Bioinformatics case studies

Reproducible sequence and structure checks, then a blinded second opinion

AI Gene Review · projects/BIOINFORMATICS · 2026

Bottom line

  • Two S. pombe cases where a family-based function was tested by script: pmp20 (peroxiredoxin) and Epe1 (JmjC demethylase).
  • pmp20: one cysteine (C43), no resolving Cys; review REMOVEs IBA thioredoxin peroxidase activity and keeps the curated NOT peroxidase activity. Epe1: 7 catalytic rows REMOVE.
  • Both were re-tested as blinded OpenScientist hypotheses (July 2026); the workflow now has 264 -hypotheses/ folders.

Why a reproducible case, not a one-off

  • A one-time analysis on one gene is weak evidence and hard to audit.
  • pmp20-bioinformatics runs end to end with just all, takes table-driven inputs, and was re-run on a second target (tpx1) to prove the scripts are general.
  • Every script compares the target with active controls, so "missing residue" is measured, not assumed.
  • It is now the engineering template for new cases.

pmp20: where the second cysteine should be

Blinded second opinion

What the reviews say now

Gene Row Evidence Action
pmp20 obsolete thioredoxin peroxidase activity (GO:0008379) IBA REMOVE
pmp20 NOT peroxidase activity (GO:0004601) IDA ACCEPT
pmp20 unfolded protein holdase activity (GO:0140309) IDA ACCEPT
Epe1 histone demethylase activity (GO:0032452) IBA REMOVE
Epe1 histone H3K9 demethylase activity (GO:0032454) IDA, EXP REMOVE

Read from genes/SCHPO/pmp20/pmp20-ai-review.yaml and genes/SCHPO/Epe1/Epe1-ai-review.yaml.

Lessons

  1. Blinding pays: the Epe1 run found the real defect (Tyr370 at the third Fe(II) ligand) that the older local script missed.
  2. A second run can disagree usefully: one pmp20 run noted that Cp-only is the normal 1-Cys peroxiredoxin layout, so the case rests on the negative assay (PMID:20356456), with sequence evidence as support.
  3. Structure models were used as monomers only; distances are context, not activity evidence.

Status and next steps

  • ✅ Two cases written up; both reviews cite their file: RESULTS and the OpenScientist reports.
  • ⬜ The case-study page still lists the Epe1 run as "expected"; add its outcome as BIO-002 findings.
  • ⬜ Add new cases using the template section of the page.

Read more: projects/BIOINFORMATICS.md · genes/SCHPO/pmp20/pmp20-bioinformatics/RESULTS.md · genes/SCHPO/*/…-hypotheses/