Lal Claude Opus 4.6 commited on
Commit
a87b36e
·
1 Parent(s): 6d765d9

Add split distributions and fix loading code

Browse files

- Add human split distribution (train/val/test counts)
- Add mouse split distribution (train/val/test counts)
- Fix repo_type="dataset" missing from hf_hub_download
- Update publication reference to Linder et al. 2025

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Files changed (1) hide show
  1. README.md +30 -6
README.md CHANGED
@@ -13,10 +13,10 @@ size_categories:
13
  # borzoi-data
14
 
15
  ## Dataset Summary
16
- This dataset contains the specific genomic intervals used for training, validating, and testing the Borzoi model, a deep learning architecture for predicting functional genomic tracks from DNA sequence. The intervals are provided for both human and mouse genomes. We modified the intervals provided in the original source by extending the input sequence to 524,288bp to create the full interval that was supplied to the model.
17
 
18
- - **Publication:** [Avsec, Ž., et al. "Effective gene expression prediction from sequence by integrating long-range interactions." Nat Methods 18, 1196–1203 (2021).](https://www.nature.com/articles/s41592-021-01252-x)
19
- - **Original Source** https://github.com/calico/borzoi/tree/main/data
20
  - **Genome Builds:**
21
  - Human: hg38
22
  - Mouse: mm10
@@ -49,15 +49,39 @@ Both files follow a standard genomic interval format:
49
  | `human_intervals.tsv` | 55,497 | hg38 |
50
  | `mouse_intervals.tsv` | 49,369 | mm10 |
51
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
52
  ## Usage
53
 
54
  ```python
55
  from huggingface_hub import hf_hub_download
56
  import pandas as pd
57
 
58
- file_path = hf_hub_download(repo_id="Genentech/borzoi-data", filename="human_intervals.tsv")
 
 
 
 
59
  df_human = pd.read_csv(file_path, sep='\t')
60
 
61
- file_path = hf_hub_download(repo_id="Genentech/borzoi-data", filename="mouse_intervals.tsv")
 
 
 
 
62
  df_mouse = pd.read_csv(file_path, sep='\t')
63
- ```
 
13
  # borzoi-data
14
 
15
  ## Dataset Summary
16
+ This dataset contains the specific genomic intervals used for training, validating, and testing the Borzoi model, a deep learning architecture for predicting functional genomic tracks from DNA sequence. The intervals are provided for both human and mouse genomes. We modified the intervals provided in the original source by extending the input sequence to 524,288 bp to create the full interval that was supplied to the model.
17
 
18
+ - **Publication:** [Linder, J. et al. Nature Genetics (2025)](https://www.nature.com/articles/s41588-024-02053-6)
19
+ - **Original Source:** https://github.com/calico/borzoi/tree/main/data
20
  - **Genome Builds:**
21
  - Human: hg38
22
  - Mouse: mm10
 
49
  | `human_intervals.tsv` | 55,497 | hg38 |
50
  | `mouse_intervals.tsv` | 49,369 | mm10 |
51
 
52
+ ### Data Splits
53
+
54
+ **Human (hg38)**
55
+ | Split | Count |
56
+ |-------|-------|
57
+ | train | 41,699 |
58
+ | val | 6,910 |
59
+ | test | 6,888 |
60
+
61
+ **Mouse (mm10)**
62
+ | Split | Count |
63
+ |-------|-------|
64
+ | train | 36,950 |
65
+ | val | 6,318 |
66
+ | test | 6,101 |
67
+
68
  ## Usage
69
 
70
  ```python
71
  from huggingface_hub import hf_hub_download
72
  import pandas as pd
73
 
74
+ file_path = hf_hub_download(
75
+ repo_id="Genentech/borzoi-data",
76
+ filename="human_intervals.tsv",
77
+ repo_type="dataset"
78
+ )
79
  df_human = pd.read_csv(file_path, sep='\t')
80
 
81
+ file_path = hf_hub_download(
82
+ repo_id="Genentech/borzoi-data",
83
+ filename="mouse_intervals.tsv",
84
+ repo_type="dataset"
85
+ )
86
  df_mouse = pd.read_csv(file_path, sep='\t')
87
+ ```