NetSolP: predicting protein solubility in Escherichia coli using language models.

Area of Science:

• Biotechnology
• Computational Biology
• Protein Engineering

Background:

• Protein solubility and expression levels are critical bottlenecks in large-scale biological studies and industrial protein production.
• Accurate prediction of these properties from amino acid sequences can significantly increase the success rate of experimental work.

Purpose of the Study:

• To develop and validate a deep learning model, NetSolP, for predicting protein solubility and purification usability directly from amino acid sequences.
• To address dataset bias in existing prediction methods by curating datasets with strict sequence-identity partitioning.

Main Methods:

• Utilized transformer-based deep learning protein language models to build the NetSolP predictor.
• Curated existing datasets using strict sequence-identity partitioning to minimize bias.
• Evaluated model performance, focusing on state-of-the-art results and improved extrapolation capabilities.

Main Results:

• NetSolP achieves state-of-the-art performance in predicting protein solubility and purification usability.
• The model demonstrates improved extrapolation across different datasets due to unbiased data curation.
• The predictor, data, and open-sourced code are publicly available.

Conclusions:

• Deep learning models like NetSolP can accurately predict protein solubility and purification potential from sequence data.
• Addressing dataset bias is crucial for developing robust and generalizable protein property prediction tools.
• NetSolP offers a valuable resource for researchers and industry to optimize protein expression and purification strategies.