EH14 X

22.2 CPU hours measuring · 1,145 values · 26 reports · 05 November 2026, 13:05 CET

From the other models: 8 on BGC-1 and 9 on EH18 Designer.

task	ours	baseline		margin
chimera or artefact	0.9683	0.5015	4-mer counting in both halves	0.4668
marine novelty vs GenomeOcean-500M	0.7528	0.4631	GenomeOcean-500M	0.2897
marine novelty vs GROVER	0.7528	0.4790	GROVER	0.2738
marine novelty vs nucleotide-transformer-500m-human-ref	0.7528	0.4971	nucleotide-transformer-500m-human-ref	0.2557
marine novelty vs evo2-40b (NVIDIA, hosted)	0.7528	0.5396	evo2-40b (NVIDIA, hosted)	0.2132
marine novelty vs DNA_bert_6	0.7528	0.5520	DNA_bert_6	0.2008
benchmark: dummy_mouse_enhancers_ensembl	0.8019	0.6944	4-mer counting	0.1075
compound class: EH14 X, 512-base windows	0.5031	0.4625	nearest known neighbour	0.0406
taxonomic placement vs nucleotide-transformer-500m-human-ref	0.2925	0.2525	nucleotide-transformer-500m-human-ref	0.0400
chimera vs the tool labs run	0.9735	0.9442	UCHIME (vsearch)	0.0293
benchmark: drosophila_enhancers_stark	0.6916	0.6649	4-mer counting	0.0267
benchmark: human_nontata_promoters	0.9087	0.8856	4-mer counting	0.0231
benchmark: demo_coding_vs_intergenomic_seqs	0.9596	0.9405	4-mer counting	0.0191
benchmark: human_enhancers_ensembl	0.7993	0.7938	4-mer counting	0.0055
benchmark: demo_human_or_worm	0.9796	0.9776	4-mer counting	0.0020
benchmark: human_ocr_ensembl	0.7087	0.7082	4-mer counting	0.0005
taxonomic placement vs EH14 X	0.2925	0.2925	EH14 X	0.0000
benchmark: human_enhancers_cohn	0.7780	0.7899	4-mer counting	-0.0119
marine novelty, mean over 40 held-out taxa (model ahead on 14)	0.6656	0.7034	4-mer counting	-0.0379
variant effect	0.3687	0.5000	chance	-0.1313
taxonomic placement vs DNA_bert_6	0.2925	0.4825	DNA_bert_6	-0.1900
taxonomic placement vs GROVER	0.2925	0.5250	GROVER	-0.2325
generative design (18S, marine): ciliophora	0.1333	0.5167	Markov chain	-0.3834
taxonomic placement	0.2925	0.7025	4-mer counting	-0.4100
taxonomic placement vs 4-mer counting	0.2925	0.7025	4-mer counting	-0.4100
generative design (18S, marine): dinoflagellata	0.1000	0.6167	Markov chain	-0.5167
generative design (18S, marine): metazoa	0.0667	0.7167	Markov chain	-0.6500

eh14x69

window	2,048
vocab	6
tokenizer	per base, no BPE
tasks › marine novelty › baseline	4-mer counting
tasks › marine novelty › baseline value at 300 bases	0.7764
tasks › marine novelty › baseline value at 512 bases	0.8227
tasks › marine novelty › measured at 300 bases, SMALL MODEL	0.6386
tasks › marine novelty › measured at 512 bases, SMALL MODEL	0.7093
tasks › marine novelty › measured at 300 bases	0.7528
tasks › marine novelty › measured at 512 bases	0.7969
tasks › marine novelty › at length	300
tasks › taxonomic placement › baseline	4-mer counting
tasks › taxonomic placement › baseline value	0.7025
tasks › taxonomic placement › measured	0.2925
tasks › genomic benchmarks › baseline	4-mers + linear layer, same split
tasks › genomic benchmarks › baseline value	—
tasks › genomic benchmarks › measured	4 of 8 won demonstrably
tasks › chimera or artefact › baseline value	0.5015
tasks › chimera or artefact › model value	0.9683
tasks › chimera or artefact › difference	0.4668
tasks › chimera or artefact › interval	0.4527 – 0.4866
tasks › chimera or artefact › controls › labels through each other	0.4409
tasks › chimera or artefact › controls › only the first quarter, byte for byte identical	0.5037
tasks › chimera or artefact › controls › only the difference between the halves	0.9672
tasks › chimera or artefact › against the real tool › uchime_vsearch_2.32	0.9442
tasks › chimera or artefact › against the real tool › our head	0.9735
tasks › chimera or artefact › against the real tool › difference	0.0293
tasks › chimera or artefact › against the real tool › interval	0.0208 – 0.0385
tasks › variant effect › baseline	chance
tasks › variant effect › baseline value	0.5
tasks › variant effect › measured	0.3687
competitors on marine novelty › 4-mer counting	0.7764
competitors on marine novelty › EH14 X (257 M, SERVED)	0.7528
competitors on marine novelty › EH14 X (6.7 M, no longer served)	0.6386
competitors on marine novelty › GenomeOcean-4B	0.5715
competitors on marine novelty › DNABERT-6	0.552
competitors on marine novelty › Evo 2 (40 mrd)	0.5396
competitors on marine novelty › GenomeOcean-4B-bgcFM	0.5054
competitors on marine novelty › chance	0.5
competitors on marine novelty › NT-500M	0.4971
competitors on marine novelty › GROVER	0.479
competitors on marine novelty › GenomeOcean-500M	0.4631
competitors on marine novelty › at length	300
at the site	yes
next model plan › name	eh14x-next
next model plan › parameters	250,000,000
next model plan › window	2,048
next model plan › target bases	6,000,000,000
next model plan › tokens per parameter	24
nu › serveert › parameters	257,298,562
nu › serveert › window	512
nu › serveert › vocab	6
nu › serveert › bases read	4,499,996,672
nu › serveert › marine novelty	0.7969
nu › serveert › interval	0.765 – 0.8251
nu › serveert › since	2026-09-21, checked against the live API on 18 October 2026
nu › previous version not longer served › parameters	6,699,138
nu › previous version not longer served › window	512
nu › previous version not longer served › vocab	6
nu › previous version not longer served › bases read	937,164,800
nu › previous version not longer served › marine novelty	0.7093
nu › previous version not longer served › interval	0.675 – 0.7436
nu › baseline 4-mers at 512	0.8227
nu › bigger helps › difference	0.0876
nu › bigger helps › intervals separated	yes
nu › marine novelty at 300 bases › served model	0.7528
nu › marine novelty at 300 bases › interval	0.722 – 0.7828
nu › marine novelty at 300 bases › baseline 4-mers at 300	0.7764
nu › marine novelty at 300 bases › measured	18 oktober 2026, gereedschap/getallen_gelijktrekken.py

uchime_vergelijk9

checkpoint	EH14 X, 6.7 M (previous version)
n real	4,000
n chimera	2,000
uchime	0.9442
uchime anything found	3,928
baseline 4-mers	0.4989
chimera head	0.9735
difference head min uchime	0.0293
interval	0.0208 – 0.0385

chimeerkop15

checkpoint	EH14 X, 6.7 M (previous version)
groups held-out	90
cut points[0]	0.25
cut points[1]	0.375
cut points[2]	0.5
cut points[3]	0.625
cut points[4]	0.75
baseline 4-mers	0.5
head at frozen body	0.9683
control labels through each other	0.4409
control only first half	0.7746
control first quarter identical	0.5
control only difference	0.9672
difference model min baseline	0.4683
interval	0.4511 – 0.4851

ranglijst134

when	2026-11-02 17:32
how	frozen model, linear layer, no finetuning
tasks › demo_coding_vs_intergenomic_seqs › task	demo_coding_vs_intergenomic_seqs
tasks › demo_coding_vs_intergenomic_seqs › train	4,000
tasks › demo_coding_vs_intergenomic_seqs › test	4,000
tasks › demo_coding_vs_intergenomic_seqs › counting › auc	0.9405
tasks › demo_coding_vs_intergenomic_seqs › counting › interval	0.933 – 0.9473
tasks › demo_coding_vs_intergenomic_seqs › model › auc	0.9596
tasks › demo_coding_vs_intergenomic_seqs › model › interval	0.9546 – 0.9649
tasks › demo_coding_vs_intergenomic_seqs › difference, paired on the same sequences › difference	0.0191
tasks › demo_coding_vs_intergenomic_seqs › difference, paired on the same sequences › ci95 difference	0.013 – 0.0253
tasks › demo_coding_vs_intergenomic_seqs › difference, paired on the same sequences › share of bootstrap rounds where A scores higher	1
tasks › demo_coding_vs_intergenomic_seqs › difference, paired on the same sequences › bootstrap rounds	2,000
tasks › demo_coding_vs_intergenomic_seqs › difference, paired on the same sequences › ci95_a	0.9543 – 0.9648
tasks › demo_coding_vs_intergenomic_seqs › difference, paired on the same sequences › ci95_b	0.9334 – 0.947
tasks › demo_coding_vs_intergenomic_seqs › difference, paired on the same sequences › holds until the interval is this many times wider	3.13
tasks › demo_coding_vs_intergenomic_seqs › when	2026-11-02 17:40
tasks › demo_coding_vs_intergenomic_seqs › published cnn	0.892
tasks › demo_coding_vs_intergenomic_seqs › minutes	8.1
tasks › demo_human_or_worm › task	demo_human_or_worm
tasks › demo_human_or_worm › train	4,000
tasks › demo_human_or_worm › test	4,000
tasks › demo_human_or_worm › counting › auc	0.9776
tasks › demo_human_or_worm › counting › interval	0.9738 – 0.9814
tasks › demo_human_or_worm › model › auc	0.9796
tasks › demo_human_or_worm › model › interval	0.976 – 0.9832
tasks › demo_human_or_worm › difference, paired on the same sequences › difference	0.002
tasks › demo_human_or_worm › difference, paired on the same sequences › ci95 difference	-0.0016 – 0.0057
tasks › demo_human_or_worm › difference, paired on the same sequences › share of bootstrap rounds where A scores higher	0.857
tasks › demo_human_or_worm › difference, paired on the same sequences › bootstrap rounds	2,000
tasks › demo_human_or_worm › difference, paired on the same sequences › ci95_a	0.976 – 0.9831
tasks › demo_human_or_worm › difference, paired on the same sequences › ci95_b	0.9736 – 0.9813
tasks › demo_human_or_worm › when	2026-11-02 17:47
tasks › demo_human_or_worm › published cnn	0.961
tasks › demo_human_or_worm › minutes	6.8
tasks › human_enhancers_cohn › task	human_enhancers_cohn
tasks › human_enhancers_cohn › train	4,000
tasks › human_enhancers_cohn › test	4,000
tasks › human_enhancers_cohn › counting › auc	0.7899
tasks › human_enhancers_cohn › counting › interval	0.7771 – 0.8045
tasks › human_enhancers_cohn › model › auc	0.778
tasks › human_enhancers_cohn › model › interval	0.7648 – 0.7915
tasks › human_enhancers_cohn › difference, paired on the same sequences › difference	-0.0118
tasks › human_enhancers_cohn › difference, paired on the same sequences › ci95 difference	-0.0226 – -0.0014
tasks › human_enhancers_cohn › difference, paired on the same sequences › share of bootstrap rounds where A scores higher	0.0145
tasks › human_enhancers_cohn › difference, paired on the same sequences › bootstrap rounds	2,000
tasks › human_enhancers_cohn › difference, paired on the same sequences › ci95_a	0.7637 – 0.7919
tasks › human_enhancers_cohn › difference, paired on the same sequences › ci95_b	0.7761 – 0.804
tasks › human_enhancers_cohn › difference, paired on the same sequences › holds until the interval is this many times wider	1.13
tasks › human_enhancers_cohn › when	2026-11-02 17:52
tasks › human_enhancers_cohn › published cnn	0.715
tasks › human_enhancers_cohn › minutes	5.6
tasks › human_enhancers_ensembl › task	human_enhancers_ensembl
tasks › human_enhancers_ensembl › train	4,000
tasks › human_enhancers_ensembl › test	4,000
tasks › human_enhancers_ensembl › counting › auc	0.7938
tasks › human_enhancers_ensembl › counting › interval	0.7807 – 0.8077
tasks › human_enhancers_ensembl › model › auc	0.7993
tasks › human_enhancers_ensembl › model › interval	0.7866 – 0.8113
tasks › human_enhancers_ensembl › difference, paired on the same sequences › difference	0.0055
tasks › human_enhancers_ensembl › difference, paired on the same sequences › ci95 difference	-0.0085 – 0.0198
tasks › human_enhancers_ensembl › difference, paired on the same sequences › share of bootstrap rounds where A scores higher	0.784
tasks › human_enhancers_ensembl › difference, paired on the same sequences › bootstrap rounds	2,000
tasks › human_enhancers_ensembl › difference, paired on the same sequences › ci95_a	0.7857 – 0.8124
tasks › human_enhancers_ensembl › difference, paired on the same sequences › ci95_b	0.7798 – 0.8065
tasks › human_enhancers_ensembl › when	2026-11-02 17:57
tasks › human_enhancers_ensembl › published cnn	0.744
tasks › human_enhancers_ensembl › minutes	5
tasks › human_nontata_promoters › task	human_nontata_promoters
tasks › human_nontata_promoters › train	4,000
tasks › human_nontata_promoters › test	4,000
tasks › human_nontata_promoters › counting › auc	0.8856
tasks › human_nontata_promoters › counting › interval	0.8745 – 0.8965
tasks › human_nontata_promoters › model › auc	0.9087
tasks › human_nontata_promoters › model › interval	0.9 – 0.9178
tasks › human_nontata_promoters › difference, paired on the same sequences › difference	0.0232
tasks › human_nontata_promoters › difference, paired on the same sequences › ci95 difference	0.0149 – 0.0314
tasks › human_nontata_promoters › difference, paired on the same sequences › share of bootstrap rounds where A scores higher	1
tasks › human_nontata_promoters › difference, paired on the same sequences › bootstrap rounds	2,000
tasks › human_nontata_promoters › difference, paired on the same sequences › ci95_a	0.8992 – 0.9178
tasks › human_nontata_promoters › difference, paired on the same sequences › ci95_b	0.8744 – 0.8959
tasks › human_nontata_promoters › difference, paired on the same sequences › holds until the interval is this many times wider	2.8
tasks › human_nontata_promoters › when	2026-11-02 18:02
tasks › human_nontata_promoters › published cnn	0.846
tasks › human_nontata_promoters › minutes	4.8
tasks › human_ocr_ensembl › task	human_ocr_ensembl
tasks › human_ocr_ensembl › train	4,000
tasks › human_ocr_ensembl › test	4,000
tasks › human_ocr_ensembl › counting › auc	0.7082
tasks › human_ocr_ensembl › counting › interval	0.6927 – 0.7247
tasks › human_ocr_ensembl › model › auc	0.7087
tasks › human_ocr_ensembl › model › interval	0.6929 – 0.7244
tasks › human_ocr_ensembl › difference, paired on the same sequences › difference	0.0005
tasks › human_ocr_ensembl › difference, paired on the same sequences › ci95 difference	-0.0155 – 0.0164
tasks › human_ocr_ensembl › difference, paired on the same sequences › share of bootstrap rounds where A scores higher	0.5305
tasks › human_ocr_ensembl › difference, paired on the same sequences › bootstrap rounds	2,000
tasks › human_ocr_ensembl › difference, paired on the same sequences › ci95_a	0.6933 – 0.7247
tasks › human_ocr_ensembl › difference, paired on the same sequences › ci95_b	0.692 – 0.7243
tasks › human_ocr_ensembl › when	2026-11-02 18:07
tasks › human_ocr_ensembl › published cnn	0.68
tasks › human_ocr_ensembl › minutes	4.9
tasks › drosophila_enhancers_stark › task	drosophila_enhancers_stark
tasks › drosophila_enhancers_stark › train	4,000
tasks › drosophila_enhancers_stark › test	1,730
tasks › drosophila_enhancers_stark › counting › auc	0.6649
tasks › drosophila_enhancers_stark › counting › interval	0.6384 – 0.687
tasks › drosophila_enhancers_stark › model › auc	0.6916
tasks › drosophila_enhancers_stark › model › interval	0.6658 – 0.7161
tasks › drosophila_enhancers_stark › difference, paired on the same sequences › difference	0.0267
tasks › drosophila_enhancers_stark › difference, paired on the same sequences › ci95 difference	-0.0017 – 0.054
tasks › drosophila_enhancers_stark › difference, paired on the same sequences › share of bootstrap rounds where A scores higher	0.969
tasks › drosophila_enhancers_stark › difference, paired on the same sequences › bootstrap rounds	2,000
tasks › drosophila_enhancers_stark › difference, paired on the same sequences › ci95_a	0.6671 – 0.7148
tasks › drosophila_enhancers_stark › difference, paired on the same sequences › ci95_b	0.6389 – 0.6893
tasks › drosophila_enhancers_stark › when	2026-11-02 18:11
tasks › drosophila_enhancers_stark › published cnn	0.586
tasks › drosophila_enhancers_stark › minutes	4.3
tasks › dummy_mouse_enhancers_ensembl › task	dummy_mouse_enhancers_ensembl
tasks › dummy_mouse_enhancers_ensembl › train	968
tasks › dummy_mouse_enhancers_ensembl › test	242
tasks › dummy_mouse_enhancers_ensembl › counting › auc	0.6944
tasks › dummy_mouse_enhancers_ensembl › counting › interval	0.635 – 0.7614
tasks › dummy_mouse_enhancers_ensembl › model › auc	0.8019
tasks › dummy_mouse_enhancers_ensembl › model › interval	0.7429 – 0.8628
tasks › dummy_mouse_enhancers_ensembl › difference, paired on the same sequences › difference	0.1076
tasks › dummy_mouse_enhancers_ensembl › difference, paired on the same sequences › ci95 difference	0.0368 – 0.1734
tasks › dummy_mouse_enhancers_ensembl › difference, paired on the same sequences › share of bootstrap rounds where A scores higher	0.999
tasks › dummy_mouse_enhancers_ensembl › difference, paired on the same sequences › bootstrap rounds	2,000
tasks › dummy_mouse_enhancers_ensembl › difference, paired on the same sequences › ci95_a	0.7438 – 0.8587
tasks › dummy_mouse_enhancers_ensembl › difference, paired on the same sequences › ci95_b	0.6261 – 0.7609
tasks › dummy_mouse_enhancers_ensembl › difference, paired on the same sequences › holds until the interval is this many times wider	1.52
tasks › dummy_mouse_enhancers_ensembl › when	2026-11-02 18:12
tasks › dummy_mouse_enhancers_ensembl › published cnn	0.715
tasks › dummy_mouse_enhancers_ensembl › minutes	1

concurrenten25

grover › model	PoetschLab/GROVER
grover › auc	0.479
grover › ci95	0.4361 – 0.5143
grover › known	400
grover › novel	400
grover › masked_positions	24
grover › length	300
grover › when	2026-09-22 19:15
nt-500m › model	InstaDeepAI/nucleotide-transformer-500m-human-ref
nt-500m › auc	0.4971
nt-500m › ci95	0.4614 – 0.5366
nt-500m › known	400
nt-500m › novel	400
nt-500m › masked_positions	24
nt-500m › length	300
nt-500m › when	2026-09-22 20:45
dnabert-6 › model	zhihan1996/DNA_bert_6
dnabert-6 › auc	0.552
dnabert-6 › ci95	0.5117 – 0.5914
dnabert-6 › known	400
dnabert-6 › novel	400
dnabert-6 › masked_positions	24
dnabert-6 › length	300
dnabert-6 › when	2026-09-22 22:09
measured with	EH14 X, 255 M – EH14 X, 6.7 M (previous version)

zeemeting13

model	EH14 X, 6.7 M (previous version)
bases read	937,164,800
train_loss	0.7524
known	800
novel	800
held-out taxa	40
window	512
auc	0.7309
ci95	0.707 – 0.7532
baseline › 4-mer counting	0.8227
baseline › chance	0.5
baseline › at length	512
minutes	0.7

besmettingsproef25

arms › contaminated › bases read	135,168,000
arms › contaminated › train_loss	0.7698
arms › contaminated › auc	0.5654
arms › contaminated › ci95	0.5259 – 0.6036
arms › contaminated › known	400
arms › contaminated › novel	400
arms › contaminated › window	512
arms › contaminated › held-out taxa	40
arms › clean › bases read	937,164,800
arms › clean › train_loss	0.7524
arms › clean › auc	0.7093
arms › clean › ci95	0.675 – 0.7436
arms › clean › known	400
arms › clean › novel	400
arms › clean › window	512
arms › clean › held-out taxa	40
paired, resampled by group › taxa in this measurement	34
paired, resampled by group › taxa held out	40
paired, resampled by group › difference	0.1439
paired, resampled by group › resampled by row › ci95	0.1272 – 0.1609
paired, resampled by group › resampled by row › width	0.0337
paired, resampled by group › resampled by taxon › ci95	0.0838 – 0.2011
paired, resampled by group › resampled by taxon › width	0.1173
paired, resampled by group › resampled by taxon › groups on the novel side	34
paired, resampled by group › how much wider than by row	3.48

breedtest160

background broad	40,000
background design	1,476
differences allowed	1
probes[0] › genus	Candida
probes[0] › which	counted
probes[0] › background at design	0.06
probes[0] › background broad	0.0576
probes[0] › times as many	1
probes[0] › tested against	39,956
probes[1] › genus	Candida
probes[1] › which	designed
probes[1] › background at design	0.0025
probes[1] › background broad	0.0013
probes[1] › times as many	0.5
probes[1] › tested against	39,956
probes[2] › genus	Aspergillus
probes[2] › which	counted
probes[2] › background at design	0.01
probes[2] › background broad	0.0124
probes[2] › times as many	1.2
probes[2] › tested against	39,960
probes[3] › genus	Aspergillus
probes[3] › which	designed
probes[3] › background at design	0.0025
probes[3] › background broad	0.0019
probes[3] › times as many	0.8
probes[3] › tested against	39,960
probes[4] › genus	Cryptosporidium
probes[4] › which	counted
probes[4] › background at design	0.0375
probes[4] › background broad	0.0339
probes[4] › times as many	0.9
probes[4] › tested against	39,937
probes[5] › genus	Cryptosporidium
probes[5] › which	designed
probes[5] › background at design	0.0125
probes[5] › background broad	0.0132
probes[5] › times as many	1.1
probes[5] › tested against	39,937
probes[6] › genus	Plasmodium
probes[6] › which	counted
probes[6] › background at design	0
probes[6] › background broad	0.0001
probes[6] › times as many	—
probes[6] › tested against	39,979
probes[7] › genus	Plasmodium
probes[7] › which	designed
probes[7] › background at design	0
probes[7] › background broad	0.0001
probes[7] › times as many	—
probes[7] › tested against	39,979
probes[8] › genus	Toxoplasma
probes[8] › which	counted
probes[8] › background at design	0.01
probes[8] › background broad	0.0072
probes[8] › times as many	0.7
probes[8] › tested against	39,994
probes[9] › genus	Toxoplasma
probes[9] › which	designed
probes[9] › background at design	0
probes[9] › background broad	0.0012
probes[9] › times as many	—
probes[9] › tested against	39,994
probes[10] › genus	Trypanosoma
probes[10] › which	counted
probes[10] › background at design	0
probes[10] › background broad	0.0011
probes[10] › times as many	—
probes[10] › tested against	39,968
probes[11] › genus	Trypanosoma
probes[11] › which	designed
probes[11] › background at design	0
probes[11] › background broad	0.0011
probes[11] › times as many	—
probes[11] › tested against	39,968
probes[12] › genus	Leishmania
probes[12] › which	counted
probes[12] › background at design	0
probes[12] › background broad	0.0012
probes[12] › times as many	—
probes[12] › tested against	39,994
probes[13] › genus	Leishmania
probes[13] › which	designed
probes[13] › background at design	0
probes[13] › background broad	0.0012
probes[13] › times as many	—
probes[13] › tested against	39,994
probes[14] › genus	Entamoeba
probes[14] › which	counted
probes[14] › background at design	0.0025
probes[14] › background broad	0.0024
probes[14] › times as many	0.9
probes[14] › tested against	39,995
probes[15] › genus	Entamoeba
probes[15] › which	designed
probes[15] › background at design	0
probes[15] › background broad	0.0012
probes[15] › times as many	—
probes[15] › tested against	39,995
probes[16] › genus	Naegleria
probes[16] › which	counted
probes[16] › background at design	0
probes[16] › background broad	0
probes[16] › times as many	—
probes[16] › tested against	39,996
probes[17] › genus	Naegleria
probes[17] › which	designed
probes[17] › background at design	0
probes[17] › background broad	0
probes[17] › times as many	—
probes[17] › tested against	39,996
probes[18] › genus	Acanthamoeba
probes[18] › which	counted
probes[18] › background at design	0.0175
probes[18] › background broad	0.0274
probes[18] › times as many	1.6
probes[18] › tested against	39,980
probes[19] › genus	Acanthamoeba
probes[19] › which	designed
probes[19] › background at design	0
probes[19] › background broad	0.0003
probes[19] › times as many	—
probes[19] › tested against	39,980
probes[20] › genus	Babesia
probes[20] › which	counted
probes[20] › background at design	0.135
probes[20] › background broad	0.1258
probes[20] › times as many	0.9
probes[20] › tested against	39,976
probes[21] › genus	Babesia
probes[21] › which	designed
probes[21] › background at design	0
probes[21] › background broad	0.0026
probes[21] › times as many	—
probes[21] › tested against	39,976
probes[22] › genus	Trichomonas
probes[22] › which	counted
probes[22] › background at design	0
probes[22] › background broad	0.0005
probes[22] › times as many	—
probes[22] › tested against	39,989
probes[23] › genus	Trichomonas
probes[23] › which	designed
probes[23] › background at design	0
probes[23] › background broad	0.0005
probes[23] › times as many	—
probes[23] › tested against	39,989
probes[24] › genus	Blastocystis
probes[24] › which	counted
probes[24] › background at design	0
probes[24] › background broad	0.0003
probes[24] › times as many	—
probes[24] › tested against	39,984
probes[25] › genus	Blastocystis
probes[25] › which	designed
probes[25] › background at design	0
probes[25] › background broad	0.0001
probes[25] › times as many	—
probes[25] › tested against	39,984
minutes	0.4

clustermeting15

clusters	1,604
families	1,534
train	1,284
test	320
measurement › largest class guessing › correct	0.2281
measurement › largest class guessing › interval	0.1812 – 0.2751
measurement › 4-mers + linear layer › correct	0.4594
measurement › 4-mers + linear layer › interval	0.4093 – 0.5125
measurement › EH14 X, 512-base windows › correct	0.5031
measurement › EH14 X, 512-base windows › interval	0.4531 – 0.5563
measurement › BGC-1, 512-base windows › correct	0.6937
measurement › BGC-1, 512-base windows › interval	0.6468 – 0.7438
measurement › BGC-1, 2048-base windows › correct	0.7406
measurement › BGC-1, 2048-base windows › interval	0.6906 – 0.7875
measured with	EH14 X, 6.7 M (previous version) – BGC-1

concurrenten_plaatsing36

grover › model	PoetschLab/GROVER
grover › accuracy	0.525
grover › ci95	0.4761 – 0.5735
grover › queries	400
grover › groups	60
grover › chance	0.0167
grover › when	2026-09-22 22:57
dnabert-6 › model	zhihan1996/DNA_bert_6
dnabert-6 › accuracy	0.4825
dnabert-6 › ci95	0.4339 – 0.5314
dnabert-6 › queries	400
dnabert-6 › groups	60
dnabert-6 › chance	0.0167
dnabert-6 › when	2026-09-22 23:18
eh14x › model	EH14 X
eh14x › accuracy	0.2925
eh14x › ci95	0.2501 – 0.3389
eh14x › queries	400
eh14x › groups	60
eh14x › chance	0.0167
eh14x › when	2026-09-22 23:26
kmer › model	4-mer counting
kmer › accuracy	0.7025
kmer › ci95	0.6559 – 0.7452
kmer › queries	400
kmer › groups	60
kmer › chance	0.0167
kmer › when	2026-09-22 23:48
nt-500m › model	InstaDeepAI/nucleotide-transformer-500m-human-ref
nt-500m › accuracy	0.2525
nt-500m › ci95	0.2124 – 0.2973
nt-500m › queries	400
nt-500m › groups	60
nt-500m › chance	0.0167
nt-500m › when	2026-09-23 00:10
measured with	EH14 X, 6.7 M (previous version)

corpus_telling4

size	7,171,796,732
rules	12,824,758
on	02-10-2026 13:18
measured with	no model

eh14x-50M20

model	EH14 X, 60 M
parameters	59,719,809
window	512
tokens_G	1.9
final loss	0.6879
skipped steps	0
duration minutes	172
card	RTX 4090
marine novelty › correct	0.7313
marine novelty › interval	0.6957 – 0.7618
marine novelty › baseline	0.7764
marine novelty › baseline method	4-mer counting, 300 bases
marine novelty › difference	-0.0451
marine novelty › at length	300
curve correct measured › 0.57	0.6911
curve correct measured › 1.33	0.7267
curve correct measured › 1.90	0.7313
curve with padding › 0.57	0.6802
curve with padding › 1.33	0.5282
curve with padding › 1.90	0.5228

evo2_vergelijk9

measured on	150 known, 150 new, 40 held-out taxa
length	300
evo2 › model	evo2-40b (NVIDIA, hosted)
evo2 › auc	0.5396
evo2 › ci95	0.4699 – 0.6042
evo2 › failed	0
counting › model	4-mer counting
counting › auc	0.7764
we	—

genomeocean13

model	DOEJGI/GenomeOcean-500M
size	500M
auc	0.4631
ci95	0.402 – 0.526
length	300
measured on	200 known, 200 new, 40 held-out taxa
baseline › 4-mer counting	0.7764
baseline › chance	0.5
baseline › at length	300
ours › model	EH14 X, 6.7 M (previous version)
ours › auc	0.6386
ours › ci95	0.602 – 0.6709
ours › window	300

gpu-uren7

bgc-1 › gpu hours	180.85
bgc-1 › gpu minutes	10,851
bgc-1 › cpu hours	0
eh14x › gpu hours	0
eh14x › gpu minutes	0
eh14x › cpu hours	22.2
eh14x › cpu minutes	1,331

heldout11

version	biox-dna-0
bases_read	106,496,000
million_bases	106.5
train_loss	1.1533
fresh_loss	1.1628
fresh_perplexity	3.1989
gap	0.0095
fresh_loss_se	0.0085
gap_in_se	1.13
windows	960
genome_seen_fraction	0.0345

novel_kmer174

method	k-mer (k=4), gecentreerd
centred	yes
reference_records	190,379
levels[0] › rank_removed	7
levels[0] › graded_at_rank	6
levels[0] › groups	10
levels[0] › records	3,801
levels[0] › accuracy	0.3002
levels[0] › chance	0.0046
levels[0] › lift	65.11
levels[0] › per_group[0] › group	glomus
levels[0] › per_group[0] › parent	glomerales
levels[0] › per_group[0] › n	1,351
levels[0] › per_group[0] › accuracy	0.5426
levels[0] › per_group[0] › chance	0.0013
levels[0] › per_group[1] › group	maxillopoda_x
levels[0] › per_group[1] › parent	maxillopoda
levels[0] › per_group[1] › n	956
levels[0] › per_group[1] › accuracy	0
levels[0] › per_group[1] › chance	0
levels[0] › per_group[2] › group	cryptococcus
levels[0] › per_group[2] › parent	tremellomycetes
levels[0] › per_group[2] › n	544
levels[0] › per_group[2] › accuracy	0
levels[0] › per_group[2] › chance	0
levels[0] › per_group[3] › group	cladosporium
levels[0] › per_group[3] › parent	dothideomycetes
levels[0] › per_group[3] › n	256
levels[0] › per_group[3] › accuracy	0.6719
levels[0] › per_group[3] › chance	0.0347
levels[0] › per_group[4] › group	desmodesmus
levels[0] › per_group[4] › parent	sphaeropleales_x
levels[0] › per_group[4] › n	183
levels[0] › per_group[4] › accuracy	0
levels[0] › per_group[4] › chance	0
levels[0] › per_group[5] › group	kinetoplastea_xxx
levels[0] › per_group[5] › parent	kinetoplastea_xx
levels[0] › per_group[5] › n	162
levels[0] › per_group[5] › accuracy	0.5926
levels[0] › per_group[5] › chance	0.0021
levels[0] › per_group[6] › group	solanum
levels[0] › per_group[6] › parent	embryophyceae_xx
levels[0] › per_group[6] › n	109
levels[0] › per_group[6] › accuracy	0.2385
levels[0] › per_group[6] › chance	0.0302
levels[0] › per_group[7] › group	davidiella
levels[0] › per_group[7] › parent	dothideomycetes
levels[0] › per_group[7] › n	91
levels[0] › per_group[7] › accuracy	0.4615
levels[0] › per_group[7] › chance	0.0347
levels[0] › per_group[8] › group	smittium
levels[0] › per_group[8] › parent	harpellales
levels[0] › per_group[8] › n	75
levels[0] › per_group[8] › accuracy	0
levels[0] › per_group[8] › chance	0
levels[0] › per_group[9] › group	tardigrada_xxx
levels[0] › per_group[9] › parent	tardigrada_xx
levels[0] › per_group[9] › n	74
levels[0] › per_group[9] › accuracy	0.973
levels[0] › per_group[9] › chance	0.0013
levels[1] › rank_removed	6
levels[1] › graded_at_rank	5
levels[1] › groups	10
levels[1] › records	9,470
levels[1] › accuracy	0.4456
levels[1] › chance	0.0554
levels[1] › lift	8.04
levels[1] › per_group[0] › group	eurotiomycetes
levels[1] › per_group[0] › parent	pezizomycotina
levels[1] › per_group[0] › n	4,157
levels[1] › per_group[0] › accuracy	0.504
levels[1] › per_group[0] › chance	0.1142
levels[1] › per_group[1] › group	maxillopoda
levels[1] › per_group[1] › parent	crustacea
levels[1] › per_group[1] › n	2,266
levels[1] › per_group[1] › accuracy	0.0071
levels[1] › per_group[1] › chance	0.0131
levels[1] › per_group[2] › group	sphaeropleales_x
levels[1] › per_group[2] › parent	sphaeropleales
levels[1] › per_group[2] › n	1,387
levels[1] › per_group[2] › accuracy	0.5948
levels[1] › per_group[2] › chance	0.0007
levels[1] › per_group[3] › group	glomeromycotina_x
levels[1] › per_group[3] › parent	glomeromycotina
levels[1] › per_group[3] › n	570
levels[1] › per_group[3] › accuracy	0.7561
levels[1] › per_group[3] › chance	0.0034
levels[1] › per_group[4] › group	dino-group-ii-clade-1
levels[1] › per_group[4] › parent	dino-group-ii
levels[1] › per_group[4] › n	362
levels[1] › per_group[4] › accuracy	0.6906
levels[1] › per_group[4] › chance	0.0238
levels[1] › per_group[5] › group	globigerinoidea
levels[1] › per_group[5] › parent	rotaliida
levels[1] › per_group[5] › n	306
levels[1] › per_group[5] › accuracy	0.9118
levels[1] › per_group[5] › chance	0.0153
levels[1] › per_group[6] › group	dictyosteliida-group-iv
levels[1] › per_group[6] › parent	dictyosteliida
levels[1] › per_group[6] › n	152
levels[1] › per_group[6] › accuracy	0.875
levels[1] › per_group[6] › chance	0.0009
levels[1] › per_group[7] › group	peronosporomycetes_xx
levels[1] › per_group[7] › parent	peronosporomycetes_x
levels[1] › per_group[7] › n	115
levels[1] › per_group[7] › accuracy	0.7391
levels[1] › per_group[7] › chance	0.0037
levels[1] › per_group[8] › group	dino-group-ii-clade-15
levels[1] › per_group[8] › parent	dino-group-ii
levels[1] › per_group[8] › n	78
levels[1] › per_group[8] › accuracy	0.4615
levels[1] › per_group[8] › chance	0.0238
levels[1] › per_group[9] › group	dino-group-ii-clade-2
levels[1] › per_group[9] › parent	dino-group-ii
levels[1] › per_group[9] › n	77
levels[1] › per_group[9] › accuracy	0.9091
levels[1] › per_group[9] › chance	0.0238
levels[2] › rank_removed	5
levels[2] › graded_at_rank	4
levels[2] › groups	10
levels[2] › records	11,883
levels[2] › accuracy	0.2567
levels[2] › chance	0.0221
levels[2] › lift	11.61
levels[2] › per_group[0] › group	agaricomycotina
levels[2] › per_group[0] › parent	basidiomycota
levels[2] › per_group[0] › n	4,873
levels[2] › per_group[0] › accuracy	0.0646
levels[2] › per_group[0] › chance	0.031
levels[2] › per_group[1] › group	dino-group-i
levels[2] › per_group[1] › parent	syndiniales
levels[2] › per_group[1] › n	3,200
levels[2] › per_group[1] › accuracy	0.3634
levels[2] › per_group[1] › chance	0.0261
levels[2] › per_group[2] › group	chlorellales
levels[2] › per_group[2] › parent	trebouxiophyceae
levels[2] › per_group[2] › n	1,679
levels[2] › per_group[2] › accuracy	0.5027
levels[2] › per_group[2] › chance	0.0058
levels[2] › per_group[3] › group	sagenista_x
levels[2] › per_group[3] › parent	sagenista
levels[2] › per_group[3] › n	732
levels[2] › per_group[3] › accuracy	0.0956
levels[2] › per_group[3] › chance	0.0063
levels[2] › per_group[4] › group	haemosporida
levels[2] › per_group[4] › parent	coccidiomorphea
levels[2] › per_group[4] › n	478
levels[2] › per_group[4] › accuracy	0.0732
levels[2] › per_group[4] › chance	0.0145
levels[2] › per_group[5] › group	mediophyceae_x
levels[2] › per_group[5] › parent	mediophyceae
levels[2] › per_group[5] › n	429
levels[2] › per_group[5] › accuracy	0.8695
levels[2] › per_group[5] › chance	0.0093
levels[2] › per_group[6] › group	kickxellomycotina
levels[2] › per_group[6] › parent	zoopagomycota
levels[2] › per_group[6] › n	203
levels[2] › per_group[6] › accuracy	0
levels[2] › per_group[6] › chance	0.0008
levels[2] › per_group[7] › group	cryptophyceae_x
levels[2] › per_group[7] › parent	cryptophyceae
levels[2] › per_group[7] › n	142
levels[2] › per_group[7] › accuracy	0.9085
levels[2] › per_group[7] › chance	0.0049
levels[2] › per_group[8] › group	coccolithales
levels[2] › per_group[8] › parent	prymnesiophyceae
levels[2] › per_group[8] › n	87
levels[2] › per_group[8] › accuracy	0.9655
levels[2] › per_group[8] › chance	0.0046
levels[2] › per_group[9] › group	dino-group-iv
levels[2] › per_group[9] › parent	syndiniales
levels[2] › per_group[9] › n	60
levels[2] › per_group[9] › accuracy	0.6333
levels[2] › per_group[9] › chance	0.0261

ocean_gap46

title	the catalogue against the water
occurrences › bands[0] › band	0-100 m
occurrences › bands[0] › records	82,221
occurrences › bands[0] › distinct_taxa	9,132
occurrences › bands[0] › distinct_species	6,109
occurrences › bands[0] › identified_to_species	49,294
occurrences › bands[0] › pct_to_species	60
occurrences › bands[1] › band	100-500 m
occurrences › bands[1] › records	19,277
occurrences › bands[1] › distinct_taxa	3,221
occurrences › bands[1] › distinct_species	2,125
occurrences › bands[1] › identified_to_species	14,478
occurrences › bands[1] › pct_to_species	75.1
occurrences › bands[2] › band	500-1000 m
occurrences › bands[2] › records	2,672
occurrences › bands[2] › distinct_taxa	1,060
occurrences › bands[2] › distinct_species	595
occurrences › bands[2] › identified_to_species	1,151
occurrences › bands[2] › pct_to_species	43.1
occurrences › bands[3] › band	1000+ m
occurrences › bands[3] › records	3,356
occurrences › bands[3] › distinct_taxa	1,110
occurrences › bands[3] › distinct_species	491
occurrences › bands[3] › identified_to_species	1,004
occurrences › bands[3] › pct_to_species	29.9
occurrences › excluded_negative_depth	170
occurrences › excluded_no_depth	102,299
sequences › groups[0] › group	shallow
sequences › groups[0] › sequences	80
sequences › groups[0] › named_pct	20
sequences › groups[0] › unnamed_pct	80
sequences › groups[0] › unnamed_pct_of_reads	77.8
sequences › groups[1] › group	deep
sequences › groups[1] › sequences	80
sequences › groups[1] › named_pct	6.2
sequences › groups[1] › unnamed_pct	93.8
sequences › groups[1] › unnamed_pct_of_reads	98.2
sequences › study	PRJEB3400
sequences › marker	18S V9
agreement › occurrence_pct_to_species_shallow	60
agreement › occurrence_pct_to_species_deep	29.9
agreement › occurrence_drop_pct_points	30.1
agreement › edna_unnamed_shallow_pct	80
agreement › edna_unnamed_deep_pct	93.8
agreement › edna_rise_pct_points	13.8
agreement › same_direction	yes

pertaxon205

model	EH14 X, 6.7 M (previous version)
bases read	498,073,600
known	200
taxa	40
model wins at	14
rows[0] › taxon	AKIW781
rows[0] › n	30
rows[0] › model	0.8907
rows[0] › counting	0.7868
rows[0] › difference	0.1038
rows[1] › taxon	Platyhelminthes
rows[1] › n	30
rows[1] › model	0.8437
rows[1] › counting	0.7527
rows[1] › difference	0.091
rows[2] › taxon	Pedosphaeraceae
rows[2] › n	30
rows[2] › model	0.8747
rows[2] › counting	0.792
rows[2] › difference	0.0827
rows[3] › taxon	Acanthocephala
rows[3] › n	30
rows[3] › model	0.9527
rows[3] › counting	0.9153
rows[3] › difference	0.0373
rows[4] › taxon	Puniceicoccaceae
rows[4] › n	30
rows[4] › model	0.8692
rows[4] › counting	0.8338
rows[4] › difference	0.0353
rows[5] › taxon	Oscillatoriaceae
rows[5] › n	30
rows[5] › model	0.4807
rows[5] › counting	0.4457
rows[5] › difference	0.035
rows[6] › taxon	Pirellulaceae
rows[6] › n	30
rows[6] › model	0.9107
rows[6] › counting	0.8762
rows[6] › difference	0.0345
rows[7] › taxon	Opitutaceae
rows[7] › n	30
rows[7] › model	0.853
rows[7] › counting	0.827
rows[7] › difference	0.026
rows[8] › taxon	Mollicutes
rows[8] › n	30
rows[8] › model	0.7698
rows[8] › counting	0.7503
rows[8] › difference	0.0195
rows[9] › taxon	CONThreeP
rows[9] › n	30
rows[9] › model	0.809
rows[9] › counting	0.794
rows[9] › difference	0.015
rows[10] › taxon	Moraxellaceae
rows[10] › n	30
rows[10] › model	0.5467
rows[10] › counting	0.5402
rows[10] › difference	0.0065
rows[11] › taxon	Mediophyceae:plas
rows[11] › n	30
rows[11] › model	0.3452
rows[11] › counting	0.3393
rows[11] › difference	0.0058
rows[12] › taxon	Symbiontida
rows[12] › n	30
rows[12] › model	0.999
rows[12] › counting	0.9972
rows[12] › difference	0.0018
rows[13] › taxon	Isosphaeraceae
rows[13] › n	30
rows[13] › model	0.9137
rows[13] › counting	0.9122
rows[13] › difference	0.0015
rows[14] › taxon	Melioribacteraceae
rows[14] › n	30
rows[14] › model	0.8265
rows[14] › counting	0.833
rows[14] › difference	-0.0065
rows[15] › taxon	Stappiaceae
rows[15] › n	30
rows[15] › model	0.2938
rows[15] › counting	0.3007
rows[15] › difference	-0.0068
rows[16] › taxon	Ulvophyceae
rows[16] › n	30
rows[16] › model	0.7192
rows[16] › counting	0.7273
rows[16] › difference	-0.0082
rows[17] › taxon	Babeliaceae
rows[17] › n	30
rows[17] › model	0.7485
rows[17] › counting	0.7615
rows[17] › difference	-0.013
rows[18] › taxon	Eubacteriaceae
rows[18] › n	30
rows[18] › model	0.7137
rows[18] › counting	0.7328
rows[18] › difference	-0.0192
rows[19] › taxon	Alcaligenaceae
rows[19] › n	30
rows[19] › model	0.5055
rows[19] › counting	0.5348
rows[19] › difference	-0.0293
rows[20] › taxon	E6aC02
rows[20] › n	30
rows[20] › model	0.6858
rows[20] › counting	0.7187
rows[20] › difference	-0.0328
rows[21] › taxon	Defluviitaleaceae
rows[21] › n	30
rows[21] › model	0.552
rows[21] › counting	0.585
rows[21] › difference	-0.033
rows[22] › taxon	Oligoflexaceae
rows[22] › n	30
rows[22] › model	0.702
rows[22] › counting	0.7362
rows[22] › difference	-0.0342
rows[23] › taxon	Thalassospiraceae
rows[23] › n	30
rows[23] › model	0.4868
rows[23] › counting	0.5365
rows[23] › difference	-0.0497
rows[24] › taxon	Acidaminobacteraceae
rows[24] › n	30
rows[24] › model	0.562
rows[24] › counting	0.6157
rows[24] › difference	-0.0537
rows[25] › taxon	Cytophagaceae
rows[25] › n	30
rows[25] › model	0.5713
rows[25] › counting	0.6368
rows[25] › difference	-0.0655
rows[26] › taxon	Rhodomicrobiaceae
rows[26] › n	30
rows[26] › model	0.2415
rows[26] › counting	0.3138
rows[26] › difference	-0.0723
rows[27] › taxon	Amoebophilaceae
rows[27] › n	30
rows[27] › model	0.7293
rows[27] › counting	0.8032
rows[27] › difference	-0.0738
rows[28] › taxon	Sporomusaceae
rows[28] › n	30
rows[28] › model	0.702
rows[28] › counting	0.776
rows[28] › difference	-0.074
rows[29] › taxon	Defluviicoccaceae
rows[29] › n	30
rows[29] › model	0.4102
rows[29] › counting	0.4848
rows[29] › difference	-0.0747
rows[30] › taxon	Syntrophaceae
rows[30] › n	30
rows[30] › model	0.6045
rows[30] › counting	0.6815
rows[30] › difference	-0.077
rows[31] › taxon	Rikenellaceae
rows[31] › n	30
rows[31] › model	0.7087
rows[31] › counting	0.7868
rows[31] › difference	-0.0782
rows[32] › taxon	Prolixibacteraceae
rows[32] › n	30
rows[32] › model	0.5605
rows[32] › counting	0.6435
rows[32] › difference	-0.083
rows[33] › taxon	Actinomarinaceae
rows[33] › n	30
rows[33] › model	0.7628
rows[33] › counting	0.8562
rows[33] › difference	-0.0933
rows[34] › taxon	Victivallaceae
rows[34] › n	30
rows[34] › model	0.6725
rows[34] › counting	0.7757
rows[34] › difference	-0.1032
rows[35] › taxon	Pseudanabaenaceae
rows[35] › n	30
rows[35] › model	0.517
rows[35] › counting	0.649
rows[35] › difference	-0.132
rows[36] › taxon	Methanobacteriaceae
rows[36] › n	30
rows[36] › model	0.7915
rows[36] › counting	0.9443
rows[36] › difference	-0.1528
rows[37] › taxon	Clostridiaceae
rows[37] › n	30
rows[37] › model	0.3518
rows[37] › counting	0.5422
rows[37] › difference	-0.1903
rows[38] › taxon	MALV-I
rows[38] › n	30
rows[38] › model	0.5435
rows[38] › counting	0.7582
rows[38] › difference	-0.2147
rows[39] › taxon	Pelagophyceae:plas
rows[39] › n	30
rows[39] › model	0.6017
rows[39] › counting	0.8407
rows[39] › difference	-0.239

telmeting27

300 › length	300
300 › known	400
300 › novel	400
300 › taxa	40
300 › taxa in this measurement	34
300 › fitted at	20,000
300 › seed	1
300 › ways › nearest reference › auc	0.7764
300 › ways › nearest reference › ci95	0.7471 – 0.8071
300 › ways › nearest reference › 95% interval, resampled by taxon	0.6938 – 0.8651
300 › ways › order-3 Markov, log-probability › auc	0.494
300 › ways › order-3 Markov, log-probability › ci95	0.4527 – 0.5351
300 › ways › order-3 Markov, log-probability › 95% interval, resampled by taxon	0.3708 – 0.631
512 › length	512
512 › known	400
512 › novel	400
512 › taxa	40
512 › taxa in this measurement	34
512 › fitted at	20,000
512 › seed	1
512 › ways › nearest reference › auc	0.8227
512 › ways › nearest reference › ci95	0.7958 – 0.8527
512 › ways › nearest reference › 95% interval, resampled by taxon	0.7458 – 0.9013
512 › ways › order-3 Markov, log-probability › auc	0.5144
512 › ways › order-3 Markov, log-probability › ci95	0.4753 – 0.5549
512 › ways › order-3 Markov, log-probability › 95% interval, resampled by taxon	0.3788 – 0.6505
measured with	no model

twee-instrumenten-controles32

controls › evo 2 against 4-mer counting › evo2_auc	0.5396
controls › evo 2 against 4-mer counting › evo 2, ci95 on rows	0.4699 – 0.6042
controls › evo 2 against 4-mer counting › evo 2, ci95 on taxa, estimated	0.3788 – 0.6953
controls › evo 2 against 4-mer counting › rule-of-thumb factor	2.357
controls › evo 2 against 4-mer counting › 4-mer counting, as it stands live	0.7764
controls › evo 2 against 4-mer counting › 4-mer counting live, n per side	400
controls › evo 2 against 4-mer counting › 4-mer counting, same sample as evo 2	0.7426
controls › evo 2 against 4-mer counting › 4-mer counting, ci95 on rows	0.6807 – 0.7971
controls › evo 2 against 4-mer counting › 4-mer counting, ci95 on taxa	0.639 – 0.8521
controls › evo 2 against 4-mer counting › n per side	150
controls › evo 2 against 4-mer counting › taxa in this measurement	27
controls › evo 2 against 4-mer counting › intervals overlap on taxa	yes
controls › evo 2 against 4-mer counting › margin	-0.0563
controls › competitors at 300 bases › ours	0.7528
controls › competitors at 300 bases › ours, ci95 on rows	0.722 – 0.7828
controls › competitors at 300 bases › ours, ci95 on taxa, estimated	0.6481 – 0.8567
controls › competitors at 300 bases › competitors › grover › auc	0.479
controls › competitors at 300 bases › competitors › grover › ci95 on rows	0.4361 – 0.5143
controls › competitors at 300 bases › competitors › grover › ci95 on taxa, estimated	0.3411 – 0.6093
controls › competitors at 300 bases › competitors › grover › holds on taxa	yes
controls › competitors at 300 bases › competitors › grover › margin	0.0388
controls › competitors at 300 bases › competitors › nt-500m › auc	0.4971
controls › competitors at 300 bases › competitors › nt-500m › ci95 on rows	0.4614 – 0.5366
controls › competitors at 300 bases › competitors › nt-500m › ci95 on taxa, estimated	0.37 – 0.628
controls › competitors at 300 bases › competitors › nt-500m › holds on taxa	yes
controls › competitors at 300 bases › competitors › nt-500m › margin	0.0202
controls › competitors at 300 bases › competitors › dnabert-6 › auc	0.552
controls › competitors at 300 bases › competitors › dnabert-6 › ci95 on rows	0.5117 – 0.5914
controls › competitors at 300 bases › competitors › dnabert-6 › ci95 on taxa, estimated	0.4149 – 0.6882
controls › competitors at 300 bases › competitors › dnabert-6 › holds on taxa	no
controls › competitors at 300 bases › competitors › dnabert-6 › margin	-0.0401
controls › competitors at 300 bases › smallest margin	-0.0401

zeecurve-besmet18

length	512
baseline › 4-mer counting	0.8227
baseline › chance	0.5
baseline › at length	512
points[0] › million bases	50
points[0] › actual million	56.5
points[0] › train_loss	0.8002
points[0] › auc	0.5668
points[0] › ci95	0.5318 – 0.6035
points[0] › on	2026-09-22 00:41
points[1] › million bases	100
points[1] › actual million	107.3
points[1] › train_loss	0.8025
points[1] › auc	0.5765
points[1] › ci95	0.5399 – 0.6147
points[1] › on	2026-09-22 02:28
direction	from 50M to 100M bases: AUC 0.5668 -> 0.5765 (+0.0097)
withdrawn	yes, the 40 held-out taxa were in the training corpus

zeecurve30

length	512
baseline › 4-mer counting	0.8227
baseline › chance	0.5
baseline › at length	512
points[0] › million bases	50
points[0] › actual million	55.7
points[0] › train_loss	0.8305
points[0] › auc	0.6796
points[0] › ci95	0.6455 – 0.7161
points[0] › on	2026-09-22 05:25
points[1] › million bases	100
points[1] › actual million	105.7
points[1] › train_loss	0.8127
points[1] › auc	0.6977
points[1] › ci95	0.6636 – 0.7328
points[1] › on	2026-09-22 06:56
points[2] › million bases	250
points[2] › actual million	254
points[2] › train_loss	0.803
points[2] › auc	0.6989
points[2] › ci95	0.6613 – 0.734
points[2] › on	2026-09-22 11:26
points[3] › million bases	500
points[3] › actual million	500.5
points[3] › train_loss	0.7599
points[3] › auc	0.7063
points[3] › ci95	0.6705 – 0.7401
points[3] › on	2026-09-22 20:59
direction	from 50M to 500M bases: AUC 0.6796 -> 0.7063 (+0.0267)
measured with	EH14 X, 6.7 M (previous version)

zeemeting-besmet14

model	EH14 X, 6.7 M (previous version)
bases read	1,638,400
train_loss	1.3499
known	120
novel	120
held-out taxa	40
window	512
auc	0.5208
ci95	0.4463 – 0.5866
baseline › 4-mer counting	0.8227
baseline › chance	0.5
baseline › at length	512
minutes	0.1
withdrawn	yes, the 40 held-out taxa were in the training corpus

zeroshot11

variants	2,000
pathogenic	252
held_out_genes	505
zero_shot › model	biox-dna-0
zero_shot › auc	0.4618
zero_shot › ci95	0.4281 – 0.5007
zero_shot › labels_seen	0
supervised › model	atlas-1
supervised › auc	0.8448
supervised › ci95	0.8156 – 0.8735
supervised › labels_seen	1,370,524

zeroshot_curve23

what	zero-shot variant AUC against bases read, biox-dna-0
reference › atlas-1 (supervised)	0.8448
reference › guessing	0.5
points[0] › million_bases	90.1
points[0] › train_loss	1.1585
points[0] › zero_shot_auc	0.4589
points[0] › ci95	0.424 – 0.4977
points[0] › variants	2,000
points[0] › at	2026-09-20 16:20
points[1] › million_bases	100
points[1] › train_loss	1.1533
points[1] › zero_shot_auc	0.4668
points[1] › ci95	0.4272 – 0.5073
points[1] › variants	2,000
points[1] › at	2026-09-20 17:15
points[1] › actual_million_bases	106.5
points[2] › million_bases	150
points[2] › train_loss	1.1408
points[2] › zero_shot_auc	0.4618
points[2] › ci95	0.4281 – 0.5007
points[2] › variants	2,000
points[2] › at	2026-09-20 18:59
points[2] › actual_million_bases	155.6