Introduction

Pipeline to predict HiC matrices from ATAC-seq fastq files using maxATAC and C.Origami

Setup

If you are working on Ultraviolet (aka BigPuprle) you need to setup an environment. Which means you need to install snakemake and mamba (so a conda within a conda).

module load anaconda3/gpu
conda create -n snakemake -c conda-forge -c bioconda mamba snakemake

To configure conda to work on Ultraviolet you can edit the ultraviolet.yaml that comes with this repo to match your liking and copy it to your .config directory as shown.

mkdir -p ~/.config/snakemake/ultraviolet/
cp ultraviolet.yaml ~/.config/snakemake/ultraviolet/config.yaml

Execute

To run the pipeline you need to:

1. create a samplesheet with information about your samples
2. setup a directory with with data to run C.Origami
3. edit the config/config.yaml file to specify the paths to the relevant directories.

Samplesheet

The samplesheet should have information about sample and replicate names and the path to the fastq files. See the config/sample_meta.csv that comes with this directory for an example.

Alternatively you can specify a column called Run with the SRR ids of the samples and the pipeline should automatically download them for you (NOT TESTED).

C.Origami Directory

The C.Origami directory should look like this:

<corigami_base>/
├── data
│   ├── <genome>
│   │   ├── centrotelo.bed
│   │   └── dna_sequence
│   │       ├── chr10.fa.gz
│   │       ├── chr11.fa.gz
│   │       ├── ...
│   │       ├── chrX.fa.gz
│   │       └── chrY.fa.gz
│   └── <genome>_tiles.bed
└── model_weights
    └── <corigami_model>.ckpt

Where corigami_base, genome and corigami_model are specified in config/config.yaml.

The <corigami_base>/data directory can be:

• Downloaded from here (you will need to untar it)
• If you work from within Ultraviolet, symlinked from here: /gpfs/data/tsirigoslab/home/jt3545/hic_prediction/C.Origami-release/corigami_data/

To get the model weights you need to:

• Train your own model and save the checkpoint. Ask Javier Rodriguez Hernaez for details.
• Download a pretrained hg38 model checkpoint created by Javier from here. If in Ultraviolet symlink/copy the following path: /gpfs/home/rodrij92/PROJECTS/SHARE/epoch=53-step=64260.ckpt

config.yaml

The main parameter you may need to specify are:

• genome: either hg38 or mm10
• sample_meta: path to samplesheet
• corigami_base: directory with C.Origami data
• corigami_model: name of checkpoint file under <corigami_base>/model_weights

Run

If you have specify everything correctly you can launch the pipeline by executing the following commands on Ultraviolet:

conda activate snakemake  # activate environment you created in Ultraviolet if you don't have snakemake
snakemake --profile ultraviolet

Pipeline Graph