Exploiting MEDLINE for gene molecular function prediction via NMF based multi-label classification.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene Ontology (GO) is crucial for describing gene functions across model organisms.
• Manual GO annotation is labor-intensive and time-consuming.
• Existing automated methods often underutilize valuable information from scientific literature.

Purpose of the Study:

• To develop and evaluate an innovative predictive system for automatic molecular function (GO term) assignment to genes.
• To leverage the biomedical literature for gene annotation.
• To address the multi-label nature of gene function prediction.

Main Methods:

• The study framed GO molecular function annotation as a multi-label classification problem.
• Non-negative matrix factorization (NMF) was employed for feature reduction.
• The binary-relevance method was used to handle multi-label data, combined with a K-nearest neighbor (KNN) classifier.

Main Results:

• The developed system automatically assigns molecular functions to genes using literature data.
• The combination of binary-relevance and KNN classifier demonstrated superior performance.
• An F1-measure of 0.84 was achieved on the UniProtKB/Swiss-Prot dataset.

Conclusions:

• The proposed automated system effectively assigns Gene Ontology molecular function terms to genes.
• This approach offers a more efficient alternative to manual GO annotation.
• The integration of biomedical literature significantly enhances gene function prediction accuracy.