SOMRuler: a novel interpretable transmembrane helices predictor.

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Biology

Background:

• Transmembrane helix (TMH) identification is crucial for membrane protein structure prediction.
• Current TMH predictors often lack interpretability, functioning as "black boxes".

Purpose of the Study:

• To introduce SOMRuler, a novel TMH predictor that combines high accuracy with excellent interpretability.
• To develop a method for extracting human-interpretable fuzzy rules from learned knowledge.

Main Methods:

• Utilized a self-organizing map (SOM) to learn helix distribution patterns from training data.
• Extracted fuzzy rules from the SOM's codebook vectors, reducing computational burden and enhancing reliability.
• Analyzed the validity of extracted fuzzy rules both qualitatively and quantitatively.

Main Results:

• SOMRuler demonstrated high prediction accuracy, outperforming most existing popular TMH predictors on a benchmark dataset.
• The method successfully generated interpretable fuzzy rules from learned helix distribution knowledge.
• The extracted fuzzy rules were shown to be reliable and computationally efficient.

Conclusions:

• SOMRuler provides an interpretable yet accurate approach to TMH identification in membrane protein structure prediction.
• The method's flexibility makes it suitable for diverse bioinformatics problems.
• SOMRuler enhances the understanding of TMH prediction by providing transparent, rule-based insights.