[us-commits] [ehb54/ultrascan3] 2616d4: Add psv group-contribution probe (analysis for tic...

[emre brookes]

  Branch: refs/heads/ehb54-issue-978
  Home:   https://github.com/ehb54/ultrascan3
  Commit: 2616d4c1ebdfc9a2920da2b7f3e2f0be113dee78
      https://github.com/ehb54/ultrascan3/commit/2616d4c1ebdfc9a2920da2b7f3e2f0be113dee78
  Author: ehb54 <brookes at uthscsa.edu>
  Date:   2026-07-21 (Tue, 21 Jul 2026)

  Changed paths:
    A us_somo/develop/perceiver/tools/psv_model.py

  Log Message:
  -----------
  Add psv group-contribution probe (analysis for ticket #980)

Tests Mattia's suggestion that a rough psv can be suggested for a non-coded
residue from similarity to the hybridizations already in somo.residue, by
fitting a Cohn-Edsall / Durchschlag-Zipper style group-contribution model over
the ~100 genuine polyatomic residues somo.residue codes and scoring it
leave-one-out.

Result (MAE in vbar, cm^3/g; typical vbar ~0.73):
  global mean                     0.110
  sum-of-vdW-spheres (emitted now) 0.193   <- worst option available
  residue-type group average      0.070
  hybrid group-contribution       0.054   (organic only: 0.036, metals: 0.222)

So the estimate the perceiver currently emits is worse than guessing the mean,
the group-contribution model is good for organic residues, and it should not
be trusted for metal-containing ones. Findings recorded on
ehb54/ultrascan-tickets#980.

NOT pushed: this belongs with the #980 work rather than the #978 PR.

Co-Authored-By: Claude Opus 4.8 <noreply at anthropic.com>



To unsubscribe from these emails, change your notification settings at https://github.com/ehb54/ultrascan3/settings/notifications