An iterative deep learning-guided algorithm for directed protein evolution.

Area of Science:

• Biochemistry
• Computational Biology
• Protein Engineering

Background:

• Deep learning shows promise for protein optimization but faces limitations in improving protein activity.
• Current algorithms often lack rigorous iterative evaluation, a key component of protein engineering.

Purpose of the Study:

• To introduce DeepDE, an iterative deep learning-guided algorithm for protein optimization.
• To evaluate DeepDE's effectiveness in enhancing protein activity using a focused library and triple mutants.

Main Methods:

• Developed DeepDE, an iterative deep learning algorithm utilizing triple mutants as building blocks.
• Trained the algorithm on a compact library of approximately 1,000 mutants.
• Applied DeepDE to optimize Green Fluorescent Protein (GFP) from *Aequorea victoria* over four rounds of evolution.

Main Results:

• DeepDE achieved a 74.3-fold increase in GFP activity, surpassing the benchmark superfolder GFP.
• The use of triple mutants enabled exploration of a larger sequence space per iteration.
• Limited screening of around 1,000 variants per round significantly improved DeepDE's performance.

Conclusions:

• DeepDE offers a robust and iterative approach to protein engineering, significantly enhancing protein activity.
• The algorithm effectively addresses data sparsity challenges in protein engineering through focused library screening.
• This deep learning-guided strategy represents a promising advancement over traditional methods for protein optimization.