We provide several example configurations to give you a chance to start quickly. All of the examples use our trained convolutional autoencoder models.
We provide three simple and three more advanced examples that you can start
by calling make <NAME>.
| Name | Dtypes (Num Tracks) | Window Size | Step Freq | Chromosomes |
|---|---|---|---|---|
| example-3kb | DNase | 3 kb | 2 | chr21 |
| example-12kb | DNase | 12 kb | 3 | chr20-21 |
| example-120kb | DNase | 120 kb | 6 | chr17-21 |
| encode-e11-5-limb | DNase, H3K27ac | 3 kb | 2 | chr12 |
| encode-e11-5-face-hindbrian | DNase | 3 kb | 2 | chr1 |
| roadmap-e116-gm12878 | DNase, H3K27ac, H3K4me1, H3K4me3 | 3 kb | 2 | chr20-21 |
Each example consists of a python script for downloading the approriate datasets
(download-<NAME>.py) and the accompanying track configuration (config-<NAME>.json).
Note: The demos only allow prepare data for the specified chromosomes. If you want to search for patterns outside those chromosomes please adjust the track configuration approriately.
We have trained 6 autoencoder on DNase-seq and histone mark ChIP-seq datasets. The DNase-seq autoencoders were trained on 120 datasets from ENCODE and the ChIP-seq autoencoders were trained on 343 datasets each targeting one of the following histone marks:
- H3K4me1
- H3K4me3
- H3K27ac
- H3K9ac
- H3K27me3
- H3K9me3
- H3K36me3
For each data type we trained 3 autoencoders on the following window sizes and binning.
| Dtype | Window Size | Binning |
|---|---|---|
| ChIP | 3 kb | 25 bp |
| ChIP | 12 kb | 100 bp |
| ChIP | 120 kb | 1000 bp |
| DNase | 3 kb | 25 bp |
| DNase | 12 kb | 100 bp |
| DNase | 120 kb | 1000 bp |
Download: https://zenodo.org/record/2609763
Please see our preprint for a detailed description of the training and parameter settings:
Lekschas et al., Peax: Interactive Visual Pattern Search in Sequential Data Using Unsupervised Deep Representation Learning, bioRxiv, 2019, doi: 10.1101/597518.