A Multi-Layer AI-Assisted in silico Pipeline for Generalizable Enzyme Engineering via Computational Directed Evolution

Overview

Enzymes are widely used across healthcare, textiles, and environmental technologies, requiring optimization through enzyme engineering. Traditional approaches rely on wet-lab trial-and-error experimentation, which is costly, time-consuming, and dependent on advanced laboratory equipment, limiting accessibility and enzyme-based technology advancement. This project developed a generalizable, computational pipeline and platform to address these limitations by minimizing wet-lab dependency. It consists of four stages. First, enzyme data from the BRENDA database was filtered to construct a balanced dataset. Second, five regression algorithms were trained and benchmarked; the best-performing model was selected to predict whether variants achieve the adaptation. Third, computational directed evolution generated variant sequences through literature-guided mutations applied to the wild-type sequence, with each iteration evaluated against the target property. Achieving target adaptation, however, doesn't guarantee evolutionary plausibility. Therefore, a fourth stage implemented a multi-layer validation system: ESM-2, a protein language model, assessed sequence-level plausibility, while ESMFold evaluated structure-level confidence. Variants scoring high in both layers represent successfully adapted sequences with strong evolutionary plausibility. As a case study, cold adaptation (55°C to 35°C) was performed on Tc-Cut1 (T. cellulosilytica) and heat adaptation (48°C to 68°C) on Pa-LipA (P. aeruginosa), generating new variants with desired adaptation. This low-cost, modular pipeline establishes an accessible framework for computational enzyme engineering, reducing dependence on costly laboratory infrastructure and making advanced protein design accesible to researchers.

Competition history

• ISEF 2026 Computational Biology and Bioinformatics · Entry CBIO002

Resources

• ISEF abstract page ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.