What was checked

The ten checks

Six probe checks fit one sequence feature against the class labels and report how well it separates them. Three integrity checks count duplicates and unique bases. Together those nine run per split. The last, the leakage scan, runs across pairs of splits.

This page is what they returned on this corpus, and how to read the two scales they return it on. What each check computes, in full, is documented by the tool rather than restated here.

What a flag is

Each check returns one of four flags on one split: Pass, Warning, Fail, Unknown. The first three are scored outcomes and are ordered by severity. Unknown is not the fourth degree of that order: it means the check ran and could not be scored. A blank — a check with no flag at all — is a third thing again: the check had nothing to run on.

Keeping those apart matters more than it looks. Reading Unknown or a blank as Pass converts an absence of measurement into a reassuring result, which is the single easiest way to make an audit say more than it found. Every figure on this site draws Unknown as a hatch rather than a fourth colour, and draws nothing at all for a blank.

Pass is the narrowest of the four. It says this check did not separate the classes on this split. It does not say the dataset is free of structure a model could use, because a check that did not find something is not evidence there is nothing to find. No combination of the ten flags licenses that conclusion either.

The boundaries a probe flag is drawn at · How the tool decides one, including the simulations the 0.60 boundary comes from

The split these numbers are on

A dataset ships several splits and a check runs on each of them. The single flag, AUROC and overlap this site quotes per dataset are the values on that dataset’s reference split — train for every dataset but one, where the training split is not released and test stands in.

They are not maxima over splits. A maximum would make a dataset with more splits look worse for having been measured more times, and would name no split a reader could go and check. Every figure here names the split its numbers were measured on, and every dataset page shows all splits alongside the reference one.

Across the corpus

What each check returned

The share of the 93 datasets carrying each flag on each check, measured on each dataset’s reference split. The trailing figure counts the datasets flagged Warning or Fail.

Table view

Datasets per flag per check.
Check	Pass	Warning	Fail	No result	Flagged
Split leakage	17	52	23	1	80.6%
Per sequence dinucleotide content	37	30	26		60.2%
Sequence Duplications within Labels	52	36	5		44.1%
Per sequence nucleotide content	55	21	17		40.9%
Per sequence GC content	58	13	22		37.6%
Per position nucleotide content	66	17	10		29.0%
Per position reversed nucleotide content	68	15	10		26.9%
Duplicate Sequences between Labels	77	0	16		17.2%
Unique bases	86	0	7		7.5%
Sequence lengths	89	4	0		4.3%

The same counts as a table, with the two absences kept apart: Unknown is a check that ran and could not be scored, blank is a check with nothing to run on. **Every count is a link** to those datasets in the catalogue.
Check	Kind	Pass	Warning	Fail	Unknown	Blank	Scored
Per sequence GC content	probe	58	13	22	—	—	93
Per sequence nucleotide content	probe	55	21	17	—	—	93
Per sequence dinucleotide content	probe	37	30	26	—	—	93
Sequence lengths	probe	89	4	—	—	—	93
Per position nucleotide content	probe	66	17	10	—	—	93
Per position reversed nucleotide content	probe	68	15	10	—	—	93
Sequence Duplications within Labels	integrity	52	36	5	—	—	93
Duplicate Sequences between Labels	integrity	77	—	16	—	—	93
Unique bases	integrity	86	—	7	—	—	93
Split leakage	leakage scan	17	52	23	—	1	92

The probe scale

Reading a probe AUROC

A probe is one feature and one fit — sequence length, GC content, a nucleotide or dinucleotide frequency, a per-position profile. The AUROC it reaches is a lower bound on what is recoverable from the sequences: a stronger feature, or a model, may recover more. It is never an upper bound, and a low value is not evidence that a model would also score low.

Across the 93 probed datasets the median strongest probe reaches 0.67, and 34 datasets reach 0.70 or more from a single feature.

The figure below is the strongest probe per dataset, one mark each. Read a mark as a lower bound on what a model could recover from the sequences alone, and the spread as the range this corpus covers rather than as a ranking.

The strongest single-feature probe AUROC on each dataset’s reference split, over the 93 datasets where a probe was measured. Every value is a lower bound: it is what one simple feature recovered, not the most that could be.

The dashed line marks 0.7, the threshold the Core Set uses; it is a selection cut, not a claim about the datasets either side of it.

Table view

Strongest single-feature probe, binned.
Bin	Datasets	Which
0.50–0.52	3	GUE_v2:transcription_factor_prediction_mouse_0, GUE_v2:transcription_factor_prediction_mouse_4, miRBench:AGO2_CLASH_Hejret2023
0.52–0.54	5	GUE_v2:transcription_factor_prediction_mouse_1, GUE_v2:transcription_factor_prediction_mouse_2, GUE_v2:transcription_factor_prediction_mouse_3, miRBench:AGO2_eCLIP_Klimentova2022, miRBench:AGO2_eCLIP_Manakov2022
0.54–0.56	5	GUE_v2:enhancer_promoter_interaction_IMR90, GUE_v2:enhancer_promoter_interaction_NHEK, GUE_v2:transcription_factor_prediction_human_3, mRNABench:mrl-sugimoto, nucleotide_benchmarks:H3K9me3
0.56–0.58	5	GUE_v2:enhancer_promoter_interaction_K562, OmniGenBench:lncrna_s_lycopersicum, OmniGenBench:lncrna_t_aestivum, genomic-benchmarks:drosophila_enhancers_stark, genomic-benchmarks:human_ocr_ensembl
0.58–0.60	6	GUE_v2:enhancer_promoter_interaction_GM12878, GUE_v2:enhancer_promoter_interaction_HeLa_S3, GUE_v2:epigenetic_marks_prediction_H3K14ac, GUE_v2:epigenetic_marks_prediction_H3K4me3, OmniGenBench:lncrna_g_max, OmniGenBench:promoter_strength_protoplast
0.60–0.62	5	BEACON:CRISPROnTarget, GUE_v2:enhancer_promoter_interaction_HUVEC, GUE_v2:epigenetic_marks_prediction_H3K4me1, OmniGenBench:lncrna_m_esculenta, mRNABench:mrl-sample-varying
0.62–0.64	11	BEACON:MeanRibosomeLoading, GUE_v2:epigenetic_marks_prediction_H4ac, OmniGenBench:lncrna_s_bicolor, OmniGenBench:lncrna_z_mays, OmniGenBench:pro_seq_m_esculenta, OmniGenBench:te_prediction_rice, OmniGenBench:terminator_strength_leaf, OmniGenBench:terminator_strength_protoplast, mRNABench:mrl-sample-mcherry, mRNABench:rnahl-human, mRNABench:rnahl-mouse
0.64–0.66	4	GUE_v2:epigenetic_marks_prediction_H3K36me3, GUE_v2:transcription_factor_prediction_human_2, mRNABench:translation-efficiency-human, mRNABench:translation-efficiency-mouse
0.66–0.68	5	GUE_v2:epigenetic_marks_prediction_H3K4me2, OmniGenBench:te_prediction_arabidopsis, mRNABench:mrl-sample-egfp-m1pseudo, mRNABench:mrl-sample-egfp-pseudo, mRNABench:mrl-sample-egfp-unmod
0.68–0.70	10	GUE_v2:epigenetic_marks_prediction_H3K79me3, GUE_v2:epigenetic_marks_prediction_H3K9ac, GUE_v2:transcription_factor_prediction_human_0, GUE_v2:transcription_factor_prediction_human_1, GUE_v2:transcription_factor_prediction_human_4, OmniGenBench:poly_a_chlamydomonas_reinhardtii, genomic-benchmarks:human_enhancers_ensembl, nucleotide_benchmarks:enhancers, nucleotide_benchmarks:splice_sites_acceptors, nucleotide_benchmarks:splice_sites_donors
0.70–0.72	3	GUE_v2:core_promoter_detection_tata, GUE_v2:promoter_detection_300_tata, nucleotide_benchmarks:H3K4me1
0.72–0.74	6	OmniGenBench:poly_a_arabidopsis_thaliana, OmniGenBench:poly_a_medicago_truncatula, OmniGenBench:splicing_arabidopsis_thaliana_acceptor, mRNABench:mrl-sample-designed, nucleotide_benchmarks:H2AFZ, nucleotide_benchmarks:H3K27ac
0.74–0.76	6	BEACON:Isoform, GUE_v2:epigenetic_marks_prediction_H4, OmniGenBench:poly_a_trifolium_pratense, OmniGenBench:promoter_strength_leaf, nucleotide_benchmarks:H3K36me3, nucleotide_benchmarks:H3K4me2
0.76–0.78	4	GUE_v2:core_promoter_detection_all, OmniGenBench:splicing_arabidopsis_thaliana_donor, genomic-benchmarks:human_enhancers_cohn, nucleotide_benchmarks:H3K9ac
0.78–0.80	1	nucleotide_benchmarks:H3K27me3
0.80–0.82	2	GUE_v2:core_promoter_detection_notata, OmniGenBench:poly_a_oryza_sativa_japonica_group
0.82–0.84	5	GUE_v2:epigenetic_marks_prediction_H3, GUE_v2:promoter_detection_300_all, OmniGenBench:poly_a_oryza_sativa_indica_group, nucleotide_benchmarks:H4K20me1, nucleotide_benchmarks:promoter_tata
0.84–0.86	3	GUE_v2:promoter_detection_300_notata, genomic-benchmarks:demo_coding_vs_intergenomic_seqs, genomic-benchmarks:human_nontata_promoters
0.86–0.88	2	genomic-benchmarks:demo_human_or_worm, nucleotide_benchmarks:H3K4me3
0.88–0.90	1	nucleotide_benchmarks:promoter_all
0.90–0.92	1	nucleotide_benchmarks:promoter_no_tata

What the same features reach read together rather than one at a time

The overlap scale

Reading a split overlap

The leakage scan compares two splits and reports, in each direction, the percentage of one split’s sequences that were also present in the other. Both directions are shown because the two differ whenever the splits differ in size, and quoting only the larger would describe neither split.

92 of 93 datasets were scanned; One ships a single split and so has no pair to compare. 75 have at least one sequence in two splits, 6 have more than half of one split present in another, and 5 have all of it.

The bands below count datasets rather than implying a density, because the structure worth seeing is all at one end: most of the corpus sits in the narrowest band and a handful sit at the top of the range.

For each of the 92 datasets the leakage scan could compare, the largest share of a held-out split whose sequences also appear in the training split. Measured on exact sequence identity.

The bands are uneven, so the bar lengths count datasets and are not a density. A dataset in “none detected” had no exact sequence in common between the splits the scan compared.

Table view

Datasets per band of held-out overlap.
Overlap	Datasets	Which
none detected	17	BEACON:CRISPROnTarget, BEACON:MeanRibosomeLoading, GUE_v2:epigenetic_marks_prediction_H3, GUE_v2:epigenetic_marks_prediction_H3K4me3, GUE_v2:transcription_factor_prediction_mouse_0, GUE_v2:transcription_factor_prediction_mouse_2, GUE_v2:transcription_factor_prediction_mouse_4, OmniGenBench:poly_a_arabidopsis_thaliana, OmniGenBench:poly_a_chlamydomonas_reinhardtii, OmniGenBench:poly_a_medicago_truncatula, OmniGenBench:poly_a_trifolium_pratense, OmniGenBench:pro_seq_m_esculenta, OmniGenBench:splicing_arabidopsis_thaliana_acceptor, OmniGenBench:splicing_arabidopsis_thaliana_donor, mRNABench:translation-efficiency-human, mRNABench:translation-efficiency-mouse, nucleotide_benchmarks:promoter_tata
up to 0.5%	19	GUE_v2:epigenetic_marks_prediction_H3K14ac, GUE_v2:epigenetic_marks_prediction_H3K36me3, GUE_v2:epigenetic_marks_prediction_H3K4me1, GUE_v2:epigenetic_marks_prediction_H3K4me2, GUE_v2:epigenetic_marks_prediction_H3K79me3, GUE_v2:epigenetic_marks_prediction_H3K9ac, GUE_v2:epigenetic_marks_prediction_H4, GUE_v2:epigenetic_marks_prediction_H4ac, GUE_v2:transcription_factor_prediction_mouse_1, GUE_v2:transcription_factor_prediction_mouse_3, OmniGenBench:lncrna_s_bicolor, OmniGenBench:poly_a_oryza_sativa_indica_group, OmniGenBench:poly_a_oryza_sativa_japonica_group, OmniGenBench:te_prediction_arabidopsis, OmniGenBench:terminator_strength_protoplast, mRNABench:rnahl-human, mRNABench:rnahl-mouse, miRBench:AGO2_CLASH_Hejret2023, nucleotide_benchmarks:H3K4me3
0.5% to 1%	19	OmniGenBench:lncrna_m_esculenta, OmniGenBench:lncrna_s_lycopersicum, OmniGenBench:promoter_strength_leaf, OmniGenBench:promoter_strength_protoplast, OmniGenBench:terminator_strength_leaf, genomic-benchmarks:drosophila_enhancers_stark, genomic-benchmarks:human_enhancers_cohn, nucleotide_benchmarks:H2AFZ, nucleotide_benchmarks:H3K27ac, nucleotide_benchmarks:H3K27me3, nucleotide_benchmarks:H3K36me3, nucleotide_benchmarks:H3K4me1, nucleotide_benchmarks:H3K4me2, nucleotide_benchmarks:H3K9ac, nucleotide_benchmarks:H4K20me1, nucleotide_benchmarks:promoter_all, nucleotide_benchmarks:promoter_no_tata, nucleotide_benchmarks:splice_sites_acceptors, nucleotide_benchmarks:splice_sites_donors
1% to 2%	14	GUE_v2:core_promoter_detection_all, GUE_v2:core_promoter_detection_notata, GUE_v2:enhancer_promoter_interaction_GM12878, GUE_v2:enhancer_promoter_interaction_HUVEC, GUE_v2:enhancer_promoter_interaction_HeLa_S3, GUE_v2:enhancer_promoter_interaction_K562, GUE_v2:promoter_detection_300_all, GUE_v2:promoter_detection_300_notata, OmniGenBench:te_prediction_rice, genomic-benchmarks:demo_human_or_worm, genomic-benchmarks:human_ocr_ensembl, miRBench:AGO2_eCLIP_Manakov2022, nucleotide_benchmarks:H3K9me3, nucleotide_benchmarks:enhancers
2% to 5%	7	GUE_v2:enhancer_promoter_interaction_IMR90, GUE_v2:enhancer_promoter_interaction_NHEK, GUE_v2:transcription_factor_prediction_human_0, GUE_v2:transcription_factor_prediction_human_1, GUE_v2:transcription_factor_prediction_human_2, GUE_v2:transcription_factor_prediction_human_3, OmniGenBench:lncrna_g_max
5% to 10%	4	GUE_v2:core_promoter_detection_tata, GUE_v2:promoter_detection_300_tata, OmniGenBench:lncrna_z_mays, genomic-benchmarks:demo_coding_vs_intergenomic_seqs
10% to 25%	2	OmniGenBench:lncrna_t_aestivum, mRNABench:mrl-sugimoto
25% to 50%	4	BEACON:Isoform, GUE_v2:transcription_factor_prediction_human_4, genomic-benchmarks:human_enhancers_ensembl, genomic-benchmarks:human_nontata_promoters
over 50%	6	mRNABench:mrl-sample-designed, mRNABench:mrl-sample-egfp-m1pseudo, mRNABench:mrl-sample-egfp-pseudo, mRNABench:mrl-sample-egfp-unmod, mRNABench:mrl-sample-mcherry, mRNABench:mrl-sample-varying

How the scan matches sequences across a pair of splits

Every cell

Every dataset, every check

93 rows by 10 columns, grouped by collection and ordered worst-first inside each group. Each cell is the flag that check returned on that dataset's reference split, and every mark is a link.

One row per dataset, one column per check, grouped by collection. Each cell is the flag that check returned on that dataset's reference split.
Dataset	GC content	Nucleotide content	Dinucleotide content	Sequence lengths	Per position nucleotide	Per position from 3′	Duplicates within label	Duplicates across labels	Unique bases	Split leakage
	probe checks						integrity checks			leakage scan
BEACON3
Isoform	Warning	Fail	Fail	Pass	Fail	Fail	Pass	Pass	Pass	Fail
CRISPROnTarget	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass
MeanRibosomeLoading	Pass	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Pass
GUE_v232
core_promoter_detection_tata	Fail	Fail	Fail	Pass	Fail	Fail	Warning	Fail	Pass	Fail
promoter_detection_300_notata	Fail	Fail	Fail	Pass	Warning	Warning	Warning	Pass	Fail	Warning
promoter_detection_300_tata	Warning	Warning	Warning	Pass	Fail	Fail	Warning	Fail	Pass	Fail
core_promoter_detection_all	Fail	Warning	Fail	Pass	Warning	Warning	Warning	Fail	Pass	Warning
core_promoter_detection_notata	Fail	Warning	Fail	Pass	Warning	Warning	Warning	Fail	Pass	Warning
promoter_detection_300_all	Warning	Warning	Fail	Pass	Warning	Warning	Warning	Fail	Pass	Warning
epigenetic_marks_prediction_H4	Warning	Warning	Fail	Pass	Pass	Pass	Pass	Pass	Pass	Warning
epigenetic_marks_prediction_H3K4me2	Warning	Warning	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Warning
epigenetic_marks_prediction_H3K9ac	Warning	Warning	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Warning
epigenetic_marks_prediction_H3	Fail	Fail	Fail	Pass	Pass	Pass	Pass	Pass	Pass	Pass
enhancer_promoter_interaction_NHEK	Pass	Pass	Pass	Pass	Pass	Pass	Warning	Pass	Fail	Fail
transcription_factor_prediction_human_4	Pass	Pass	Warning	Pass	Pass	Pass	Fail	Pass	Pass	Fail
enhancer_promoter_interaction_HUVEC	Pass	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Fail	Warning
transcription_factor_prediction_human_0	Pass	Pass	Warning	Pass	Pass	Pass	Warning	Pass	Pass	Fail
transcription_factor_prediction_human_2	Pass	Pass	Warning	Pass	Pass	Pass	Warning	Pass	Pass	Fail
epigenetic_marks_prediction_H4ac	Warning	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Warning
enhancer_promoter_interaction_IMR90	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Fail	Fail
transcription_factor_prediction_human_1	Pass	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Fail
transcription_factor_prediction_human_3	Pass	Pass	Pass	Pass	Pass	Pass	Warning	Pass	Pass	Fail
epigenetic_marks_prediction_H3K36me3	Pass	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Warning
epigenetic_marks_prediction_H3K4me1	Pass	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Warning
epigenetic_marks_prediction_H3K79me3	Pass	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Warning
enhancer_promoter_interaction_GM12878	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Warning
enhancer_promoter_interaction_HeLa_S3	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Warning
enhancer_promoter_interaction_K562	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Warning
epigenetic_marks_prediction_H3K14ac	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Warning
transcription_factor_prediction_mouse_1	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Warning
transcription_factor_prediction_mouse_3	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Warning
epigenetic_marks_prediction_H3K4me3	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass
transcription_factor_prediction_mouse_0	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass
transcription_factor_prediction_mouse_2	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass
transcription_factor_prediction_mouse_4	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass
OmniGenBench21
te_prediction_arabidopsis	Warning	Warning	Warning	Pass	Pass	Pass	Warning	Fail	Pass	Warning
terminator_strength_leaf	Warning	Warning	Warning	Pass	Warning	Warning	Pass	Pass	Pass	Warning
lncrna_z_mays	Pass	Pass	Pass	Pass	Warning	Warning	Fail	Fail	Pass	Fail
lncrna_s_lycopersicum	Pass	Pass	Pass	Pass	Pass	Pass	Warning	Fail	Fail	Warning
promoter_strength_leaf	Fail	Warning	Fail	Pass	Pass	Pass	Pass	Pass	Pass	Warning
lncrna_m_esculenta	Pass	Pass	Pass	Pass	Warning	Pass	Warning	Fail	Pass	Warning
lncrna_s_bicolor	Pass	Pass	Pass	Pass	Warning	Pass	Warning	Fail	Pass	Warning
poly_a_medicago_truncatula	Pass	Pass	Pass	Pass	Fail	Fail	Pass	Fail	Pass	Pass
lncrna_g_max	Pass	Pass	Pass	Pass	Pass	Pass	Warning	Fail	Pass	Fail
lncrna_t_aestivum	Pass	Pass	Pass	Pass	Pass	Pass	Warning	Fail	Pass	Fail
poly_a_arabidopsis_thaliana	Pass	Pass	Pass	Pass	Fail	Fail	Warning	Pass	Pass	Pass
poly_a_oryza_sativa_indica_group	Pass	Pass	Pass	Pass	Fail	Fail	Pass	Pass	Pass	Warning
poly_a_oryza_sativa_japonica_group	Pass	Pass	Pass	Pass	Fail	Fail	Pass	Pass	Pass	Warning
poly_a_trifolium_pratense	Pass	Pass	Pass	Pass	Fail	Fail	Warning	Pass	Pass	Pass
splicing_arabidopsis_thaliana_donor	Pass	Pass	Warning	Pass	Fail	Fail	Pass	Pass	Pass	Pass
poly_a_chlamydomonas_reinhardtii	Pass	Pass	Pass	Pass	Warning	Warning	Pass	Fail	Pass	Pass
te_prediction_rice	Pass	Warning	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Warning
splicing_arabidopsis_thaliana_acceptor	Pass	Pass	Pass	Pass	Fail	Fail	Pass	Pass	Pass	Pass
terminator_strength_protoplast	Pass	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Warning
pro_seq_m_esculenta	Pass	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Pass
promoter_strength_protoplast	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Warning
genomic-benchmarks7
human_enhancers_ensembl	Warning	Warning	Warning	Pass	Warning	Warning	Fail	Fail	Fail	Fail
human_nontata_promoters	Fail	Fail	Fail	Pass	Warning	Warning	Pass	Pass	Pass	Fail
demo_human_or_worm	Warning	Warning	Fail	Pass	Pass	Pass	Fail	Pass	Fail	Warning
demo_coding_vs_intergenomic_seqs	Fail	Fail	Fail	Pass	Pass	Pass	Warning	Pass	Pass	Fail
human_enhancers_cohn	Fail	Fail	Fail	Pass	Pass	Pass	Pass	Pass	Pass	Warning
human_ocr_ensembl	Pass	Pass	Pass	Pass	Pass	Pass	Warning	Pass	Pass	Warning
drosophila_enhancers_stark	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Warning
mRNABench11
mrl-sample-designed	Fail	Fail	Fail	Pass	Pass	Pass	Pass	Pass	Pass	Fail
mrl-sugimoto	Pass	Pass	Pass	Pass	Pass	Pass	Fail	Fail	Pass	Fail
translation-efficiency-human	Pass	Warning	Warning	Warning	Pass	Pass	Pass	Pass	Pass	Pass
translation-efficiency-mouse	Pass	Warning	Warning	Warning	Pass	Pass	Pass	Pass	Pass	Pass
mrl-sample-egfp-m1pseudo	Pass	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Fail
mrl-sample-egfp-pseudo	Pass	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Fail
mrl-sample-egfp-unmod	Pass	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Fail
mrl-sample-mcherry	Pass	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Fail
mrl-sample-varying	Pass	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Pass	Fail
rnahl-human	Pass	Pass	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Warning
rnahl-mouse	Pass	Pass	Pass	Warning	Pass	Pass	Pass	Pass	Pass	Warning
miRBench3
AGO2_CLASH_Hejret2023	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Warning
AGO2_eCLIP_Manakov2022	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Warning
AGO2_eCLIP_Klimentova2022	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	Pass	no result recorded
nucleotide_benchmarks16
promoter_all	Fail	Fail	Fail	Pass	Warning	Warning	Warning	Pass	Pass	Warning
promoter_no_tata	Fail	Fail	Fail	Pass	Warning	Warning	Warning	Pass	Pass	Warning
enhancers	Warning	Warning	Warning	Pass	Warning	Warning	Warning	Pass	Pass	Warning
splice_sites_donors	Warning	Warning	Warning	Pass	Warning	Warning	Warning	Pass	Pass	Warning
promoter_tata	Fail	Fail	Fail	Pass	Warning	Warning	Warning	Pass	Pass	Pass
H3K27me3	Fail	Fail	Fail	Pass	Pass	Pass	Warning	Pass	Pass	Warning
H3K36me3	Fail	Fail	Fail	Pass	Pass	Pass	Warning	Pass	Pass	Warning
H3K4me2	Fail	Fail	Fail	Pass	Pass	Pass	Warning	Pass	Pass	Warning
H3K4me3	Fail	Fail	Fail	Pass	Pass	Pass	Warning	Pass	Pass	Warning
H3K9ac	Fail	Fail	Fail	Pass	Pass	Pass	Warning	Pass	Pass	Warning
H4K20me1	Fail	Fail	Fail	Pass	Pass	Pass	Warning	Pass	Pass	Warning
H2AFZ	Fail	Warning	Fail	Pass	Pass	Pass	Warning	Pass	Pass	Warning
H3K27ac	Fail	Warning	Fail	Pass	Pass	Pass	Warning	Pass	Pass	Warning
H3K4me1	Fail	Warning	Fail	Pass	Pass	Pass	Warning	Pass	Pass	Warning
splice_sites_acceptors	Pass	Pass	Warning	Pass	Warning	Warning	Warning	Pass	Pass	Warning
H3K9me3	Pass	Pass	Pass	Pass	Pass	Pass	Warning	Pass	Pass	Warning

Pass
Warning
Fail
the check had nothing to run on

Every mark here is a link. A cell opens that dataset’s 10 checks, a name opens its page, a collection opens the catalogue filtered to it, and a column heading opens what that check returned across the corpus. Flags are on the reference split; grey is a check that had nothing to run on, which is not the same statement as Pass.

The ten checks

What a flag is

The split these numbers are on

What each check returned

Reading a probe AUROC

Reading a split overlap

Every dataset, every check

Move the boundaries

Where to go next