Add sliding-window and allelic-skew inference
Browse files
README.md
CHANGED
|
@@ -11,8 +11,6 @@ tags:
|
|
| 11 |
pipeline_tag: other
|
| 12 |
---
|
| 13 |
|
| 14 |
-
(windowing forthcoming once internal flip as opposed to total flip at inference time is confirmed as intended behavior)
|
| 15 |
-
|
| 16 |
# MPAC
|
| 17 |
|
| 18 |
MPAC (Malinois with Parallel Aggregated Cross-validation) predicts cis-regulatory
|
|
@@ -50,7 +48,17 @@ than a reverse complement of the whole 600 bp construct.
|
|
| 50 |
MPAC covers autosomes only; `from_pretrained` raises on chrX, chrY and anything else
|
| 51 |
with no held-out fold.
|
| 52 |
|
| 53 |
-
For
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 54 |
[john-c-butts/MPAC](https://github.com/john-c-butts/MPAC).
|
| 55 |
|
| 56 |
## Citation
|
|
|
|
| 11 |
pipeline_tag: other
|
| 12 |
---
|
| 13 |
|
|
|
|
|
|
|
| 14 |
# MPAC
|
| 15 |
|
| 16 |
MPAC (Malinois with Parallel Aggregated Cross-validation) predicts cis-regulatory
|
|
|
|
| 48 |
MPAC covers autosomes only; `from_pretrained` raises on chrX, chrY and anything else
|
| 49 |
with no held-out fold.
|
| 50 |
|
| 51 |
+
For allelic skew, pass matched reference and alternate contexts of 371 bp -- 180 bp
|
| 52 |
+
upstream of the variant, the variant, 190 bp downstream. Each is tiled into eighteen
|
| 53 |
+
200 bp windows at stride 10 and averaged, reproducing the scheme behind the
|
| 54 |
+
published predictions.
|
| 55 |
+
|
| 56 |
+
```python
|
| 57 |
+
out = ensemble.predict_skew(ref_contexts, alt_contexts, device="cuda")
|
| 58 |
+
out["skew"] # (n, 3), alt minus ref
|
| 59 |
+
```
|
| 60 |
+
|
| 61 |
+
Command-line tooling for VCF-scale runs lives at
|
| 62 |
[john-c-butts/MPAC](https://github.com/john-c-butts/MPAC).
|
| 63 |
|
| 64 |
## Citation
|