|
|
--- |
|
|
license: agpl-3.0 |
|
|
tags: |
|
|
- EpiATLAS |
|
|
- IHEC |
|
|
- epigenetics |
|
|
- EpiClass |
|
|
- pytorch |
|
|
--- |
|
|
# Epigenomic Classifer - Assay |
|
|
|
|
|
Epigenome Assay/Target classifier trained on the [EpiATLAS dataset](https://ihec-epigenomes.org/epiatlas/data/). The classes correspond to the minimal elements of the [IHEC reference epigenome](https://ihec-epigenomes.org/research/reference-epigenome-standards/index.html): |
|
|
- Seven (7) ChIP: assays H3k27ac, H3k27me3, H3k36me3, H3k4me1, H3k4me3, H3k9me3, input |
|
|
- RNA-Seq (two protocols, mrna/total) |
|
|
- WGBS (two protocols, pbat/standard) |
|
|
|
|
|
The model is a simple dense feedforward neural network, with one hidden layer of 3000 nodes. The model was trained using PyTorch Lightning. See Github repository [labjacquespe/EpiClass](https://github.com/labjacquespe/epiclass/blob/master/src/python/epiclass/core/model_pytorch.py) for model code. |
|
|
|
|
|
See the .o and .e files for training details. More information is also available on Comet ML, in the rabyj/epiclass project. The ID of this training run is [0f8e5eb996114868a17057bebe64f87c](https://www.comet.com/rabyj/epiclass/0f8e5eb996114868a17057bebe64f87c) |
|
|
|
|
|
For more context, see the associated publication: [Leveraging a large harmonized epigenomic data collection for metadata prediction to validate and augment over 350,000 public epigenomic datasets](https://www.biorxiv.org/content/10.1101/2025.09.04.670545v1) |
