AutoHiC

AutoHiC is a deep learning tool that uses Hi-C data to support genome assembly. It can automatically correct errors during genome assembly and generate chromosome-level genome.

Author: Zijie Jiang

Email: jzjlab@163.com

Content

• AutoHiC
  • Content
  • Notes
  • Overview of AutoHiC
  • Citations
  • Installation
  • conda
  • Docker
  • Singularity
  • Pre-trained model download
  • Usages
  • Data Preparation
  • Configs
  • Run
  • Results
  • Example
  • data
  • run
  • result
  • Plot HiC interaction map
  • One Setp AutoHiC (optional)
  • example
  • Split chromosome (optional)
  • License

Notes

• Currently AutoHiC has integrated 3d-dna into the complete process. If you are using YaHS, SALSA, Pin_hic etc, please read this document: Other tools

• AutoHiC updates very fast. If you have already cloned AutoHiC, please delete the AutoHiC folder and clone it again.

  ​

Overview of AutoHiC

​

Citations

If you used AutoHiC in your research, please cite us:

Zijie Jiang, Zhixiang Peng, Zhaoyuan Wei, Jiahe Sun, Yongjiang Luo, Lingzi Bie, Guoqing Zhang, Yi Wang, A deep learning-based method enables the automatic and accurate assembly of chromosome-level genomes, Nucleic Acids Research, 2024;, gkae789, https://doi.org/10.1093/nar/gkae789

Installation

conda

# clone AutoHiC
git clone https://github.com/Jwindler/AutoHiC.git

# cd AutoHiC
cd AutoHiC

# create AutoHiC env
conda env create -f autohic.yaml

# activate AutoHiC
conda activate autohic

# configuration environment
cd ./src/models/swin

# install dependencies
pip install -e . -i https://pypi.tuna.tsinghua.edu.cn/simple/

Note:

1. If src/straw.cpp:34:10: fatal error: curl/curl.h: No such file or directory is encountered during installation, enter the following command sudo apt-get install libcurl-dev libcurl4-openssl-dev libssl-dev in the terminal or refer: https://stackoverflow.com/questions/11471690/curl-h-no-such-file-or-directory.

2. Either GPU or CPU can be installed according to the above steps, and the program will automatically identify the running configuration and environment.

3. If you want to use GPU, please install CUDA-11.3 and cuDNN-8.2 before.

​

Docker

# pull images
sudo docker pull jwindler/autohic:main

# start container
sudo docker run -it -v $(pwd):/home/autohic jwindler/autohic:main bash

# You need to use mounts (-v) to exchange files between the host filesystem on which your user can write and the container filesystem. ( Default "./" )

# clone AutoHiC
git clone https://github.com/Jwindler/AutoHiC.git

# cd AutoHiC
cd AutoHiC

# activate AutoHiC
conda activate autohic

# configuration environment
cd ./src/models/swin

# install dependencies
pip install -e . -i https://pypi.tuna.tsinghua.edu.cn/simple/

Singularity

Considering that many users run AutoHiC on HPC, the build dependency environment may not be very free, and Docker has root restrictions, we provide a singularity version. Detailed documentation: doc

Pre-trained model download

Please select your most convenient download link below, You need to download error_model.pth, chr_model.pth , Juicer and 3d-dna for the configuration of subsequent configuration files

​

Usages

Data Preparation

• Contig level genome
• Hi-C reads
• directory structure (as below)

species_name/
├── rawdata
│   └── fastq
│       ├── SRR_X_R1.fastq.gz
│       ├── SRR_X_R2.fastq.gz
└── references
    └── contig.fasta

Notes:

1. The directory structure must be consistent with the above image.
2. Paired-end sequences must end with X_R1.fastq.gz and X_R2.fastq.gz ( also supports uncompressed formats such as : X_R1.fastq and X_R2.fastq )

​

Configs

Copy and edit the configuration file cft-autohic.txt in your local folder.

cft-autohic.txt example files are available in the AutoHiC directory.

To ensure run properly, please refrain from adding any additional lines to cft-autohic.txt .

• Setting the configuration file

Notes:

1. PRETRAINED_MODEL and CHR_PRETRAINED_MODEL parameters come from the download path of your pre-trained model before and after
2. JUICER_DIR and TD_DNA_DIR parameters come from the path you downloaded and decompressed respectively (If you have already installed it in advance, you can configure it directly)

​

Run

# cd AutoHiC directory
# Please modify according to your installation directory
cd /home/AutoHiC  

# run 
nohup python3.9 autohic.py -c cfg-autohic.txt > log.txt 2>&1 &

# nohup: Run the program ignoring pending signals

Notes:

1. Please specify the absolute path of the cft-autohic.txt
2. It is recommended to specify a directory for the log.txt, It will record the running information of AutoHiC
3. Delete the nohup command if you don't want the program to run in the background.
4. If you modify the configuration file and re-run AutoHiC, you must manually delete the previously generated result file.
5. If a warning (like the image below) appears in the log while you are using it, this is normal and the program is running normally. You just have to wait for the results.

​

Results

After the AutoHiC operation is completed, the following results will be obtained.

species_name/
├── AutoHiC
│   ├── autohic_results
│   │   ├── 0
│   │   ├── 1
│   │   ├── 2
│   │   ├── 3
│   │   ├── 4
│   │   └── chromosome
│   ├── data
│   │   ├── reference
│   │   └── restriction_sites
│   ├── hic_results
│   │   ├── 3d-dna
│   │   └── juicer
│   ├── logs
│   │   ├── 3d-dna.log
│   │   ├── 3_epoch.log
│   │   ├── 4_epoch.log
│   │   ├── bwa_index.log
│   │   ├── chromosome_epoch.log
│   │   └── juicer.log
│   ├── quast_output
│   │   ├── chromosome
│   │   └── contig
│   ├── chromosome_autohic.fasta 
│   └── result.html
├── cfg-autohic.txt

The main output:

1. fasta file with a "_autohic" suffix containing the output scaffolds at the chromosome level. If the size of genome_autohic.fasta differs significantly from the genome, it is recommended to use genome.FINAL.fasta (Path:/path/AutoHiC/autohic_results/chromosome) for optimal results. If the genome scaffolding effect is poor during chromosome assignment, the model may have errors in identifying chromosomes, which can lead to issues. Other scaffolding software can be used, followed by the use of AutoHiC to correct and assign the chromosomes. Please refer to Other tools.

2. The result.html file, which provides detailed information before and after genome correction, where the error occurred, and a heat map of HiC interaction and chromosome length before and after.

3. Please see this document for detailed results description.

Example

If you want to run AutoHiC with sample data, you can choose from the following data.

data

Please follow the link provided for the selected species to download the appropriate data and organize it into the required format, can refer to : Data Preparation.

• Reference genome : Sample genome files are available at the example_genome file in the pre-trained model download link : Pre-trained model download
• The default enzyme used for example data is DpnII

run

cd AutoHiC

nohup python3.9 autohic.py -c cfg-autohic.txt > log.txt 2>&1 &

• Please modify the cfg-autohic.txt file according to the actual situation, can refer to : Configs.

result

The main results of AutoHiC are genome and assembly reports at the chromosome level. For a detailed description of the results, please refer to Results. At the same time, we also upload the assembly report to Google Drive for users to retrieve and view.

​

Plot HiC interaction map

AutoHiC also provides a script to visualise the HiC interaction matrix separately.

python3.9 visualizer.py -hic example.hic

For detailed commands, please refer to the help documentation (--help)

• result

One Setp AutoHiC (optional)

If you have already run Juicer and 3d-dna, you can use the following extended script to use AutoHiC to help you detect HiC assembly errors and generate adjusted assembly files.

# Enter the AutoHiC directory.
cd /home/ubuntu/AutoHic  

# run onehic
python3.9 onehic.py -hic test.hic -asy test.assembly -autohic /home/ubuntu/AutoHic -p pretrained.pth -out ./

# run 3d-dna to get fasta
bash run-asm-pipeline-post-review.sh -r adjusted.assembly genome.fasta merged_nodups.txt 

# Please specify the absolute path of each file
# adjusted.assembly is output from onehic.py
# merged_nodups.txt is output from Juicer

Notes:

1. .hic and .assembly : can be obtained from 3d-dna results
2. -autohic : the parameter represents the path of AutoHiC
3. -p: the path to the error pretrained model you downloaded before

example

If you want to run onehic.py with example data, please get the corresponding data from the previously linked Pre-trained model download example_onehic file.

​

Split chromosome (optional)

If your genome is very complex, the model may not be very accurate in assigning the chromosomes. It is recommended that you import the last adjustment file into Juicxbox to manually split chromosomes.

The .hic and .assembly files you need to use can be obtained from the chromosome folder under the autohic_results directory.

​

License

AutoHiC Copyright (c) 2022 Wang lab. All rights reserved.

This software is distributed under the MIT License (MIT).