PyPEF-An Integrated Framework for Data-Driven Protein Engineering.

Area of Science:

• Biotechnology and Bioinformatics
• Computational Biology and Cheminformatics

Background:

• Data-driven strategies are increasingly vital in protein engineering, driven by advances in experimental data, sequencing, high-throughput screening, and AI.
• Predicting beneficial amino acid substitutions and their impact on protein function remains a key challenge in developing proteins for biocatalysis, biomedicine, and life sciences.

Purpose of the Study:

• To introduce PyPEF (pythonic protein engineering framework), a general-purpose framework for data-driven protein engineering.
• To enable efficient identification and selection of beneficial protein variants within a defined sequence space using machine learning and signal processing techniques.

Main Methods:

• Development of the PyPEF framework integrating machine learning, signal processing, and statistical physics.
• Systematic and random exploration of protein variant fitness and sampling of random evolution pathways.
• Evaluation of PyPEF's predictive accuracy and throughput on four public protein and enzyme datasets using regression models.

Main Results:

• PyPEF efficiently predicts protein sequence fitness for various properties, achieving coefficients of determination from 0.58 to 0.92.
• The framework enables screening over 500,000 protein sequence variants in minutes on a personal computer.
• Demonstrated significant accuracy across diverse datasets, highlighting the integration of data-driven technologies for protein evolution.

Conclusions:

• PyPEF offers a powerful solution for sequence exploration and combinatorial challenges in protein engineering.
• The framework facilitates exhaustive in silico screening of sequence spaces, accelerating the discovery of proteins with desired functions.
• PyPEF supports different data-driven philosophies, from high-accuracy prediction of epistatic effects to capturing general mutation trends in directed evolution.