Facilitating Machine Learning-Guided Protein Engineering with Smart Library Design and Massively Parallel Assays.

Area of Science:

• Protein engineering and computational biology.
• Biophysics and molecular biology.

Background:

• Protein design is crucial for medical advances like antibody and vaccine development.
• Traditional methods like mutational screens and directed evolution are limited by scale for combinatorial mutations.
• High-throughput screening and Next-Generation Sequencing increase read-out scale but cannot survey all variants.

Purpose of the Study:

• To describe advances in massive-scale variant screens.
• To explain the integration of machine learning and mutagenesis for accelerated protein engineering.
• To examine strategies for economical, informative, and effective variant discovery.

Main Methods:

• Utilizing in-silico approaches with machine learning to predict variant fitness.
• Employing a subset of empirical measurements to train machine learning models.
• Integrating machine learning with mutagenesis strategies and massive-scale variant screens.

Main Results:

• Machine learning models predict fitness of novel protein variants.
• Models can guide the search for high-fitness variants and resolve epistasis.
• Machine learning facilitates the creation of focused libraries, reducing experimental burden.

Conclusions:

• Machine learning-guided protein engineering accelerates the optimization process.
• Integrated strategies enhance the economy, informativeness, and effectiveness of variant discovery.
• This approach fast-tracks the development of useful protein variants for various applications.