Files changed (1) hide show
  1. README.md +84 -3
README.md CHANGED
@@ -1,3 +1,84 @@
1
- ---
2
- license: apache-2.0
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ pretty_name: OpenGenome2
4
+ tags:
5
+ - huggingscience
6
+ - science
7
+ - biology
8
+ - genomics
9
+ - dna
10
+ - sequence-modeling
11
+ - long-context
12
+ - metagenomics
13
+ - microbiology
14
+ - bacteriophage
15
+ - plasmid
16
+ - mrna
17
+ - promoter
18
+ - ncrna
19
+ - organelle-genomes
20
+ - gtdb
21
+ - imgvr
22
+ - imgpr
23
+ - ncbi
24
+ - json
25
+ - fasta
26
+ ---
27
+
28
+ # OpenGenome2
29
+
30
+ OpenGenome2 is a large-scale, diverse DNA corpus spanning prokaryotes (bacteria & archaea), eukaryotes, organelles, bacteriophages, plasmids, and metagenomes. It was assembled to train long-context genomic foundation models (e.g., Evo 2) and emphasizes **functionally informative regions** (e.g., genic windows, promoters, noncoding RNAs) while providing comprehensive whole-genome coverage across the tree of life.
31
+
32
+ * **License:** Apache-2.0
33
+ * **Scope:** trillions of nucleotides aggregated across multiple public sources (e.g., GTDB v220, IMG/VR, IMG/PR, NCBI eukaryotic references) with curation and quality filtering to support genome-scale modeling.
34
+ * **Formats:** JSONL (token-ready sequences) and FASTA (raw sequences).
35
+ * **Splits:** `train`, `validation`, `test` are provided for most JSONL subsets.
36
+
37
+ ## Contents & Source Mapping
38
+
39
+ The on-hub layout mirrors data provenance and training phase:
40
+
41
+ * **`json/pretraining_or_both_phases/`**
42
+ Functional and broad-coverage sources used during pretraining (or used in both pre- and mid-training):
43
+
44
+ * `eukaryotic_genic_windows/` – windows enriched for coding/genic content
45
+ * `gtdb_v220_imgpr/` – prokaryotes from **GTDB v220** plus curated plasmids/phages from **IMG/PR**
46
+ * `imgvr_untagged/` – bacteriophage sequences from **IMG/VR**
47
+ * `metagenomes/` – metagenomic contigs
48
+ * `mrna/`, `mrna_splice_promoter/` – mRNA and splicing/promoter-focused windows
49
+ * `ncrna/` – noncoding RNAs
50
+ * `organelle/` – organellar genomes
51
+ * `promoters/` – promoter regions
52
+ * **`json/midtraining_specific/`**
53
+ Additional curated data emphasized during mid-training:
54
+
55
+ * `gtdb_v220_stitched/` – stitched windows / references from **GTDB v220**
56
+ * `imgpr/` – curated plasmids & phages from **IMG/PR**
57
+ * `ncbi_eukaryotic_genomes/` – curated eukaryotic references from **NCBI**
58
+ * **`fasta/`** mirrors these sources for users preferring raw FASTA (e.g., `fasta/gtdb_v220`, `fasta/mrna`, `fasta/organelles`, `fasta/promoters`, etc.).
59
+
60
+ ## Recommended Uses
61
+
62
+ * Pretraining/continued-pretraining **DNA language models** (autoregressive or masked).
63
+ * Long-context modeling (100 kbp–1 Mbp) for **regulatory architecture** and genome context tasks.
64
+ * Zero-shot **variant effect** predictions on promoters/genic windows.
65
+ * Embedding-based **functional clustering** across genomes, plasmids, and phages.
66
+ * Generative modeling for **protein–DNA/RNA system design** when coupled with appropriate safety review.
67
+
68
+ ## Citation
69
+
70
+ If you use OpenGenome2 or models trained on it, please cite:
71
+
72
+ * **Evo 2** (trained on OpenGenome2):
73
+ Brixi *et al.*, “Genome modeling and design across all domains of life with Evo 2,” *bioRxiv* (2025).
74
+ [https://doi.org/10.1101/2025.02.18.638918](https://doi.org/10.1101/2025.02.18.638918)
75
+
76
+ * **Data sources**:
77
+
78
+ * GTDB updated references (v220)
79
+ * IMG/VR (bacteriophages) and IMG/PR (plasmids/phages)
80
+ * NCBI eukaryotic genomes/references
81
+
82
+ ## Attribution
83
+
84
+ * Dataset maintained by the **Arc Institute**. See the Hub repo for updates and discussions.