Predicting the pathogenicity of protein coding mutations using Natural Language Processing.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• DNA sequencing identifies numerous tumor mutations, but distinguishing pathogenic variants from neutral ones is a significant challenge.
• Current manual annotation of cancer-driving mutations is time-consuming, costly, and inefficient.
• Automating the identification of functionally significant genetic variations is crucial for cancer research.

Purpose of the Study:

• To introduce NLP-SNPPred, a novel method for predicting pathogenic genetic variations using scientific literature.
• To leverage Natural Language Processing (NLP) and machine learning to automate the identification of cancer-related mutations.
• To develop a computational tool that reduces the manual effort in variant annotation.

Main Methods:

• Utilizing state-of-the-art NLP techniques (sent2vec, word2vec, tf-idf) to create vector representations of biomedical literature.
• Ingesting bio-medical literature to learn implicit features distinguishing pathogenic variations.
• Training machine learning predictors on these representations, using OncoKB data and evaluating on benchmark datasets.

Main Results:

• NLP-SNPPred effectively predicts the functional impact of protein-coding variations.
• The method outperforms existing state-of-the-art function prediction approaches.
• Minimal complementary biological features are required, demonstrating the power of NLP in this domain.

Conclusions:

• Natural Language Processing (NLP) can be effectively applied to predict the functional impact of genetic variations.
• Encoding biological knowledge into machine learning models automates laborious manual annotation processes.
• NLP-SNPPred offers a promising, automated solution for identifying pathogenic mutations in cancer research.