henhaobio
Sign in

bioRxivGenomicsPreprint

Regional missense constraint improves variant deleteriousness prediction

Kaitlin E. SamochaiD, Jack A. KosmickiiD, Konrad J. KarczewskiiD, Anne O’Donnell‐LuriaiD, Emma Pierce‐Hoffman, Daniel G. MacArthuriD, Benjamin M. NealeiD, Mark J. DalyiD

PDFDOI
In-depth full-text digest available · Jump to digest ↓

Abstract

Given increasing numbers of patients who are undergoing exome or genome sequencing, it is critical to establish tools and methods to interpret the impact of genetic variation. While the ability to predict deleteriousness for any given variant is limited, missense variants remain a particularly challenging class of variation to interpret, since they can have drastically different effects depending on both the precise location and specific amino acid substitution of the variant. In order to better evaluate missense variation, we leveraged the exome sequencing data of 60,706 individuals from the Exome Aggregation Consortium (ExAC) dataset to identify sub-genic regions that are depleted of missense variation. We further used this depletion as part of a novel missense deleteriousness metric named MPC. We applied MPC to de novo missense variants and identified a category of de novo missense variants with the same impact on neurodevelopmental disorders as truncating mutations in intolerant genes, supporting the value of incorporating regional missense constraint in variant interpretation.

In-depth summary

Claude Opus 5.5 · xhigh effort · 17 Oct 2026 · from the full text of version 1 (12 Jun 2017) · CC BY 4.0

Sub-genic missense depletion in ExAC anchors MPC, a deleteriousness score; de novo missense variants with MPC ≥ 2 occur in neurodevelopmental cases at 5.79 times the control rate.

Missense effects depend on both position and the exact amino acid change, and earlier predictors (SIFT, PolyPhen-2, CADD) rely mostly on cross-species conservation and protein structure. This work adds within-gene depletion of missense variation across 60,706 Exome Aggregation Consortium (ExAC) exomes; few earlier scores used human constraint, and one that did, M-CAP, includes gene-level intolerance (RVIS) as one of many features.

Key results

  1. ShownDe novo missense variants with high MPC scores carry neurodevelopmental-disorder risk close to that of de novo protein-truncating variants in loss-of-function-intolerant genes, while low-MPC variants show no excess.

    5620 cases vs 2078 controls: MPC ≥ 2, 0.162 vs 0.028 de novo missense variants per exome (rate ratio 5.79, p < 10-50); 1 ≤ MPC < 2, rate ratio 1.49 (p = 3.70x10-9); MPC < 1, 0.507 vs 0.504 (rate ratio 1.01, p = 0.885). Earlier work put de novo truncating variants absent from ExAC in genes with pLI ≥ 0.9 at a rate ratio of 6.7. Control rates were scaled by ~1.14 for the higher case synonymous rate. · Using MPC to evaluate the deleteriousness of de…; Discussion; Introduction; Correcting the rate of de novo variants in…

  2. ShownWhen case and control de novo missense variants are ranked together, the top-scoring tenth by MPC is more enriched for case variants than the top tenth by PolyPhen-2, CADD or M-CAP.

    Pooled 6382 de novo missense variants (5113 from cases, baseline case proportion 0.8); top ~10% by MPC: odds ratio 5.43 (Fisher's exact p < 10-27) vs 1.44 for PolyPhen-2 (p = 2.66x10-3). All four scores were significantly enriched, but the CADD and M-CAP odds ratios are in Table S13, which is not in this text. · Using MPC to evaluate the deleteriousness of de…

  3. ShownAbout 15% of human transcripts contain sub-regions with significantly different tolerance to missense variation.

    Poisson likelihood-ratio tests on observed vs expected rare (minor allele frequency, MAF < 0.1%) ExAC missense counts, summarized as γ (fraction of expected variants observed), found regional differences in 2700 of 17,915 transcripts (15.1%): 1717 with two segments, 904 with three, 79 with four or more. These transcripts are longer (2753 vs 1530 coding bp; Wilcox p < 10-50), as expected from greater detection power. · Searching for regional missense constraint…; Likelihood ratio tests to define regional…

  4. ShownRegions keeping at most 60% of their expected missense variation make up a small share of coding sequence but hold most known pathogenic missense variants in severe haploinsufficiency genes.

    Regions with γ ≤ 0.6 span 14% of the coding region but hold 89% of 404 ClinVar pathogenic missense variants in 49 severe haploinsufficient genes (6.2-fold; p < 10-50); 32 of these genes (65%) vary regionally. In CDKL5, the first 283 amino acids keep 24% of expected missense variants vs 82% for the other 748, and 39 of 43 (91%) pathogenic or likely pathogenic missense variants sit in the constrained part. · ClinVar variants and regional depletion; ClinVar pathogenic variants

  5. ShownThe excess of de novo missense variants in neurodevelopmental cases is concentrated in missense-constrained regions, which shrinks the pool of candidate pathogenic variants.

    With corrected control rates, γ ≤ 0.6 regions carry 0.24 de novo missense variants per case exome vs 0.08 per control exome, and γ > 0.6 regions 0.58 vs 0.56; odds ratio 2.1 for 0.4 < γ ≤ 0.6 and > 4.9 for γ ≤ 0.4. Control variants spread across constraint bins in proportion to coding bases (p = 0.1262). Candidates drop from 4683 to 1374, yet 71% of case de novo missense variants lie outside constrained regions. · Using regional constraint to interpret de novo…; Correcting the rate of de novo variants in…

Methods, data and reproducibility

Poisson likelihood-ratio tests on observed vs expected rare missense counts (60,706 ExAC exomes) tested 17,915 transcripts for regions of differing depletion (γ); 'missense badness' scores the extra depletion of each amino acid substitution in constrained regions. Logistic regression (ClinVar pathogenic vs common ExAC variants) merged γ, missense badness and PolyPhen-2 into MPC, which was tested on de novo variants.

  • Exome Aggregation Consortium (ExAC) exomes: rare (MAF < 0.1%) missense variants for constraint estimation; common (MAF > 1%) missense variants as the benign training set · human · n = 60,706 individuals; benign set 82,932 variants (88,083 in Methods)
  • ClinVar (July 9, 2015 release) pathogenic and likely pathogenic missense variants in haploinsufficient genes causing severe disease · human · n = 404 variants in 49 genes (402 with all metrics)
  • De novo variants from neurodevelopmental disorder trios: intellectual disability/developmental delay (ID/DD) and epileptic encephalopathy · human · n = 5620 cases: 5264 ID/DD trios + 356 epileptic encephalopathy trios
  • De novo variants from unaffected siblings of autism spectrum disorder cases, used as controls · human · n = 2078
  • Canonical GENCODE v19 protein-coding transcripts after quality and outlier filtering · human · n = 17,915 transcripts
Methods & tools
MPC, ExAC, ClinVar, ClinGen Dosage Sensitivity Map, GENCODE v19, Variant Effect Predictor (VEP) version 81, vt normalize, PolyPhen-2, BLOSUM, Grantham, CADD, M-CAP
Reagents & environment
Annotation versions are stated (canonical GENCODE v19 transcripts from Ensembl 75, VEP version 81 on GRCh37, ClinVar release of July 9, 2015, ClinGen Dosage Sensitivity Map downloaded May 5, 2015); no software or package versions for the analysis itself.
Reproducibility gaps
No code availability statement; links in the URLs section (ExAC, ClinVar, OMIM, ClinGen, CADD) are missing from this text. Regional constraint (Table S4), missense badness (Table S10) and MPC for all possible missense variants (Table S14) are said to be supplementary tables, not included here.
Limitations (6)
  • Even 60,706 ExAC individuals give too little power to estimate constraint per base, so variant counts were aggregated by exon, with breakpoints refined to the amino acid level. · stated by the authors
  • The method treats coding sequence linearly: binding pockets formed by residues scattered along the protein, other 3D structural features, and noncoding sequence are not evaluated. · stated by the authors
  • Most (71%) case de novo missense variants lie in regions that are not missense constrained (γ > 0.6), where the case:control rate ratio is indistinguishable from 1. · stated by the authors
  • Cases were sequenced more recently than controls and have a higher de novo synonymous rate (0.287 vs 0.251 per exome); control rates were therefore scaled by ~1.14, a correction the authors call conservative. · stated by the authors
  • MPC was fit on a narrow pathogenic set (404 ClinVar missense variants in 49 severe haploinsufficient genes) against common ExAC variants, and its only independent test is neurodevelopmental de novo variants; recessive or other disease classes are not evaluated. · noted in this summary
  • The tool comparison is the authors' own design with a single top-~10% cut-off, the CADD and M-CAP odds ratios are only in Table S13 (absent here), and benign-set sizes differ between Results and Methods (82,932 vs 88,083). · noted in this summary

Read this if you interpret de novo or exome missense variants, especially in neurodevelopmental disorder cohorts, and want a constraint-aware deleteriousness score or the regional missense constraint map behind it.

Citations
351

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISWang '24Zhang '22Pérez‐Palma '20Muneeb '26Ancien '21Huang '20
86% semantic matchGenomics
Lily Wang, Katherine R. Chao, Ruchit Panchal+29 more·bioRxiv·Apr 11, 2024· 13 citations