DIProT: A deep learning based interactive toolkit for efficient and effective Protein design.

Area of Science:

• Computational Biology
• Protein Engineering
• Bioinformatics

Background:

• The protein inverse folding problem, crucial for designing novel proteins, remains a significant challenge.
• Existing methods often lack user-friendliness and integration of diverse approaches for in-silico protein design.

Purpose of the Study:

• To present DIProT, an interactive toolkit that integrates data-driven and knowledge-driven methods for protein design.
• To provide a user-friendly platform for in-silico protein design, enabling evaluation and iterative refinement.

Main Methods:

• DIProT utilizes a non-autoregressive deep generative model for solving the inverse folding problem.
• The toolkit integrates a protein structure prediction model to facilitate in-silico evaluation of designed sequences.
• It supports incorporating user prior knowledge and a virtual design loop with human feedback.

Main Results:

• The inverse folding model shows competitive effectiveness and efficiency on benchmark datasets (TS50, CATH4.2).
• Promising results were achieved in sequence recovery and inference time.
• Case studies demonstrate DIProT's utility in facilitating user-guided protein design.

Conclusions:

• DIProT offers an integrated and interactive solution for the inverse folding problem.
• The toolkit empowers users to design protein sequences efficiently and iteratively.
• DIProT advances the field of in-silico protein design by combining generative models and user feedback.