Contested result

VEST4 and ESM-1b: Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML) reverses the order

Low severity
Not yet reviewed
Ordering reversal
Different tasks across studies

Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML) ranks ESM-1b above VEST4 in 6 of 7 comparisons; other studies rank VEST4 above ESM-1b in 7 of 8.

An automatic lead, not a finding of error. The results below are shown as published, everywhere on the site, until a reviewer decides otherwise.

Which studies agree

2 papers rank VEST4 above ESM-1b in 7 of their 10 comparisons and ESM-1b higher in 1; 2 are ties. Livesey and Marsh 2025 ranks ESM-1b above VEST4 in 6 of its 7 comparisons and VEST4 higher in 1.

Rank VEST4 higher (2): An openly licensed benchmark…, Brandes et al. 2023
Rank ESM-1b higher (1): Livesey and Marsh 2025 (disagrees)
Mixed (0): none

Each paper sits where at least two thirds of its comparisons point; Mixed means neither model reaches two thirds. The larger, outlined dot is the paper that disagrees.

VEST4 higherTieESM-1b higherNumbers are counts of comparisons; the outlined bar is the paper that disagrees.

Papers compared (3)
Paper	Role	Ranks VEST4 higher	Ranks ESM-1b higher	Ties
Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML) and 1 more record · DOI	Disagrees	1	6	0
An openly licensed benchmark and per-gene calibration map for missense pathogenicity predictors on activating cancer drivers and 1 more record · DOI	Agrees	3	1	2
Brandes et al. 2023 Supplementary Table 2: full benchmark evaluations of ESM1b and all other methods · DOI	Agrees	4	0	0

Comparisons in Livesey and Marsh 2025 (7)

Protocol	Dataset	Metric	VEST4	ESM-1b	Higher
Livesey and Marsh 2025 clinical benchmark: pairwise AUROC win rate, ClinVar pathogenic versus gnomAD missense	Livesey and Marsh 2025 clinical set: ClinVar pathogenic versus gnomAD v4.1 missense (819 proteins)	Pairwise win rate (average pairwise win rate against all other VEPs, per-protein AUROC on shared variants, ClinVar P/LP versus gnomAD v4.1 missense, proteins with at least 10 of each), percent	77.1	54.3	VEST4
Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, all 36 DMS sets	Livesey and Marsh 2025 human DMS benchmark set (36 proteins)	Pairwise win rate (average pairwise win rate against all other VEPs, \|Spearman\| with DMS on shared variants, 36 human DMS sets), percent	59	65	ESM-1b
Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, all 36 DMS sets	Livesey and Marsh 2025 human DMS benchmark set (36 proteins)	Pairwise win rate (average pairwise win rate against the five top-ranked VEPs (column top_5_winrate; the five are not listed in the table), \|Spearman\| with DMS on shared variants, 36 human DMS sets), percent	2.33	14.7	ESM-1b
Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, direct assays only	Livesey and Marsh 2025 human DMS benchmark set (36 proteins)	Pairwise win rate (average pairwise win rate against all other VEPs, \|Spearman\| with DMS on shared variants, direct DMS assays only), percent	59.9	62.4	ESM-1b
Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, indirect assays only	Livesey and Marsh 2025 human DMS benchmark set (36 proteins)	Pairwise win rate (average pairwise win rate against all other VEPs, \|Spearman\| with DMS on shared variants, indirect DMS assays only), percent	58.4	67	ESM-1b
Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, SNV-accessible substitutions only	Livesey and Marsh 2025 human DMS benchmark set (36 proteins)	Pairwise win rate (average pairwise win rate against all other VEPs, \|Spearman\| with DMS on shared variants, SNV-accessible substitutions only), percent	59	64.5	ESM-1b
Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, SNV-accessible substitutions with gaps filled	Livesey and Marsh 2025 human DMS benchmark set (36 proteins)	Pairwise win rate (average pairwise win rate against all other VEPs, \|Spearman\| with DMS, SNV-accessible substitutions, missing predictions filled with the VEP's most benign score), percent	60.5	65.9	ESM-1b

Comparisons in An openly licensed benchmark… (6)

Protocol	Dataset	Metric	VEST4	ESM-1b	Higher
CGC missense oncogenicity, all genes (Lee 2026, OncoCal tool_performance.tsv)	Missense variants in 768 Cancer Gene Census genes with open oncogenicity labels	AUROC (CGC missense, all genes), unitless	0.833	0.842	ESM-1b
CGC missense oncogenicity, all genes (Lee 2026, OncoCal tool_performance.tsv)	Missense variants in 768 Cancer Gene Census genes with open oncogenicity labels	Average precision (CGC missense, all genes), unitless	0.648	0.641	VEST4
CGC missense oncogenicity, oncogenes (gain of function) (Lee 2026, OncoCal tool_performance.tsv)	Missense variants in 768 Cancer Gene Census genes with open oncogenicity labels	AUROC (CGC missense, oncogenes (gain of function)), unitless	0.84	0.841	Tie
CGC missense oncogenicity, oncogenes (gain of function) (Lee 2026, OncoCal tool_performance.tsv)	Missense variants in 768 Cancer Gene Census genes with open oncogenicity labels	Average precision (CGC missense, oncogenes (gain of function)), unitless	0.713	0.719	Tie
CGC missense oncogenicity, tumour suppressors (loss of function) (Lee 2026, OncoCal tool_performance.tsv)	Missense variants in 768 Cancer Gene Census genes with open oncogenicity labels	AUROC (CGC missense, tumour suppressors (loss of function)), unitless	0.885	0.858	VEST4
CGC missense oncogenicity, tumour suppressors (loss of function) (Lee 2026, OncoCal tool_performance.tsv)	Missense variants in 768 Cancer Gene Census genes with open oncogenicity labels	Average precision (CGC missense, tumour suppressors (loss of function)), unitless	0.626	0.594	VEST4

Comparisons in Brandes et al. 2023 (4)

Protocol	Dataset	Metric	VEST4	ESM-1b	Higher
ClinVar missense pathogenicity, ESM1b head to head with one other method on the variants both score (Brandes et al. 2023 Fig. 2g,h, Supplementary Table 2)	ClinVar missense variants scored by both ESM1b and VEST4 (Brandes et al. 2023)	AUPRC (balanced PRC-AUC over the variants both methods score: mean over 100 random downsamples to equal pathogenic and benign counts (80% of the minority class); zero-shot), unitless	0.942	0.906	VEST4
ClinVar missense pathogenicity, ESM1b head to head with one other method on the variants both score (Brandes et al. 2023 Fig. 2g,h, Supplementary Table 2)	ClinVar missense variants scored by both ESM1b and VEST4 (Brandes et al. 2023)	AUROC (global ROC-AUC over the variants both methods score; zero-shot), unitless	0.946	0.906	VEST4
HGMD/gnomAD missense pathogenicity, ESM1b head to head with one other method on the variants both score (Brandes et al. 2023 Fig. 2g,h, Supplementary Table 2)	HGMD/gnomAD missense variants scored by both ESM1b and VEST4 (Brandes et al. 2023)	AUPRC (balanced PRC-AUC over the variants both methods score: mean over 100 random downsamples to equal pathogenic and benign counts (80% of the minority class); zero-shot), unitless	0.981	0.905	VEST4
HGMD/gnomAD missense pathogenicity, ESM1b head to head with one other method on the variants both score (Brandes et al. 2023 Fig. 2g,h, Supplementary Table 2)	HGMD/gnomAD missense variants scored by both ESM1b and VEST4 (Brandes et al. 2023)	AUROC (global ROC-AUC over the variants both methods score; zero-shot), unitless	0.98	0.894	VEST4

What disagrees

Other studies: rank VEST4 above ESM-1b in 7 of 8 comparisons
This study: ranks ESM-1b above VEST4 in 6 of 7 comparisons

Compared: VEST4, ESM-1b
Study that disagrees: Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML), Livesey and Marsh 2025, Supplementary Material 2 (Tables S1-S8)
Other studies: Brandes et al. 2023 Supplementary Table 2: full benchmark evaluations of ESM1b and all other methods, An openly licensed benchmark and per-gene calibration map for missense pathogenicity predictors on activating cancer drivers, OncoCal repository, tables/tool_performance.tsv (per-tool AUROC and AUPRC by gene role)
Rule: Ordering reversal; different tasks across studies
Model identity: Matched across studies partly through links not yet reviewed

The compared numbers

Each value links to its result record and source. Compare values within one protocol; the rule compares rankings, not raw values across protocols.

In the study that disagrees (14)
Study	Tool	Protocol or benchmark	Metric	Value
Livesey and Marsh 2025, Supplementary Material 2 (Tables S1-S8); Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML)	ESM-1b (Livesey and Marsh 2025)	Livesey and Marsh 2025 clinical benchmark: pairwise AUROC win rate, ClinVar pathogenic versus gnomAD missense	Pairwise win rate (average pairwise win rate against all other VEPs, per-protein AUROC on shared variants, ClinVar P/LP versus gnomAD v4.1 missense, proteins with at least 10 of each)	54.3
Livesey and Marsh 2025, Supplementary Material 2 (Tables S1-S8); Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML)	VEST4 (Livesey and Marsh 2025)	Livesey and Marsh 2025 clinical benchmark: pairwise AUROC win rate, ClinVar pathogenic versus gnomAD missense	Pairwise win rate (average pairwise win rate against all other VEPs, per-protein AUROC on shared variants, ClinVar P/LP versus gnomAD v4.1 missense, proteins with at least 10 of each)	77.1
Livesey and Marsh 2025, Supplementary Material 2 (Tables S1-S8); Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML)	ESM-1b (Livesey and Marsh 2025)	Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, all 36 DMS sets	Pairwise win rate (average pairwise win rate against all other VEPs, \|Spearman\| with DMS on shared variants, 36 human DMS sets)	65
Livesey and Marsh 2025, Supplementary Material 2 (Tables S1-S8); Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML)	ESM-1b (Livesey and Marsh 2025)	Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, all 36 DMS sets	Pairwise win rate (average pairwise win rate against the five top-ranked VEPs (column top_5_winrate; the five are not listed in the table), \|Spearman\| with DMS on shared variants, 36 human DMS sets)	14.7
Livesey and Marsh 2025, Supplementary Material 2 (Tables S1-S8); Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML)	VEST4 (Livesey and Marsh 2025)	Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, all 36 DMS sets	Pairwise win rate (average pairwise win rate against all other VEPs, \|Spearman\| with DMS on shared variants, 36 human DMS sets)	59
Livesey and Marsh 2025, Supplementary Material 2 (Tables S1-S8); Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML)	VEST4 (Livesey and Marsh 2025)	Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, all 36 DMS sets	Pairwise win rate (average pairwise win rate against the five top-ranked VEPs (column top_5_winrate; the five are not listed in the table), \|Spearman\| with DMS on shared variants, 36 human DMS sets)	2.33
Livesey and Marsh 2025, Supplementary Material 2 (Tables S1-S8); Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML)	ESM-1b (Livesey and Marsh 2025)	Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, direct assays only	Pairwise win rate (average pairwise win rate against all other VEPs, \|Spearman\| with DMS on shared variants, direct DMS assays only)	62.4
Livesey and Marsh 2025, Supplementary Material 2 (Tables S1-S8); Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML)	VEST4 (Livesey and Marsh 2025)	Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, direct assays only	Pairwise win rate (average pairwise win rate against all other VEPs, \|Spearman\| with DMS on shared variants, direct DMS assays only)	59.9
Livesey and Marsh 2025, Supplementary Material 2 (Tables S1-S8); Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML)	ESM-1b (Livesey and Marsh 2025)	Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, indirect assays only	Pairwise win rate (average pairwise win rate against all other VEPs, \|Spearman\| with DMS on shared variants, indirect DMS assays only)	67
Livesey and Marsh 2025, Supplementary Material 2 (Tables S1-S8); Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML)	VEST4 (Livesey and Marsh 2025)	Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, indirect assays only	Pairwise win rate (average pairwise win rate against all other VEPs, \|Spearman\| with DMS on shared variants, indirect DMS assays only)	58.4
Livesey and Marsh 2025, Supplementary Material 2 (Tables S1-S8); Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML)	ESM-1b (Livesey and Marsh 2025)	Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, SNV-accessible substitutions only	Pairwise win rate (average pairwise win rate against all other VEPs, \|Spearman\| with DMS on shared variants, SNV-accessible substitutions only)	64.5
Livesey and Marsh 2025, Supplementary Material 2 (Tables S1-S8); Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML)	ESM-1b (Livesey and Marsh 2025)	Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, SNV-accessible substitutions with gaps filled	Pairwise win rate (average pairwise win rate against all other VEPs, \|Spearman\| with DMS, SNV-accessible substitutions, missing predictions filled with the VEP's most benign score)	65.9
Livesey and Marsh 2025, Supplementary Material 2 (Tables S1-S8); Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML)	VEST4 (Livesey and Marsh 2025)	Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, SNV-accessible substitutions with gaps filled	Pairwise win rate (average pairwise win rate against all other VEPs, \|Spearman\| with DMS, SNV-accessible substitutions, missing predictions filled with the VEP's most benign score)	60.5
Livesey and Marsh 2025, Supplementary Material 2 (Tables S1-S8); Livesey and Marsh 2025, Genome Biology 26:104 (Europe PMC full text XML)	VEST4 (Livesey and Marsh 2025)	Livesey and Marsh 2025 DMS benchmark: pairwise \|Spearman\| win rate, SNV-accessible substitutions only	Pairwise win rate (average pairwise win rate against all other VEPs, \|Spearman\| with DMS on shared variants, SNV-accessible substitutions only)	59

In the other studies (16)
Study	Tool	Protocol or benchmark	Metric	Value
Brandes et al. 2023 Supplementary Table 2: full benchmark evaluations of ESM1b and all other methods; Genome-wide prediction of disease variant effects with a deep protein language model (Brandes et al., Nature Genetics 2023)	ESM1b (Brandes et al. 2023 clinical benchmarks)	ClinVar missense pathogenicity, ESM1b head to head with one other method on the variants both score (Brandes et al. 2023 Fig. 2g,h, Supplementary Table 2)	AUPRC (balanced PRC-AUC over the variants both methods score: mean over 100 random downsamples to equal pathogenic and benign counts (80% of the minority class); zero-shot)	0.906
Brandes et al. 2023 Supplementary Table 2: full benchmark evaluations of ESM1b and all other methods; Genome-wide prediction of disease variant effects with a deep protein language model (Brandes et al., Nature Genetics 2023)	ESM1b (Brandes et al. 2023 clinical benchmarks)	ClinVar missense pathogenicity, ESM1b head to head with one other method on the variants both score (Brandes et al. 2023 Fig. 2g,h, Supplementary Table 2)	AUROC (global ROC-AUC over the variants both methods score; zero-shot)	0.906
Brandes et al. 2023 Supplementary Table 2: full benchmark evaluations of ESM1b and all other methods; Genome-wide prediction of disease variant effects with a deep protein language model (Brandes et al., Nature Genetics 2023)	VEST4 (Brandes et al. 2023 head-to-head comparison)	ClinVar missense pathogenicity, ESM1b head to head with one other method on the variants both score (Brandes et al. 2023 Fig. 2g,h, Supplementary Table 2)	AUPRC (balanced PRC-AUC over the variants both methods score: mean over 100 random downsamples to equal pathogenic and benign counts (80% of the minority class); zero-shot)	0.942
Brandes et al. 2023 Supplementary Table 2: full benchmark evaluations of ESM1b and all other methods; Genome-wide prediction of disease variant effects with a deep protein language model (Brandes et al., Nature Genetics 2023)	VEST4 (Brandes et al. 2023 head-to-head comparison)	ClinVar missense pathogenicity, ESM1b head to head with one other method on the variants both score (Brandes et al. 2023 Fig. 2g,h, Supplementary Table 2)	AUROC (global ROC-AUC over the variants both methods score; zero-shot)	0.946
Brandes et al. 2023 Supplementary Table 2: full benchmark evaluations of ESM1b and all other methods; Genome-wide prediction of disease variant effects with a deep protein language model (Brandes et al., Nature Genetics 2023)	ESM1b (Brandes et al. 2023 clinical benchmarks)	HGMD/gnomAD missense pathogenicity, ESM1b head to head with one other method on the variants both score (Brandes et al. 2023 Fig. 2g,h, Supplementary Table 2)	AUPRC (balanced PRC-AUC over the variants both methods score: mean over 100 random downsamples to equal pathogenic and benign counts (80% of the minority class); zero-shot)	0.905
Brandes et al. 2023 Supplementary Table 2: full benchmark evaluations of ESM1b and all other methods; Genome-wide prediction of disease variant effects with a deep protein language model (Brandes et al., Nature Genetics 2023)	ESM1b (Brandes et al. 2023 clinical benchmarks)	HGMD/gnomAD missense pathogenicity, ESM1b head to head with one other method on the variants both score (Brandes et al. 2023 Fig. 2g,h, Supplementary Table 2)	AUROC (global ROC-AUC over the variants both methods score; zero-shot)	0.894
Brandes et al. 2023 Supplementary Table 2: full benchmark evaluations of ESM1b and all other methods; Genome-wide prediction of disease variant effects with a deep protein language model (Brandes et al., Nature Genetics 2023)	VEST4 (Brandes et al. 2023 head-to-head comparison)	HGMD/gnomAD missense pathogenicity, ESM1b head to head with one other method on the variants both score (Brandes et al. 2023 Fig. 2g,h, Supplementary Table 2)	AUPRC (balanced PRC-AUC over the variants both methods score: mean over 100 random downsamples to equal pathogenic and benign counts (80% of the minority class); zero-shot)	0.981
Brandes et al. 2023 Supplementary Table 2: full benchmark evaluations of ESM1b and all other methods; Genome-wide prediction of disease variant effects with a deep protein language model (Brandes et al., Nature Genetics 2023)	VEST4 (Brandes et al. 2023 head-to-head comparison)	HGMD/gnomAD missense pathogenicity, ESM1b head to head with one other method on the variants both score (Brandes et al. 2023 Fig. 2g,h, Supplementary Table 2)	AUROC (global ROC-AUC over the variants both methods score; zero-shot)	0.98
An openly licensed benchmark and per-gene calibration map for missense pathogenicity predictors on activating cancer drivers; OncoCal repository, tables/tool_performance.tsv (per-tool AUROC and AUPRC by gene role)	ESM1b rank score, dbNSFP 5.3.1a (Lee 2026)	CGC missense oncogenicity, all genes (Lee 2026, OncoCal tool_performance.tsv)	Average precision (CGC missense, all genes)	0.641
An openly licensed benchmark and per-gene calibration map for missense pathogenicity predictors on activating cancer drivers; OncoCal repository, tables/tool_performance.tsv (per-tool AUROC and AUPRC by gene role)	ESM1b rank score, dbNSFP 5.3.1a (Lee 2026)	CGC missense oncogenicity, all genes (Lee 2026, OncoCal tool_performance.tsv)	AUROC (CGC missense, all genes)	0.842
An openly licensed benchmark and per-gene calibration map for missense pathogenicity predictors on activating cancer drivers; OncoCal repository, tables/tool_performance.tsv (per-tool AUROC and AUPRC by gene role)	ESM1b rank score, dbNSFP 5.3.1a (Lee 2026)	CGC missense oncogenicity, tumour suppressors (loss of function) (Lee 2026, OncoCal tool_performance.tsv)	Average precision (CGC missense, tumour suppressors (loss of function))	0.594
An openly licensed benchmark and per-gene calibration map for missense pathogenicity predictors on activating cancer drivers; OncoCal repository, tables/tool_performance.tsv (per-tool AUROC and AUPRC by gene role)	ESM1b rank score, dbNSFP 5.3.1a (Lee 2026)	CGC missense oncogenicity, tumour suppressors (loss of function) (Lee 2026, OncoCal tool_performance.tsv)	AUROC (CGC missense, tumour suppressors (loss of function))	0.858
An openly licensed benchmark and per-gene calibration map for missense pathogenicity predictors on activating cancer drivers; OncoCal repository, tables/tool_performance.tsv (per-tool AUROC and AUPRC by gene role)	VEST4 rank score, dbNSFP 5.3.1a (Lee 2026)	CGC missense oncogenicity, all genes (Lee 2026, OncoCal tool_performance.tsv)	Average precision (CGC missense, all genes)	0.648
An openly licensed benchmark and per-gene calibration map for missense pathogenicity predictors on activating cancer drivers; OncoCal repository, tables/tool_performance.tsv (per-tool AUROC and AUPRC by gene role)	VEST4 rank score, dbNSFP 5.3.1a (Lee 2026)	CGC missense oncogenicity, all genes (Lee 2026, OncoCal tool_performance.tsv)	AUROC (CGC missense, all genes)	0.833
An openly licensed benchmark and per-gene calibration map for missense pathogenicity predictors on activating cancer drivers; OncoCal repository, tables/tool_performance.tsv (per-tool AUROC and AUPRC by gene role)	VEST4 rank score, dbNSFP 5.3.1a (Lee 2026)	CGC missense oncogenicity, tumour suppressors (loss of function) (Lee 2026, OncoCal tool_performance.tsv)	Average precision (CGC missense, tumour suppressors (loss of function))	0.626
An openly licensed benchmark and per-gene calibration map for missense pathogenicity predictors on activating cancer drivers; OncoCal repository, tables/tool_performance.tsv (per-tool AUROC and AUPRC by gene role)	VEST4 rank score, dbNSFP 5.3.1a (Lee 2026)	CGC missense oncogenicity, tumour suppressors (loss of function) (Lee 2026, OncoCal tool_performance.tsv)	AUROC (CGC missense, tumour suppressors (loss of function))	0.885

Follow-up

Current state: Not yet reviewed.

No one has reviewed this finding yet.

How to help resolve this

If you know why these studies disagree, for example a different population, metric or preprocessing step, or you have a corrected value, tell us through the contribution form. A reviewer checks every submission against its source before anything changes, and the step is added to the follow-up above.

Finding discrepancy-ad72f05d02a61111 · All contested results · How findings are made and followed up