An iterative searching and ranking algorithm for prioritising pharmacogenomics genes.

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Pharmacogenomics (PGx) research aims to link genetic variations to drug efficacy and toxicity.
• Machine-readable drug-gene relationship knowledge is essential for computational PGx and personalized medicine.
• A curated PGx gene lexicon is vital for automated extraction of drug-gene relationships from scientific literature.

Purpose of the Study:

• To develop and evaluate a bootstrapping learning technique for ranking human genes based on their relevance to drug response.
• To create a comprehensive and accurate PGx-specific gene lexicon from scientific literature.
• To enhance the accuracy of identifying genes critical for personalized medicine.

Main Methods:

• A bootstrapping learning algorithm was employed to rank 33,310 human genes.
• The algorithm utilized a single seed pharmacogenomic gene to iteratively extract and rank co-occurring genes.
• Analysis was performed on a corpus of 20 million MEDLINE abstracts.

Main Results:

• The developed ranking method successfully prioritized PGx-specific genes among all human genes.
• The algorithm achieved high performance metrics: precision of 0.861, recall of 0.548, and F1-score of 0.662 for the top 2.5% of ranked genes.
• This performance significantly outperformed random gene ranking (precision: 0.032, recall: 0.013, F1: 0.018).

Conclusions:

• The bootstrapping learning technique is effective for generating a ranked list of PGx-relevant genes.
• This approach facilitates the creation of a valuable PGx gene lexicon for computational studies and personalized medicine.
• The method demonstrates a significant improvement in identifying genes critical for understanding drug response.