repli1D

Code used to simulate replication and to inverse MRT and RFD profiles to obtain a Probability of Origin Density Lansdcape (PODLS)

Install

git clone https://github.com/organic-chemistry/repli1D.git
cd  repli1D
conda create -c bioconda -n repli  keras tensorflow snakemake numpy joblib plotly
#conda create -c bioconda -n repli  keras tensorflow-gpu snakemake numpy joblib plotly

conda activate repli
pip install -e ./

Replicate results of the article

data for K562 GM Hela and Cerevisae are available in data/ directory at a 5000 bp resolution for human cells and 1000 bp resolution for Cerevisae. For each cell line, (here Hela for example) the following command:

snakemake -c 4 Hela_5000.sh --config output="my-folder/"

will create the file Hela_5000.sh needed to perform the optimisation. The paramater -c 4 set the number of threads used perform the replication. Then running

sh Hela.sh

will cary on the optimisation. For each iteration i, it will create three folders (inside my-folder):

• _RFD_to_init_small_opti_i # optimisation on the chromosome 2
• wholecell_i # the result of the whole cell simulation using I_i and the parameters from _RFD_to_init_small_opti_i/params.json
• RFD_to_init_nn_i+1 # with the neural network that ouput I_i+1 in the file nn_global_profiles.csv.

Running the code on a new data (refered as celltype)

This require to create a data file in the data folder name celltype_5000.csv with 5000 the resolution of the bins. It is a csv file with column: chrom,chromStart,chromEnd,MRT,OKSeq Then a new entry cellline must be added in the config.yaml file (by copying K562 of Cerevisae for example) and by adapting the parameter to the organism. Then running,

snakemake -c 4 celltype_5000.sh --config output="my-folder/"

will generate the script celltype.sh that must be run to carry on the optimisation. The paramater -c 4 set the number of threads used perform the replication

Description

The code to simulate replication and obtain RFD and MRT, given an input PODLS is in src/repli1d/fast_sim_break.py

example of use are in notebook_demo:

Typical simulation and comparison with experiment

Example with casadet

To inverse:

python src/repli1d/whole_pipeline_from_data_file.py --root /scratch/jarbona/yeast-inverse/ --speed 2.5 --max_epoch 2000 --repTime 30 --chr_sub 4 --window 41 --data data/yeast.csv  --introduction_time 5 --resolution 1

python src/repli1d/visu_signals.py --cell Hela --signal MRT , OKSeq , MCM
python src/repli1d/detect_and_simulate.py --cell Raji --ndiff 60 --experimental --name tmp/ --visu --signal peakRFDonly

python src/repli1d/detect_and_simulate.py --cell Cerevisae --signal MCM-beda  --experimental --name tmp/ --visu --resolution 5 --fspeed 0.24 --introduction_time 1 --mrt_res 5 --input --dori 1 --masking 25 --noise 0.0 --nsim 300 --wholecell --ndiff 20

python src/repli1d/whole_pipeline.py --cell Raji --savenoise --add_noise --root results/Raji_mask_bothopti_snoise_nnoise_std

Training of Neural Networks

Neural networks can be chosen by argument "ml_model" that can be specified to "jm_cnn_model", "jm_cnn_model_beta", "resnet", etc. (default value is jm_cnn_model.) for training the network on terminal, one sample code is as follows:

python src/repli1d/nn.py  --noenrichment --targets initiation --root training_dir/nn_K562_2000/ --listfile data/K562_2000_merged_histones_init.csv  --window 101 --wig 0 --predict_files data/K562_2000_merged_histones_init.csv --marks H3K4me1 H3K4me3 H3K27me3 H3K36me3 H3K9me3 H2A.Z H3K79me2 H3K9ac H3K4me2 H3K27ac H4K20me1 --datafile