Team-WB1

Meet-EU Team WB1

Topic B : Chromosome compartments

Introduction

We tried to use HMM for detecting compartments and noticed that when we use division to three compartments, two of the detected compartments have very similar epigenetic profiles (which suggest that they shouldn't have been divided to separate compartments) but lay on different chromosome arms. Then we performed Principal Component Analysis and noticed that often there is one component that corresponds to chromosome arms (it is positive on one arm and negative on the other arm). To remove the arm bias we tried running HMM on 6 first principal components, except the component which corresponds the most to the chromosome arms (so 5 out of 6 components).

Basic pipeline

You should first download intrachromosomal HiC data and epigenetic data from http://www.lcqb.upmc.fr/meetu/dataforstudent/
Then preprocess HiC matrices and the file with epigenetic marks using preprocess_mtx.py and preprocess_epi.py. Finally you can run the analysis with calc_comps.py.

Check out bash script run_analysis.sh that we used for obtaining the results shared on github.

calc_comps.py

This program detects the chromatin compartments. It builds HMM models for different number of compartments and chooses best scoring number. It can either choose a different number of compartments for each chromosome (default mode) or a consensus number that has the highest overall score (--cmp_consensus option).
For each chromosome the program checks if there is a principal component that corresponds well enough to the chromosome arms (parameter min_arm_score). If there is, then it calculates compartments before and after removing the arm component. Then it chooses the setup that results with the highest score.

  -i INPUT_DIR, --input_dir INPUT_DIR
                        Directory that contains the preprocessed HiC data
  -c CENTR_FILE, --centr_file CENTR_FILE
                        file with centromeric positions
  -e EPI_DIR, --epi_dir EPI_DIR
                        Directory that contains preprocessed data with
                        epigenetic marks
  -o OUT_DIR, --out_dir OUT_DIR
                        Directory for saving the results (compartments)
  -p PLOTS_DIR, --plots_dir PLOTS_DIR
                        Directory for saving the plots
  --cmp_consensus       Find best number of compartments for all chromosomes
                        (default: find best number for each chromosome
                        separately)
  --cmp_min CMP_MIN     Minimum number of compartments to consider
  --cmp_max CMP_MAX     Maximum number of compartments to consider
  -a MIN_ARM_SCORE, --min_arm_score MIN_ARM_SCORE
                        If arm_score of a principal component is higher than
                        min_arm_score than it can be removed (between 0 and 1,
                        recommended values: 0.6-0.9

100kb_resolution_intrachromosomal

Results

Predictions for 5 cell lines (GM12878, HMEC, HUVEC, IMR90 and NHEK) using Hi-C data at 100kB resolution can be found in results folders. The folder named "results_full" contains the results obtained using --cmp_consensus option and the folder "results_separate" contains separate results for each chromosome, allowing for differences in optimal number of compartments between them. Each folder also contains plots comparing self-developed compartmentalisation score between running HMM on 6 PC and 5 PC (without the arm component).