Back

GenomeOcean: An Efficient Genome Foundation Model Trained on Large-Scale Metagenomic Assemblies

Zhou, Z.; Riley, R.; Kautsar, S.; Wu, W.; Egan, R.; Hofmeyr, S.; Goldhaber-Gordon, S.; Yu, M.; Ho, H.; Liu, F.; Chen, F.; Morgan-Kiss, R.; Shi, L.; Liu, H.; Wang, Z.

2025-02-05 bioinformatics
10.1101/2025.01.30.635558 bioRxiv
Show abstract

Functional genomic sequences occupy an infinitesimal fraction of the astronomically large DNA sequence space, implying that evolution explores a low-dimensional manifold shaped by universal biochemical and evolutionary constraints. To investigate this underexplored Genomic Manifold Hypothesis using scalable, data-driven evidence, we developed GenomeOcean, a 4-billion-parameter generative model trained on 219 TB of diverse, co-assembled global environmental metagenomic data using an optimized transformer. The model captures both universal phylogenetic syntax and protein-coding fidelity. Furthermore, we validate the embeddings intrinsic low dimensionality. Crucially, comparisons with the independently trained Evo 2 model demonstrate a strong linear correspondence between their embedding spaces and convergent generative behaviors. These results suggest that the genomic manifold represents a fundamental biological principle, offering a unified framework for understanding evolutionary constraints and guiding synthetic biology.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

1
Bioinformatics
◐ 1204 papers in training set
Top 2%
14.8%
Similar papers in this journal
2
Nature Methods
○ 385 papers in training set
Top 0.9%
11.7%
Similar papers in this journal
3
Nature Biotechnology
○ 172 papers in training set
Top 0.2%
9.6%
Similar papers in this journal
4
Nature Communications
○ 5641 papers in training set
Top 21%
7.7%
Similar papers in this journal
5
Genome Biology
○ 637 papers in training set
Top 2%
6.6%
Similar papers in this journal
50% of probability mass above
6
Cell Systems
○ 201 papers in training set
Top 0.5%
6.6%
7
Genome Research
● 468 papers in training set
Top 0.9%
6.1%
8
Nucleic Acids Research
◐ 1281 papers in training set
Top 5%
4.0%
9
Bioinformatics Advances
◐ 203 papers in training set
Top 2%
2.6%
10
Molecular Biology and Evolution
◐ 542 papers in training set
Top 2%
2.6%
11
PLOS Computational Biology
● 1863 papers in training set
Top 11%
2.6%
12
Cell
○ 431 papers in training set
Top 4%
2.3%
13
Nature Machine Intelligence
○ 70 papers in training set
Top 1%
2.1%
14
Proceedings of the National Academy of Sciences
● 2444 papers in training set
Top 28%
1.7%
15
Nature Genetics
○ 286 papers in training set
Top 3%
1.7%
16
Nature
○ 645 papers in training set
Top 7%
1.7%
17
Nature Computational Science
○ 55 papers in training set
Top 0.7%
1.6%
18
Briefings in Bioinformatics
◐ 354 papers in training set
Top 5%
1.3%
19
NAR Genomics and Bioinformatics
◐ 242 papers in training set
Top 3%
1.1%
20
Cell Genomics
○ 172 papers in training set
Top 4%
1.0%
21
eLife
● 5828 papers in training set
Top 68%
0.8%
22
Science
● 477 papers in training set
Top 9%
0.8%
23
PeerJ
◐ 308 papers in training set
Top 13%
0.6%
24
Science Advances
● 1243 papers in training set
Top 34%
0.6%

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.