lagosproject commited on
Commit
f95f6ea
·
verified ·
1 Parent(s): 77c4ae1

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +106 -0
README.md ADDED
@@ -0,0 +1,106 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-4.0
3
+ task_categories:
4
+ - other
5
+ tags:
6
+ - genomics
7
+ - dna
8
+ - embeddings
9
+ - alphagenome
10
+ - hg38
11
+ - grch38
12
+ pretty_name: ALPHAGenome hg38 Embeddings
13
+ size_categories:
14
+ - 10K<n<100K
15
+ ---
16
+
17
+ # ALPHAGenome hg38 Embeddings
18
+
19
+ Pre-computed **ALPHAGenome** DNA foundation model embeddings for the entire human genome (hg38 / GRCh38).
20
+
21
+ The human genome is divided into **~22,000 non-overlapping 131 KB bins**. Each bin's DNA sequence is embedded into a **3,072-dimensional latent space** using the ALPHAGenome foundation model.
22
+
23
+ ## Companion project
24
+
25
+ These embeddings power the **ALPHAGenome UMAP Explorer** — an interactive browser visualization of latent relationships between genomic regions:
26
+
27
+ **→ [GitHub: DNAEmbeddings](https://github.com/lagosproject/DNAEmbeddings)**
28
+ **→ [Live Demo](https://lagosproject.github.io/DNAEmbeddings/)**
29
+
30
+ Scripts to download this dataset and re-run the UMAP pipeline are included in that repo.
31
+
32
+ ## Dataset structure
33
+
34
+ ```
35
+ data/
36
+ chr1_embeddings.npy # float32 array, shape (N, 3072)
37
+ chr1_metadata.csv # columns: chrom, start, end
38
+ chr2_embeddings.npy
39
+ chr2_metadata.csv
40
+ ... # one pair per chromosome (chr1–22, chrX, chrY)
41
+ ```
42
+
43
+ Each `.npy` file and its paired `.csv` share the same row order — row `i` in the embeddings corresponds to row `i` in the metadata.
44
+
45
+ ## Quick usage
46
+
47
+ ```python
48
+ from huggingface_hub import hf_hub_download
49
+ import numpy as np
50
+ import pandas as pd
51
+
52
+ # Download a single chromosome
53
+ emb_path = hf_hub_download("lagosproject/ALPHAGenome-Embeddings", "data/chr1_embeddings.npy", repo_type="dataset")
54
+ meta_path = hf_hub_download("lagosproject/ALPHAGenome-Embeddings", "data/chr1_metadata.csv", repo_type="dataset")
55
+
56
+ embeddings = np.load(emb_path) # (1750, 3072)
57
+ metadata = pd.read_csv(meta_path) # chrom | start | end
58
+ ```
59
+
60
+ Or download everything at once (≈312 MB):
61
+
62
+ ```python
63
+ from huggingface_hub import snapshot_download
64
+ snapshot_download("lagosproject/ALPHAGenome-Embeddings", repo_type="dataset", local_dir="res/")
65
+ ```
66
+
67
+ ## Statistics
68
+
69
+ | Chromosome | Windows | Embedding shape |
70
+ |---|---|---|
71
+ | chr1 | 1,750 | (1750, 3072) |
72
+ | chr2 | 1,829 | (1829, 3072) |
73
+ | chr3 | 1,509 | (1509, 3072) |
74
+ | chr4 | 1,439 | (1439, 3072) |
75
+ | chr5 | 1,379 | (1379, 3072) |
76
+ | chr6 | 1,294 | (1294, 3072) |
77
+ | chr7 | 1,206 | (1206, 3072) |
78
+ | chr8 | 1,100 | (1100, 3072) |
79
+ | chr9 | 915 | (915, 3072) |
80
+ | chr10 | 1,013 | (1013, 3072) |
81
+ | chr11 | 1,022 | (1022, 3072) |
82
+ | chr12 | 1,012 | (1012, 3072) |
83
+ | chr13 | 743 | (743, 3072) |
84
+ | chr14 | 687 | (687, 3072) |
85
+ | chr15 | 640 | (640, 3072) |
86
+ | chr16 | 619 | (619, 3072) |
87
+ | chr17 | 627 | (627, 3072) |
88
+ | chr18 | 609 | (609, 3072) |
89
+ | chr19 | 444 | (444, 3072) |
90
+ | chr20 | 482 | (482, 3072) |
91
+ | chr21 | 288 | (288, 3072) |
92
+ | chr22 | 285 | (285, 3072) |
93
+ | chrX | 1,170 | (1170, 3072) |
94
+ | chrY | 191 | (191, 3072) |
95
+ | **Total** | **22,253** | **3,072-dim** |
96
+
97
+ ## Data sources
98
+
99
+ | Source | Usage |
100
+ |--------|-------|
101
+ | **ALPHAGenome** | DNA sequence embeddings (3,072-dim per 131 KB bin) |
102
+ | **UCSC hg38** | Reference genome coordinates |
103
+
104
+ ## License
105
+
106
+ CC BY 4.0 — free to use for research and educational purposes with attribution.