E-SNPs and GO

E-SNPs and GO annotates human single amino acid variants to predict pathogenicity by integrating protein sequence encodings and Gene Ontology (GO) functional information.

Key Features:

• AI-based encoding: Employs artificial intelligence models to encode protein sequences, eliminating the need for traditional database searches for evolutionary data.
• Integration of Gene Ontology (GO): Encodes GO functional annotations alongside sequence information to capture functional context of variants.
• Protein language models and embeddings: Uses protein language models and embedding techniques for input encoding to represent complex relationships within protein sequences and functions.
• Predictive classification: Predicts whether a single amino acid variation (SAV) originating from single-nucleotide polymorphisms (SNPs) is associated with disease.
• Training and validation: Trained on 101,146 human protein single amino acid variants across 13,661 proteins and evaluated on a blind test set of 10,266 variants achieving a Matthews Correlation Coefficient (MCC) score of 0.72.

Scientific Applications:

• Precision medicine: Supports annotation of protein variants to inform clinical interpretation and therapeutic decision-making.
• Disease genetics: Aids in distinguishing harmful protein variations from neutral ones to study genetic contributions to disease.
• Drug target identification: Assists in identifying potential targets for drug development by highlighting pathogenic variants.

Methodology:

Encodes protein sequences and Gene Ontology annotations using artificial intelligence models and protein language model embeddings, avoids traditional evolutionary database searches, and was trained on 101,146 variants (13,661 proteins) with evaluation on a blind test set of 10,266 variants reporting an MCC of 0.72.