RP3Net: a deep learning model for predicting recombinant protein production in Escherichia coli

Area of Science:

• Biotechnology
• Computational Biology
• Protein Engineering

Background:

• Recombinant protein expression is crucial for drug discovery and biotechnology.
• Current methods face limitations in predicting protein production efficiency.
• AI offers a potential solution to optimize recombinant protein expression.

Purpose of the Study:

• To develop an AI model for predicting small-scale heterologous soluble protein expression in E. coli.
• To leverage recent advancements in protein and genomic foundational models.
• To improve the efficiency of protein reagent production for biotechnology applications.

Main Methods:

• Developed RP3Net (Recombinant Protein Production Prediction Network), an AI model.
• Utilized recent protein and genomic foundational models.
• Trained, validated, and tested RP3Net on curated data from AstraZeneca and the Structural Genomics Consortium.

Main Results:

• RP3Net demonstrated a significant increase in Area Under Receiver Operator Curve (AUROC) by 0.15 compared to baseline models.
• Experimental validation on 97 constructs showed RP3Net achieved an AUROC of 0.83.
• RP3Net accurately predicted expression in 77% of cases and identified successful constructs in 92% of cases.

Conclusions:

• RP3Net enhances the prediction of recombinant protein expression in E. coli.
• The AI model outperforms existing methods, improving efficiency in protein production.
• RP3Net is available under an MIT license, promoting accessibility for research and development.