.

Methodology:

Uses machine-learning models trained on 101,146 human single amino acid variants (13,661 proteins) with input encodings based on protein language model embeddings and Gene Ontology annotations, and evaluated on a blind test of 10,266 variants (MCC 0.72).