PPubMed27 Feb 2023Cited 58×

Accelerating crystal structure determination with iterative <i> AlphaFold </i> prediction Iterative <i> AlphaFold </i> prediction

Thomas C. Terwilliger, Pavel V. Afonine, Dorothee Liebschner, Tristan I. Croll, Airlie J. McCoy, Robert D. Oeffner, Christopher J. Williams, Billy K. Poon, Jane S. Richardson, Randy J. Read, Paul D. Adams, A. Gonzalez

Abstract

'Tristan I. Croll' 'Airlie J. McCoy' 'Robert D. Oeffner' 'Christopher J. Williams' 'Billy K. Poon' 'Jane S. Richardson' 'Randy J. Read' 'Paul D. Adams' 'A. Gonzalez'] AlphaFold predictions can be used both as a starting point for structure determination and as a method of model optimization. The Phenix*PredictAndBuild* tool automates iterative prediction and model building, yielding a density map and model starting with sequence information and crystallographic data.

A figure from Accelerating crystal structure determination with iterative <i> AlphaFold </i> prediction Iterative <i> AlphaFold </i> prediction — fig. from the paper

§ The Valyu brief

The paper, in five minutes

Read from the full text via Valyu Contents

Iterative AlphaFold predictions integrated with Phenix automate X‑ray crystal structure determination, achieving ~90% completeness on 215 challenging proteins.

What’s new

i.Developed the Phenix PredictAndBuild tool for fully automated iterative AlphaFold‑guided crystallographic model building.
ii.Demonstrated high success on a benchmark set of 215 SAD‑determined proteins, with 208 yielding molecular‑replacement solutions and average model completeness of 90%.
iii.Showed that using rebuilt models as templates improves AlphaFold predictions, achieving up to 99% residue matching within 2 Å.
iv.Released all data, scripts, and code publicly for reproducibility.

How they did it

Starting from sequence and diffraction data, the workflow cycles through AlphaFold prediction, confidence‑based trimming, molecular replacement, morphing, refinement, and automated rebuilding using Phenix tools. At least three cycles are run until model RMSD change falls below 0.25 × resolution, after which the best model is selected.

What they found

Out of 215 selected SAD structures, 208 (≈97%) produced a molecular‑replacement solution.
Average model completeness was 90%; all but one model reached at least 50% completeness.
Final rebuilt models had a mean free R value of 0.31.
Template‑guided AlphaFold predictions matched 99% of deposited residues within 2 Å, compared to 90% for initial predictions.
AlphaFold predictions showed higher clashscores (mean 27) than deposited models (mean 6).

Caveats

✕Large distortions during morphing can produce implausible models.
✕Unsuccessful cases had low completeness (average 20%) and failed to solve the structure.
✕The method is less effective for structures containing significant non‑protein components (RNA/DNA, ligands).
✕Improvements in density maps were modest compared to gains in prediction accuracy.

§ The Valyu brief

Reading the full paper and taking notes. This takes a few seconds…

§ Ask this paper

Ask a question about this paper

Valyu reads the full text and answers from what the paper actually says.

§ Related work in other archives

Accelerating crystal structure determination with iterative <i> AlphaFold </i> prediction Iterative <i> AlphaFold </i> prediction

The paper, in five minutes

Ask a question about this paper

Similar papers in all four archives

The paper, in five minutes

Similar papers in all four archives