E-SNPs&GO: embedding of protein sequence and function improves the annotation of human pathogenic variants

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Precision Medicine

Background:

• Massive DNA sequencing generates numerous human single-nucleotide polymorphisms (SNPs) in protein-coding regions.
• Distinguishing harmful protein variations from neutral ones is a key challenge in precision medicine.
• Artificial intelligence (AI) offers novel approaches for protein sequence encoding, bypassing traditional evolutionary database searches.

Purpose of the Study:

• To develop a novel computational method, E-SNPs&GO, for predicting disease-relatedness of protein variations.
• To leverage advanced protein language models and embedding techniques for efficient variant annotation.
• To provide a scalable and accurate tool for analyzing large-scale protein variant datasets.

Main Methods:

• E-SNPs&GO utilizes protein language models and embedding techniques for input encoding.
• The method encodes both protein sequences and Gene Ontology (GO) functional annotations.
• A large dataset of 101,146 human protein single amino acid variants across 13,661 proteins was used for training.

Main Results:

• The model was trained on a comprehensive dataset derived from public resources.
• On a blind test set of 10,266 variants, E-SNPs&GO achieved a Matthews Correlation Coefficient (MCC) of 0.72.
• The performance of E-SNPs&GO is comparable to existing state-of-the-art methods for variant pathogenicity prediction.

Conclusions:

• E-SNPs&GO is proposed as an efficient and accurate tool for large-scale annotation of protein variant datasets.
• The method demonstrates the potential of AI-driven approaches in identifying disease-associated genetic variations.
• The developed webserver and datasets are publicly available for broader research use.