RNN_Neo

An RNN model framework, simulates the processing of neo-peptides in vivo from proteasome cleavage, TAP transport, pMHC binding to TCR activation.

📖 Install

Dependencies

RNN_Neo currently test on x86_64 on ubuntu 16.04 with 96-core cpu

Get the RNN-Neo Source

git clone https://github.com/LiJIAJUN-G/RNN_Neo.git

Create and activate an environment using conda

conda env create -f freeze.yml
conda activate rnn_neo

Download Reference genome

bash download_hg38.sh

🌵Make sure the ref directory is as follows:

ref
├── anno
│   ├── 1000G_phase1.snps.high_confidence.hg38.vcf.gz(1,844,006kb)
│   ├── 1000G_phase1.snps.high_confidence.hg38.vcf.gz.tbi(2,079kb)
│   ├── af-only-gnomad.hg38.vcf.gz(3,109,644kb)
│   ├── af-only-gnomad.hg38.vcf.gz.tbi(2,386kb)
│   ├── dbsnp_138.hg38.vcf.gz(1,524,307kb)
│   ├── dbsnp_138.hg38.vcf.gz.tbi(2,268kb)
│   ├── Mills_and_1000G_gold_standard.indels.hg38.vcf.gz(20,202kb)
│   ├── Mills_and_1000G_gold_standard.indels.hg38.vcf.gz.tbi(1,465kb)
│   ├── small_exac_common_3.hg38.vcf.gz(1,267kb)
│   └── small_exac_common_3.hg38.vcf.gz.tbi(237kb)
├── hg38.dict(84kb)
├── hg38.fa(3,268,032kb)
├── hg38.fa.amb(22kb)
├── hg38.fa.ann(33kb)
├── hg38.fa.bwt(3,214,439kb)
├── hg38.fa.fai(24kb)
├── hg38.fa.pac(803,610kb)
├── hg38.fa.sa(1,607,220kb)
├── hg38.gtf(1,532,946kb)
├── Homo_sapiens.GRCh38.cdna.all.fa.gz(76,700kb)
├── Homo_sapiens.GRCh38.cdna.all.index(2,569,256kb)
└── TruSeq3-PE.fa(1kb)

❗Noted：The download_hg38.sh should be executed within the rnn_neo environment. TruSeq3-PE.fa is available from the source. Due to network fluctuations, file downloads may be interrupted. Ensure that the file size matches the one stated above.

✈️Software Installation

Mutation Annotation Tool

• ANNOVAR

After downloading, it is also necessary to build the reference library of hg38 within the ANNOVAR folder.

perl annotate_variation.pl -downdb -buildver hg38 -webfrom annovar refGene humandb/

❗Noted：If you're not utilizing the original input, it may not be necessary to install the software.

Feature Extraction Tool

• NetMHCpan
• NetChop
• NetCTLpan
• MixMHCpred
• PRIME
• BigMHC

Install the aforementioned software and include the paths of other software in the environment variable, excluding BigMHC. Ensure that the software can be directly invoked. Modify the path of BigMHC in the parameter.json file.

Installation tips

• Ensure that tcsh is installed
• Some tools need to download referenced data
• Some tools need to modify the path and tmp folder: NetChop, NetMHCpan, etc.
• Some tools need to be compiled using g++, such as: MixMHCpred
• For BigMHC, the environment is already set up in the rnn_neo conda environment; just download the source code.
• Tools other than BigMHC need to be added to the environment variable

export PATH=$PATH:your_path/netMHCpan-4.1

In short, please ensure you carefully read the README provided with the above tools. If you have any questions, feel free to contact us for assistance.

🔍 Usage

1. Configure the working path by modifying the parameter.json file to include the current working path.

2. Running code with instances of three input modes. Please be aware: ensure to adjust the input file and parameters according to the provided sample code.

python run.py

3. If you wish to test the original data input, execute the code in the test directory to download the sample data.

bash download_test.sh

4. The final result output patientid _ rank _ RNN.csv should be located within the calspace folder.

calspace
├── test1
│   └── test1_rank_RNN.csv
├── test2
│   └── test2_rank_RNN.csv
└── test3
    └── test3_rank_RNN.csv

📜 Authors and Contact

JiaJun Li and TianYi Qiu

[email protected] or [email protected] Fudan University, Shanghai, China