Tutorials for predicting human and mouse data modalities using EPCOT v2.

Introduction

EPCOT v2 predicts diverse molecular modalities for the central 500kb regions extracted from 600kb segments across the entire genome.

Human Data Prediction

  1. Input file preparation. The initial BAM file, is pre-processed to become the input for EPCOT v2. The name of the generated .bigWig file is formatted as ‘{cell_line_name}_atac’. Note The detailed locations (chrom id, start pos, end pos) of each 600kb region is stored in file “input_region_600kb.bed”, which can be found on Github. There are 11,123 600kb regions across the whole human genome.

## Generate the .bigWig file
bamCoverage --bam input.bam -o ${cell_line}_atac.bigWig --outFileFormat bigwig --normalizeUsing RPGC \
--effectiveGenomeSize 2913022398 --Offset 1 --binSize 10 --numberOfProcessors 12 \
--blackListFileName ../data/black_list.bed
## Run atac_process.py to generate the .pickle file which is the required input format of EPCOT v2.
python atac_process.py ${cell_line}_atac.bigWig

  1. Prediction. Please confirm the directories of input ${cell_line}_atac.pickle file and the output ${cell_line}.pickle file. You can specify which modality you want to predict in file “tt_pred_gw.py” by modifying variable “pred_modalities”.

## Run the model execution file tt_pred_gw.py to predict desired molecular modalities.
python tt_pred_gw.py

  1. Extraction of each predicted modality. Assume the output file is called ‘spleen.pickle’. Each predicted sample is the central 500kb from 600kb segments, composed of 500 1kb genomic bin.

import pickle
with open('spleen.pickle', 'rb') as file:
    spleen_pickle = pickle.load(file)
## All predicted modalities
spleen_pickle['epi']
spleen_pickle['rna'] # expected shape: (11123, 500, 3)
spleen_pickle['bru'] # expected shape: (11123, 500, 3)
spleen_pickle['microc']  # expected shape: (11123, 500, 500, 2)
spleen_pickle['hic'] # expected shape: (11123, 100, 100, 3) Hi-C is predicted at 5kb resolution.
spleen_pickle['intacthic'] # expected shape: (11123, 500, 2)
spleen_pickle['rna_strand'] # expected shape: (11123, 500, 2)
spleen_pickle['external_tf']
spleen_pickle['tt'] # expected shape: (11123, 500, 2)
spleen_pickle['groseq'] # expected shape: (11123, 500, 2)
spleen_pickle['grocap'] # expected shape: (11123, 500, 4)
spleen_pickle['proseq'] # expected shape: (11123, 500, 3)
spleen_pickle['netcage'] # expected shape: (11123, 500, 2)
spleen_pickle['starr'] # expected shape: (11123, 500, 1)

Here is the explanation of each modality that can be predicted:

  1. Epigenomic features. The list of epigenomic features can be found on Github in a file named “epigenomes.txt”.

  2. RNA-seq. 1. CAGE-seq 2. Total RNA-seq 3. PolyA+ RNA-seq

  3. Bru-seq. 1. Bru-seq 2. BruUV-seq 3. BruChase-seq

  4. Micro-c. 1. O/E normalized Micro-C 2. KR normalized Micro-C

  5. Hi-C. 1. CTCF ChIA-PET 2. RNApol2 ChIA-PET 3. Hi-C

  6. Intact Hi-C. 1. O/E normalized intact Hi-C 2. KR normalized intact Hi-C

  7. RNA Strand. 1. Total RNA-seq (forward) 2. Total RNA-seq (reverse)

  8. Additional TFs. The list of additional TFs can be found on Github in a file named unseeen_tf.txt

  9. TT-seq. 1. TT-seq (forward) 2. TT-seq (reverse)

  10. GRO-seq. 1. GRO-seq (forward) 2. GRO-seq (reverse)

  11. GRO-cap. 1. GRO-cap (forward) 2. GRO-cap (reverse) 3. GRO-cap_wTAP (forward) 4. GRO-cap_wTAP(reverse)

  12. PRO-seq. 1. PRO-seq (forward) 2. PRO-seq (reverse) 3. PRO-cap

  13. NET-CAGE. 1. NET-CAGE (forward) 2. NET-CAGE (reverse)

  14. STARR-seq. 1. STARR-seq