NAC4ED

Summary

NAC4ED automates high-throughput screening of enzyme point mutants by computing, for each candidate variant, the population of near-attack conformations (NACs) sampled during molecular dynamics — geometries in which the reactive atoms lie within van der Waals contact and near the transition-state bond angle. This avoids the cost of explicit transition-state search or QM/MM calculation. It still grounds predictions in a physical model of the rate-limiting step.[1]

Method

The platform has four sequential modules: mutation (generates mutant structures via PyMOL/AMBER), docking (uses Glide/LigPrep to dock substrate poses into the mutant, defining a search box around key active-site residues), molecular dynamics (500 ps NPT simulation of the docked complex via AMBER's tLEaP), and evaluation (CPPTRAJ-based measurement of the reactive-atom distance/angle across the trajectory, from which the fraction of frames meeting the NAC criterion is computed as a proxy for catalytic competence).[1]

Validation

• Epoxide hydrolase (TrEH) mutagenesis scan: 1,843 candidate variants (mutations at 97 residues within 12 Å of the substrate) were screened computationally in 192 h on 2 GPUs/40 CPU cores. The predicted top variant, L89Y (20.9% active-conformation population vs. 15.6% for wild type), showed 115% relative activity experimentally; a second top-ranked variant, G119I (24.3% population), could not be validated because it failed to express — illustrating that NAC4ED's activity-population score does not capture expression/solubility effects.[1]
• Amine transaminase benchmark (40 literature-curated 3FCR mutants): NAC4ED correctly classified variants as higher- or lower-activity than wild type with 92.5% accuracy (predicted 13 high/24 low vs. 15 high/25 low observed).[1]
• Throughput: a single variant's active-conformation population is computed in 13 min on one GPU + one CPU, versus ~50 min for manual setup and ~7 days for full experimental characterization (expression, purification, and activity assay) — a reported 764-fold speedup per variant, and 1,613-fold across the full 1,843-variant TrEH screen relative to exhaustive experimental characterization.[1]

Availability

Available at lujialab.org.cn/software.

See Also

• Predicting Catalytic Competence of Enzyme-Ligand Complexes — framework page; NAC4ED sits at the "does the productive pose persist dynamically" stage, between docking and full QM/MM.
• Enzyme Kinetic Parameter Prediction — the adjacent, purely sequence/ML-based route to ranking mutants that NAC4ED's physically grounded approach complements.

Citations

[1] Zhang, C., Feng, Y., Zhu, Y., Gong, L., Wei, H., Zhang, L. (2024). NAC4ED: A high-throughput computational platform for the rational design of enzyme activity and substrate selectivity. mLife, 3(4), 505–514. Supports: method and all validation/throughput figures above. Location: Abstract; Results ("Efficient screening of EH mutations"; "Generic validation of the NAC4ED platform").