InterLabelGO+ combines a deep learning model (InterLabelGO) and sequence homology search to predict a query protein's biological function in the form of Gene Ontology (GO) terms.

InterLabelGO uses the last three layers of the ESM2 large language model to extract sequence features, which are then learned by a series of neural networks to predict GO terms under a new loss function that incorporates label imbalances and inter-label dependencies.

These deep learning-predicted terms are combined with DIAMOND search results through a dynamic weighting scheme to derive consensus predictions.

InterLabelGO+ (as team Evans) was ranked among the top teams in the recent CAFA5 challenge.
Questions about the InterLabelGO+ server? Post issues on the GitHub.
Example output