ProtNote: a multimodal method for protein-function annotation.

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• The protein sequence-function relationship is crucial for protein biology and engineering.
• Less than 1% of known protein sequences have experimentally verified functions.
• Current deep learning models for protein function prediction are limited to trained functions.

Purpose of the Study:

• To introduce ProtNote, a multimodal deep-learning model for protein function prediction.
• To enable both supervised and zero-shot protein function prediction using free-form text.
• To generalize predictions to unseen and novel protein functions.

Main Methods:

• Developed a multimodal deep-learning model named ProtNote.
• Leveraged free-form text for protein function prediction.
• Evaluated performance on supervised and zero-shot tasks, including novel Gene Ontology and Enzyme Commission number predictions.

Main Results:

• ProtNote achieves near state-of-the-art performance on its training set.
• ProtNote generalizes effectively to unseen and novel protein functions in zero-shot settings.
• ProtNote outperforms baseline models in predicting novel Gene Ontology and Enzyme Commission numbers, capturing nuanced sequence-function relationships.

Conclusions:

• ProtNote enhances protein function discovery by allowing unrestricted free text inputs.
• The model unlocks biological use cases previously inaccessible due to limitations of predefined labels.
• ProtNote is a valuable tool for navigating the dynamic landscape of protein biology and engineering.