InfoHiC

InfoHiC is a method to predict 3D genome folding from complex DNA rearrangements, which enables the training and validation usage of cancer Hi-C data. InfoHiC provides a decomposed level of interaction views of multiple contigs from the cancer Hi-C matrix. InfoHiC is composed of convolutional neural network (CNNs) that outputs chromatin interactions of genomic contigs and merges them in the total Hi-C views, based on a contig-specific copy number (CSCN) encoding that represent genomic variants in the contig matrix.

Cancer Hi-C prediction models

Brain
- NPC 1Mb CSCN decoding 40kb
- NPC 2Mb CSCN decoding 40kb
Breast
- T47D 1Mb CSCN encoding 40kb
- T47D 2Mb CSCN encoding 40kb
Hematopoietic
- K562 1Mb CSCN encoding 10kb

Snakemake install

The InfoHiC workflow is run by the snakemake.
Environments are described in workflow/envs.
Rules are described in workflow/Snakefile.

wget https://github.com/conda-forge/miniforge/releases/download/24.1.2-0/Miniforge3-Linux-x86_64.sh
bash Miniforge3-Linux-x86_64.sh
mamba create -c conda-forge -c bioconda -n snakemake snakemake
conda config --set channel_priority strict
conda activate snakemake

InfoHiC repository

git clone https://github.com/DMCB-GIST/InfoHiC.git
InfoHiC_repo=${PWD}/InfoHiC
cd ${InfoHiC_repo}

After git clone, follow the steps below.

Conda environment setting
Dataset download
InfoHiC prediction

Conda environment setting

snakemake --core all --use-conda hic_mapping_env
snakemake --core all --use-conda InfoHiC_env

Dataset download

snakemake --cores all --use-conda InfoHiC_download

InfoHiC prediction

Hi-C matrices are predicted using the InfoHiC model.

Starting from WGS fastq

If you already have InfoGenomeR output, go to Starting from InfoGenomeR output
Else, test with low coverage WGS fastqs we provide first to obtain InfoGenomeR output. Then replace fastqs with yours.
Details are here https://github.com/dmcblab/InfoGenomeR

Inputs

The inputs should be structured like below

InfoHiC trained model

model
├── contig_model.data-00000-of-00001
├── contig_model.index
└── contig_model.meta

WGS fastqs
- normal fastq files are optional for somatic mode

fastq
├── normal1.fq.gz
├── normal2.fq.gz
├── tumor1.fq.gz
└── tumor2.fq.gz

workflow

InfoGenomeR workspace setting

cd $HOME

#InfoGenomeR repository
git clone https://github.com/dmcblab/InfoGenomeR.git
InfoGenomeR_repo=${PWD}/InfoGenomeR
cd ${InfoGenomeR_repo}

#environment setting
snakemake --core all --use-conda InfoGenomeR_env

#dataset download
snakemake --cores all --use-conda InfoGenomeR_download

#example dataset download
snakemake --core all --use-conda InfoGenomeR_example_download

# make a workspace directory
cd ${InfoGenomeR_repo}
workspace_dir=InfoGenomeR_workspace1
mkdir -p ${workspace_dir}

# link the reference in the workspace directory
ln -s ${PWD}/humandb/ref ${workspace_dir}/ref

# use low coverage examples data
ln -s ${PWD}/examples/fastq ${workspace_dir}/fastq

InfoGenomeR run

# Run the InfoGenomeR workflow. Select either somatic or total mode.
snakemake --core all --use-conda ${workspace_dir}/InfoGenomeR_output --config mode=somatic min_ploidy=2.5 max_ploidy=3.5

# This is InfoGenomeR output
InfoGenomeR_output=`readlink -f ${workspace_dir}/InfoGenomeR_output`

InfoHiC workspace setting

# go to the InfoHiC repo directory
cd ${InfoHiC_repo}

# make a workspace directory
workspace_dir=InfoHiC_workspace1
mkdir -p ${workspace_dir}

# Link the reference and the InfoHiC model in the workspace
ln -s ${PWD}/humandb/ref ${workspace_dir}/ref
ln -s ${PWD}/models/breast_model ${workspace_dir}/model

# take the InfoGenomeR output 
cp -r ${InfoGenomeR_output} ${workspace_dir}/InfoGenomeR_output

InfoHiC run

# set wildcards 
resolution=40000
window=2000000
cancer_type=BRCA
model=CSCN_encoding
gpu=1 # use the available gpu

# run the InfoHiC workflow
snakemake --cores all --use-conda ${workspace_dir}/InfoHiC_prediction/hic_${resolution}.window_${window}.${cancer_type}.${model}.gpu${gpu}

# run the post process
snakemake --cores all --use-conda ${workspace_dir}/InfoHiC_prediction/hic_${resolution}.window_${window}.${cancer_type}.${model}.gpu${gpu}.post_process

Outputs

InfoHiC_workspace1/InfoHiC_prediction/hic_40000.window_2000000.BRCA.CSCN_encoding.gpu1.post_process
└── euler.8.14
    ├── contig1.30385446.43357567
    │   ├── contigs_ref_coor.tsv
    │   ├── contigs.tsv
    │   ├── ens_gene.bed
    │   ├── hic_iced_loop.bed
    │   ├── hic_iced.matrix
    │   ├── hic_iced.pdf
    │   ├── hic_iced_tad.bed
    │   ├── hic.matrix
    │   ├── hic.pdf
    │   ├── hic_tad.bed
    │   ├── super_enhancer.bed
    │   └── typical_enhancer.bed
    ├── contig2.58095296.66230780
    │   ├── contigs_ref_coor.tsv
    │   ├── contigs.tsv
    │   ├── ens_gene.bed
    │   ├── hic_iced_loop.bed
    │   ├── hic_iced.matrix
    │   ├── hic_iced.pdf
    │   ├── hic_iced_tad.bed
    │   ├── hic.matrix
    │   ├── hic.pdf
    │   ├── hic_tad.bed
    │   ├── super_enhancer.bed
    │   └── typical_enhancer.bed
    ......

the post_process output folder contains chromosome sets where InfoHiC performed SV Hi-C prediction.
- euler.8.14 indicates chromosomes 8 and 14
Each euler directory contains SV contigs
- contig2.58095296.66230780 indicates contig index, contig coordinate start, and contig coordinate end in a prediction window.
Each contig folder contains predicted Hi-C information
- contigs.tsv
  - full contig information of the contig
  - contig index, breakpoint graph node index1, breakpoint graph node index2, ref chrom, ref pos, ref end, cumulative length
- contigs_ref_coor.tsv
  - the current SV prediction window in the contig
  - ref chrom, ref pos, ref end, ref orientation, contig index, contig pos, contig end, contig orientation
- ens_gene.bed
  - Ensembl gene location in the config
- super_enhancer.bed
  - SE location in the config
- typical_enhancer.bed
  - TE location in the contig
- hic_iced_loop.bed
  - neo-loop location in the iced contig
- hic_iced.matrix
  - prediction results with post iced normalization
- hic_iced_tad.bed
  - noe-TAD location in the iced contig. The forth column indicates a TAD level from spectralTAD.
- hic_iced.pdf
  - Visualization of contig Hi-C matrix, with neo-TAD, neo-loop, ens_gene, super_enhancer, and typical_enhancer annotations

Starting from InfoGenomeR output

Inputs

InfoHiC trained model
InfoGenomeR output from WGS (https://github.com/dmcblab/InfoGenomeR)

Workflow

# go to the InfoHiC base directory
cd InfoHiC

# make a workspace directory
workspace_dir=InfoHiC_workspace2
mkdir -p ${workspace_dir}

# Link the reference and the InfoHiC model in the root directory
ln -s ${PWD}/humandb/ref ${workspace_dir}/ref
ln -s ${PWD}/models/breast_model ${workspace_dir}/model

# take the InfoGenomeR output example
cp -r examples/T47D_chromosomes_8_14/InfoGenomeR_output ${workspace_dir}/InfoGenomeR_output

# set wildcards 
resolution=40000
window=2000000
cancer_type=BRCA
model=CSCN_encoding
gpu=1 # use the available gpu

# run the InfoHiC workflow
snakemake --cores all --use-conda ${workspace_dir}/InfoHiC_prediction/hic_${resolution}.window_${window}.${cancer_type}.${model}.gpu${gpu}

# run the post process
snakemake --cores all --use-conda ${workspace_dir}/InfoHiC_prediction/hic_${resolution}.window_${window}.${cancer_type}.${model}.gpu${gpu}.post_process

InfoHiC training

The training step can be skipped because we provide InfoHiC trained models.
It takes two weeks to train on 1 Mb windows (~10 epoches).
On 2 Mb windows, it takes a month.
Take the one-day example on 1 Mb windows using the subset of T47D Hi-C reads.

Starting from InfoGenomeR output

Inputs

InfoGenomeR output from WGS (https://github.com/dmcblab/InfoGenomeR)
Hi-C reads

Workflow

# go to the InfoHiC base directory
cd InfoHiC

# make the root output directory
root_dir=InfoHiC_training_output
mkdir -p ${root_dir}

# link the reference and the seed file in the root directory
ln -s ${PWD}/humandb/ref ${root_dir}/ref
ln -s ${PWD}/models/seed_file ${root_dir}/seed_file

# Place Hi-C reads and the InfoGenomeR output in the root directory
cp -r ${PWD}/examples/T47D_hic_reads_subset ${root_dir}/hic_reads
cp -r examples/T47D_chromosomes_8_14/InfoGenomeR_output ${root_dir}/InfoGenomeR_output

# set wildcards
resolution=40000

# Run Hi-C read mapping
snakemake --cores all --use-conda ${root_dir}/hic_${resolution}/3div


# set wildcards
window=1040000
split_idx=1
split_rate=0.1 # 0.1 for valid and test, and 0.8 for train

# Run Hi-C data post process and split
snakemake --cores all --use-conda ${root_dir}/hic_${resolution}/window_${window}.split${split_idx}_rate${split_rate}

# set wildcards
cancer_type=BRCA
mode=CSCN_encoding
gpu=1
epoch=1

# InfoHiC training
snakemake --cores all --use-conda ${root_dir}/InfoHiC_training.${cancer_type}/hic_${resolution}.window_${window}.split${split_idx}_rate${split_rate}/${mode}/gpu${gpu}.epoch${epoch}

Then, go to Export an InfoHiC model

Export an InfoHiC model

After training, select a model and export it


# List the models, and select a best_checkpoint saved at the last step
model_dir=${root_dir}/InfoHiC_training.${cancer_type}/hic_${resolution}.window_${window}.split${split_idx}_rate${split_rate}/${mode}/gpu${gpu}.epoch${epoch}
ls -l ${model_dir}/training_run_data_40kb_ascn_ACGT_rc_shift 

# This is an example of best_checkpoint 
model=${model_dir}/training_run_data_40kb_ascn_ACGT_rc_shift/best_checkpoint-51198

# copy the model into the model_dir base with the best_checkpoint prefix
cp -r $model.data-00000-of-00001 ${model_dir}/best_checkpoint.data-00000-of-00001
cp -r $model.index ${model_dir}/best_checkpoint.index
cp -r $model.meta ${model_dir}/best_checkpoint.meta

# export it as a contig model finally
snakemake --cores all --use-conda ${model_dir}.model

outputs

InfoHiC_training_output/InfoHiC_training.BRCA/hic_40000.window_1040000.split1_rate0.1/CSCN_encoding/gpu1.epoch1.model
├── contig_model.data-00000-of-00001
├── contig_model.index
└── contig_model.meta

Make a link to the model output in a InfoHiC workspace when running InfoHiC.

Name		Name	Last commit message	Last commit date
Latest commit History 21 Commits
HiC_processing		HiC_processing
InfoGenomeR_processing		InfoGenomeR_processing
InfoHiC_tensorflow		InfoHiC_tensorflow
doc		doc
examples		examples
ext/3divv2-master		ext/3divv2-master
humandb		humandb
models		models
post_analysis		post_analysis
workflow		workflow
LICENSE		LICENSE
README.md		README.md

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

InfoHiC

Cancer Hi-C prediction models

Snakemake install

InfoHiC repository

Conda environment setting

Dataset download

InfoHiC prediction

Starting from WGS fastq

Inputs

workflow

InfoGenomeR workspace setting

InfoGenomeR run

InfoHiC workspace setting

InfoHiC run

Outputs

Starting from InfoGenomeR output

Inputs

Workflow

InfoHiC training

Starting from InfoGenomeR output

Inputs

Workflow

Export an InfoHiC model

outputs

About

Releases

Packages

Languages

License

xtmgah/InfoHiC

Folders and files

Latest commit

History

Repository files navigation

InfoHiC

Cancer Hi-C prediction models

Snakemake install

InfoHiC repository

Conda environment setting

Dataset download

InfoHiC prediction

Starting from WGS fastq

Inputs

workflow

InfoGenomeR workspace setting

InfoGenomeR run

InfoHiC workspace setting

InfoHiC run

Outputs

Starting from InfoGenomeR output

Inputs

Workflow

InfoHiC training

Starting from InfoGenomeR output

Inputs

Workflow

Export an InfoHiC model

outputs

About

Resources

License

Stars

Watchers

Forks

Releases

Packages 0

Languages

Packages