Identifying optimal substrate classes of membrane transporters.

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Membrane transporters are crucial for molecular transport across biological membranes, impacting key biological pathways.
• Accurate identification of membrane transporters and their substrates is vital for biotechnology, pharmacology, and metabolomics.
• Existing protein functional annotations (e.g., Gene Ontology) are often too complex, redundant, and hierarchical for supervised learning.

Purpose of the Study:

• To develop an automated pipeline for selecting an optimal subset of functional annotations for membrane transporters.
• To create machine learning-ready training datasets for predicting transporter function and substrate specificity.
• To reduce annotation complexity while maintaining biological relevance and class separability.

Main Methods:

• Development of an automated pipeline employing machine learning to identify a minimal, non-redundant set of functional annotations.
• Implementation of criteria for class selection: sufficient sample size, minimal redundancy, strong separability, and transport relevance.
• Application of the pipeline to create training datasets for transmembrane transporters across various organisms (yeast, plants, bacteria, mammals).

Main Results:

• Reduced 287 functional annotations to 11 GO terms for S. cerevisiae transporters, achieving a median pairwise F1 score of 0.87±0.16.
• For a multi-organism dataset, reduced 695 annotations to 49 terms with a median F1 score of 0.92±0.10.
• Even with 67% protein coverage, the pipeline identified 30 GO terms with a high median F1 score of 0.95±0.06.

Conclusions:

• The automated pipeline effectively reduces complex functional annotation datasets for membrane transporters.
• The selected subset of annotations is suitable for training predictive models, improving the study of transporter functions and substrates.
• This approach facilitates large-scale analysis of transporter proteins across diverse biological systems.