These scripts use openmmlib (https://github.com/mirnylab/openmm-polymer-legacy) and simplex algorithm to simulate HiC map of a given section on a chromatin (e.g. hg19 chr8:129M-131M) on High Performance Clusters. It is setup to exam changes in TAD boudary strength and compartment strength between different cell lines/ conditions.
How it works:
- Use sbatch to submit driver.sh
- driver.sh submits a job on a cpu node to run driver.py
- driver.py sets parameters for simplex including initial guesses, then subprocesses optimization_iteration.sh
- optimization_iteration.sh submit initial simulation jobs then submit iteration.sh
- iteration.sh submits iteration.py on a cpu node
- In each iteration, iteration.py generates next guess parameter set based on current correlations.
Ending criteria: Reach max iterations/ correlation stop improving and parameters stop changing
Each simulation job includes:
Submit simulation as 10 array jobs, each on a GPU node
Calculate contact frequency maps from simulations (CPU node)
Calculate Spearman correlation of HiC map and simulated contact frequency (CPU node)
Simulation principle:
Use loop extrusion to model TADs (dynamic simulation)
Use sticky to model compartment interactions (steady-state simulation)
Chromatin is modeled as self avoiding chain