Mapping the glycosyltransferase fold landscape using interpretable deep learning.

Area of Science:

• Biochemistry and Structural Biology
• Bioinformatics and Computational Biology
• Glycobiology

Background:

• Glycosyltransferases (GTs) are crucial enzymes involved in synthesizing complex carbohydrates and glycosylating various substrates, impacting nearly all cellular functions.
• The vast structural and functional diversity of GTs poses a significant challenge for traditional bioinformatics methods in mapping sequence-structure-function relationships.
• Predicting GT folds is essential for understanding their evolutionary relationships and functional roles.

Purpose of the Study:

• To develop a highly accurate deep learning model for predicting glycosyltransferase (GT) folds.
• To overcome the limitations of sequence alignment-based approaches by utilizing secondary structure representations.
• To classify GT families, including those with unknown or variant folds, and expand the known GT fold landscape.

Main Methods:

• Development of a convolutional neural network with attention (CNN-attention) deep learning model.
• Leveraging simple secondary structure representations derived from primary amino acid sequences for fold prediction.
• Training the model to identify distinguishing secondary structure features independent of primary sequence alignment.

Main Results:

• The CNN-attention model achieved high accuracy in predicting GT folds.
• The model effectively learned secondary structure features, enabling classification of evolutionarily divergent GT families based on shared structural characteristics.
• The model successfully classified GT families with unknown or variant folds, identifying potential novel folds (e.g., GT91, GT96, GT97).

Conclusions:

• The developed deep learning approach provides an accurate and interpretable method for GT fold prediction.
• This study expands the known glycosyltransferase fold landscape and highlights potential novel folds.
• The findings prioritize specific GT families for future structural and functional investigations.