Exploring the landscape of enzyme design: from molecular simulations to machine learning

Abstract

Protein design and engineering are essential across biotechnology, biomedicine, and life sciences to develop enzymes with tailored properties. Yet, identifying such enzymes within the vast sequence space remains a formidable challenge. Advances in deep learning–based protein structure prediction (e.g., AlphaFold2) and molecular simulations techniques have expanded structural and mechanistic knowledge, enabling rational enzyme redesign. However, large-scale discovery and design of novel enzymes is still constrained by limited screening throughput.

In this talk, I will present my group’s research on molecular simulation and machine learning (ML) strategies to accelerate protein and enzyme design. Simulation-based approaches reveal structure–function principles and guide the design of enzymes for diverse applications—for example, stabilizing biocatalysts in non-conventional media such as organic solvents and ionic liquids, or tailoring protein interactions with materials and light. In parallel, ML enables exploration of vast sequence landscapes by uncovering sequence–fitness relationships. I will introduce our data-driven protein engineering platforms which integrate evolutionary information with ML to predict protein fitness. These methods provide predictive power even with limited datasets, significantly accelerating protein engineering experiments.

Finally, I will discuss how combining computational approaches opens new opportunities for sustainable enzyme design, and why curated datasets and interpretable algorithms are critical for responsible innovation in protein engineering.