{
  "id": 16446902,
  "title": "Two opposite failure modes in alignment-based conservation scores, one inflating and one concealing constraint, are computable before the score is used",
  "url": "https://urgent.news/2026/10/24/two-opposite-failure-modes-in-alignment-based-conservation-scores-one",
  "topic": "science",
  "section": "Science",
  "published": "2026-10-24T00:00:00.000Z",
  "source": {
    "name": "bioRxiv",
    "slug": "biorxiv",
    "url": "https://www.biorxiv.org/content/10.64898/2026.10.20.760363v1?rss=1"
  },
  "original_language": "en",
  "account": "Two opposing factors within alignment-based conservation scores can either overestimate or underestimate constraint, both of which can be calculated prior to scoring. The first phenomenon causes the reported conservation value to inflate, while the second leads to its underestimation. These failures are computable from the alignment before the score is generated.\n\nThe first failure mode is characterized by the inflation of conservation scores, as reflected in a positive correlation (rho) of +0.553 across 55,816 columns. When conservation is recalculated over ungapped residues, this inflated value drops to a more modest +0.066, providing an independent predictor for the discrepancy. AlphaMissense, an independent predictor, shows a rho of 0.551 for the gap-free recomputation compared to 0.416 for the reported value.\n\nThe second failure mode stems from clade partitioning, where divergence between subfamilies is mistaken for tolerance. By altering the comparator set, a position can shift by an average of 2.19 grades on a nine-grade scale, affecting 5,092 positions across eight families. This shift can change 2.7 to 8.3 percent of columns from variable to conserved, with the median change being 6.6 percent. A label-permutation null cannot differentiate this effect from shared ancestry, leading to nearly all variable columns being erroneously identified as conserved. A tree-aware null, which matches the per-column parsimony rate, yields a range of 4.1 to 69.1 percent, depending on the sample depth.\n\nThe gap fraction, effective comparator count, and subfamily-explained fraction can be generated as a panel of values, each displaying the expected direction of error. An interesting prediction emerges that variants in a sister subfamily's fixed state are enriched for benign effects, supported by an odds ratio of 2.46 (1.40 to 4.33) among 87 such variants out of 1,565 classified. Applying the gap correction to an existing conserved-score pipeline significantly improves the rigid score's discrimination in a human cohort, raising the AUC from 0.832 to 0.919.",
  "summary": "A conservation score is read as a property of an amino acid position, when it measures that position together with a sequence set and a depth parameter, and the same position receives opposite calls under defensible choices of the latter two. We characterise two failure modes that push in opposite directions and show that both are computable from the alignment before the score is used. Where…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}