Accurate and efficient protein sequence design through learning concise local environment of residues.

Area of Science:

• Protein engineering and computational biology.

Background:

• Computational protein sequence design is crucial for rational protein engineering.
• Improving the accuracy and efficiency of protein design methods is a key challenge.

Purpose of the Study:

• To present ProDESIGN-LE, an accurate and efficient computational approach for protein sequence design.
• To validate the performance of ProDESIGN-LE through computational predictions and experimental testing.

Main Methods:

• ProDESIGN-LE utilizes a transformer model trained on residue local environments to predict amino acid types for a given backbone structure.
• The method represents local environments concisely and learns correlations between environments and residue types.
• Designed sequences were generated for naturally occurring and hallucinated proteins, and experimentally validated for an enzyme.

Main Results:

• ProDESIGN-LE achieved high accuracy in designing protein sequences, with predicted structures closely matching target structures (average TM-score > 0.80).
• The design process was efficient, averaging under 20 seconds per protein.
• Experimental validation showed that three out of five designed enzyme sequences exhibited excellent solubility, and one yielded monomeric species with spectra consistent with the natural enzyme.

Conclusions:

• ProDESIGN-LE is an accurate and efficient tool for computational protein sequence design.
• The method demonstrates potential for designing functional proteins with high structural fidelity and solubility.