File size: 2,733 Bytes
c011f89
 
95d8a69
 
 
 
 
 
 
 
 
 
 
 
c011f89
95d8a69
c011f89
95d8a69
c011f89
95d8a69
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
---
license: other
pretty_name: IRG Repeat Coverage (ENCODE cCRE windows, v2.1 GC-matched tables)
task_categories:
- tabular-classification
tags:
- genomics
- regulatory-elements
- ENCODE
- RepeatMasker
- annotation
- hg38
size_categories:
- 100K<n<1M
---
# IRG Repeat Coverage

Per-window repeat-content annotation that accompanies [`PxALabs/encode-regulatory-elements`](https://huggingface.co/datasets/PxALabs/encode-regulatory-elements) (v2.1). For each 1,000 bp window of the three `*_gc_matched` tables it records the fraction of the window covered by UCSC RepeatMasker annotations. It exists so that models can be evaluated on test sets matched for repeat content, one of the composition shortcuts studied in the [Interpretable Regulatory Genomics](https://github.com/PxA-Labs/interpretable-regulatory-genomics) project.

## Contents
| file | description |
|:--|:--|
| `repeat_fractions.parquet` | 328,505 rows, one per window of the three `gc_matched` tables (T cell, K562, GM12878) |
| `MANIFEST.sha256` | SHA-256 of the data file |

Columns (fractions of the 1,000 bp window, in [0, 1]):
- `rep_frac`: covered by any RepeatMasker repeat
- `alu_frac`: Alu elements
- `l1_frac`: L1 elements
- `simple_frac`: simple repeats and low-complexity regions
- join keys: `id`, with `biosample` and `negative_strategy`, to the `*_gc_matched.parquet` tables

## Usage
```python
import pandas as pd
from datasets import load_dataset
rep = pd.read_parquet("repeat_fractions.parquet")
tab = load_dataset("PxALabs/encode-regulatory-elements", "gm12878_gc_matched")["train"].to_pandas()
tab = tab.merge(rep[rep.biosample == "gm12878"][["id", "rep_frac"]], on="id", how="left", validate="1:1")
```

## Provenance
- Source annotation: UCSC Genome Browser `hg38` `rmsk` track, https://hgdownload.soe.ucsc.edu/goldenPath/hg38/database/rmsk.txt.gz (Last-Modified 2022-10-19, 155,633,856 bytes).
- Computation: `tools/data/repeat_annotation.py` in the project repository. Coordinates are 0-based half-open, as in the tables.
- Data file SHA-256: `4c8ee141bca908beed14d4d4ee91ca6d243c7512822b47b6cf2abef842da580e`.
- Used by notebooks 08, 09, 11, 13, 14 and 15 of the project (repeat-matched test sets).

## Limitations
- Covers only the three `gc_matched` tables; the other negative strategies are not annotated.
- Annotation quality is that of RepeatMasker; the fractions are a content measure, not a functional label.
- Derived from a third-party annotation. Check UCSC's data terms before redistribution beyond research use; the licence field is `other` until the project owner sets one.

## Citation
Cite the project repository and the underlying RepeatMasker / UCSC resources. A manuscript is in preparation; its citation will be added here.