There is a newer version of the record available.

Published October 4, 2026 | Version v3

Neither AlphaFold3 confidence nor Boltz-2 affinity measures substrate discrimination in threonyl-tRNA synthetase: data, code and manuscript

Authors/Creators

  • 1. Department of Chemistry and Physics, McNeese State University, Lake Charles, LA, USA

Description

Complete data, code and manuscript for the preprint of the same title. Contains AlphaFold3 and Boltz-2 predictions, derived per-sample tables, measured kinetics, analysis code and a self-checking reproduce.py that regenerates every number reported in the paper using the Python standard library alone.

What it shows. (1) Across 18 non-cognate amino acids, no AlphaFold3 confidence observable tracks the published substrate discrimination factors of threonyl-tRNA synthetase; against the binding-step component all three take the wrong sign. (2) They track residue volume instead (rho = -0.58 to -0.63), and controlling for volume leaves nothing behind, so the scores read steric similarity to the cognate rather than discrimination. Serine, the non-cognate this enzyme rejects best at the binding step, scores highest of all 18. (3) A trained affinity head (Boltz-2, 5 seeds) behaves differently: correct sign in 5 seeds of 5 (rho = +0.470 against I-prime, p at 0.05; rho = -0.600 against D for the binder probability, p = 0.009) and no steric confound (rho against volume -0.140; partial correlation given volume undiminished at +0.458). It still does not resolve discrimination, ranking serine 11th of 18 and placing cysteine and tryptophan above the cognate. (4) Neither readout is blind to catalysis: alanine-substituting the three zinc ligands collapses the threonine-over-valine preference by 0.112 (95% CI 0.106-0.117) and inverts it, with no change in predicted fold.

Measured discrimination factors are transcribed with attribution from Freist, Sternbach & Cramer (1994) Eur J Biochem 220:745-752, doi:10.1111/j.1432-1033.1994.tb18675.x. Residue volumes from Zamyatnin (1972) Prog Biophys Mol Biol 24:107-123. Boltz-2 from Passaro et al. (2025), doi:10.1101/2025.06.14.659707.

Notes

reproduce.py runs on the Python standard library alone and self-checks 38 reported values against the shipped figure source data.

Files

aars-af3-calibration-v1_code-data-manuscript.zip

Files (225.6 MB)

Name Size
md5:a71866e0b5cc3cb2108cca49d40c3d84
1.1 MB Preview Download
md5:07bc40ad7a8b6ba3ae022b4f3ebefafe
91.9 MB Download
md5:68c21d52bc89ef5ac37032fd63ad5231
40.4 MB Download
md5:b66a3a87ddcec75ed30864290107a0b5
11.7 kB Preview Download
md5:25d32d5f790db9865cf2edb27df9f9a9
7.2 kB Download
md5:7cf2cf568800c32998f20363fde38cb9
62.7 MB Download
md5:21fa6eace98f5b943798e565b22313cb
29.4 MB Download