NetSolP

NetSolP predicts protein solubility and usability for Escherichia coli expression systems directly from amino acid sequences.

Key Features:

• Deep Learning Architecture: Uses transformer-based protein language models to learn sequence-derived representations for solubility and usability prediction.
• Performance: Demonstrates high predictive accuracy compared to existing sequence-based methods.
• Bias Mitigation in Datasets: Employs strict sequence-identity partitioning during dataset curation to reduce dataset bias.
• Extrapolation Across Datasets: Shows improved extrapolation capabilities when applied to datasets distinct from the training data.

Scientific Applications:

• Optimizing Protein Expression: Enables selection or engineering of sequences with higher predicted solubility for Escherichia coli expression.
• Reducing Experimental Failures: Identifies sequences likely to produce insoluble or unusable protein to prioritize experimental efforts.
• Facilitating Protein Engineering: Guides sequence modification strategies aimed at improving solubility and usability.

Methodology:

NetSolP employs transformer-based protein language models trained on curated datasets with strict sequence-identity partitioning to process amino acid sequences and predict solubility and usability.