SELFprot: Effective and Efficient Multitask Finetuning Methods for Protein Parameter Prediction.

Area of Science:

• Computational Biology
• Machine Learning
• Bioengineering

Background:

• Predicting protein-ligand interactions and enzymatic kinetics is crucial but challenging.
• Existing computational methods often require extensive parameters and computational resources.

Purpose of the Study:

• To introduce SELFprot, a novel suite of transformer-based machine learning models.
• To improve the accuracy and efficiency of predicting biochemical interactions, including enzymatic kinetics and binding affinities.

Main Methods:

• Utilized ESM2-35M for protein and small molecule embeddings.
• Employed multitask learning and parameter-efficient finetuning (low-rank adaptation).
• Implemented ensemble learning techniques for enhanced robustness and reduced prediction variance.

Main Results:

• Achieved competitive performance on BindingDB and CatPred-DB datasets.
• Demonstrated notable improvements in parameter-efficient prediction of kinetic (k, K, IC, EC) and binding values.
• Showcased comparable accuracy to existing models with an order of magnitude fewer parameters.

Conclusions:

• SELFprot offers a versatile and efficient solution for protein-ligand interaction studies.
• The model's parameter efficiency makes it a valuable tool for bioengineering applications.
• SELFprot advances the prediction of complex biochemical interactions.