Predict Protein NMR Chemical Shifts Effortlessly with LEGOLAS on Vecura

What is LEGOLAS?

LEGOLAS (Learning Ensemble Graph-based Optimization for Learned Atomic Shifts) is a high-performance machine learning model designed to predict protein NMR chemical shifts directly from three-dimensional atomic coordinates. Built on PyTorch and utilizing an ensemble of 30 neural networks, it provides rapid, accurate predictions for the six most informative atom types: H, HA, CA, CB, C, and N. It helps users bridge the gap between computational structural biology and experimental NMR spectroscopy by allowing for the assessment of structural consistency at the atomic level. It is especially useful for structural biologists and computational chemists needing to validate structural models, screen molecular dynamics trajectories, or cross-check homology models.

What can users do with LEGOLAS on Vecura?

With LEGOLAS on Vecura, users can:

• Generate per-residue NMR chemical shift predictions from PDB structures or molecular dynamics trajectories.
• Assess structural model quality by comparing predicted shifts against experimental BMRB data.
• Screen large MD ensembles to identify frames that are most consistent with experimental spectra.
• Use the ensemble standard deviation as a confidence indicator to pinpoint residues that warrant closer scrutiny or experimental follow-up.

What the output means

The output provides a detailed table containing the ensemble mean chemical shift (in ppm) and the ensemble standard deviation for each residue and atom type. Additionally, for PDB inputs, it generates annotated PDB files where these predicted values are encoded into the B-factor column, facilitating direct visual inspection in molecular viewers.

This output should be used to support scientific decision making. It does not replace experimental validation.

Why this matters

The ability to predict NMR chemical shifts accurately and quickly is a cornerstone of modern structural biology, serving as a powerful link between experimental data and computational models. By enabling researchers to validate structural hypotheses in silico before committing to expensive and time-consuming bench experiments, tools like LEGOLAS significantly accelerate the discovery process.

By providing not just a prediction but also an uncertainty metric (the standard deviation), LEGOLAS allows researchers to quantify the reliability of their models, ensuring that scientific conclusions are based on robust and high-confidence data.

• Developed by: Roitberg Group
• Source: Official GitHub Repository
• Reference: J. Chem. Theory Comput. 2025, 21(8), 4266-4275