UV-DDR-3D analysis

---

This repository contains exploratory and hierarchical genome architecture analysis code, and research code for dataset generation, training, inference for GNN algorithm, as described in the manuscript titled "UV-induced reorganization of 3D genome mediates DNA damage response".

All analysis was performed on a Linux system (Ubuntu 22.04.4 LTS). To reproduce the analysis and the figures in the manuscript, code is particularly deposited in the form of Jupyter notebooks (Python), while also preserving code outputs.

---

This code repo contains two sections:

1. Python notebooks under 3D folder contain genome architecture analysis:
   1. boundaries/tads_rework.ipynb contains analysis code with respect to TADs and boundaries/insulation (Fig3 a,b,c,d,e,f,g, Supp.Fig2a,b,c,d,e,f).
   2. compartments/comp_ev_check.ipynb contains the correlation analysis of compartment profiles (EV) and phasing tracks (Supp Table).
   3. compartments/comp_switch.ipynb contains the analysis code for correlation of compartment profiles with respect to timecourse DEG clusters. (Fig2b,c,d)
   4. compartments/compartments.ipynb contains analysis code with respect to compartments, saddle plots saddle data analysis, compartment strength. (Fig1d,e,f, Supp.Fig1a)
   5. compartments/ev-change_deseq.ipynb & ev-tpm.ipynb EV change profiles and DEGs between samples, and TPMs per compartment profiles (Supp.Fig1c,d).
   6. compartments/raincloud.ipynb Top 20% compartment profiles interactions change (Fig1g).
   7. RNA contains R scripts for DESeq time series/DEG analysis and rna_3D.ipynb for annotating GNN analysis with respect to RNA-Seq data (Fig6e,f). Also, source data files of coseq results for expression clustering related analysis.
   8. unibind contains shell scripts to run TFBS differential enrichment analysis on regions of GNN comparison and chromatin loop anchors, also contains a notebook for visualization (Fig4h, Fig6c,d).
   9. distance_decay.ipynb notebook is for the analysis for distance decay and powerlaw estimations (Fig1 b,c).
   10. gnn_analyse/gnn_results_analyse.ipynb notebook is for the analysis of comparisons from GNN results (Fig5c,6a,b).
   11. gnn_analyse/gnn_dnase_faire.ipynb notebook is for the annotation of comparison groups with DNAse-Seq and FAIRE-Seq signals (Fig5d).
   12. loops/loops.ipynb notebook is for loop calling, loop and anchor analysis (Fig4a,b,c,d,e,f,g Supp.Fig3).
   13. repair_ops.ipynb notebook is for downstream processing of XR-Seq and Damage-Seq (and simulations) bed files generated with xr-ds-snakemake Snakemake pipeline, and converting to bigwigs.
2. GNN folder contain dataset generation, training and inference code for the GNN workflow described in the manuscript.
   1. Follow save_obs_exp_qt.ipynb in order to perform preprocessing steps for the matrices (Both megamap, and Hi-C matrices of our own). (You can easily modify for your own .mcool/.cool files.)
   2. Follow training_dataset.ipynb in order to generate in-memory datasets to sample target and query graphs from, for training the GNN model. (You can easily modify for your own data.)
   3. Use train.sh script to start training. (An example slurm batch script is provided)
   4. Use predict.sh for the inference of the model on the contact matrices to be compared. (GNN Model checkpoint used in the analysis can be found under chekpoints/last.ckpt, also comparison raw results between Control and 12min as reference point can be found as results_t0_t12.tsv ) (An example slurm batch script is provided)

Input Data

Processed files in order to reproduce the analysis code were made available on GEO database. Multi-resolution contact matrices in mcool format can be found with the identifier GSE268350 for Hi-C data, and TPM quantification data can be found with the identifier GSE268349 for RNA-seq data.

Also, all raw sequencing data generated in this study have been submitted to the NCBI SRA database, under accession number PRJNA1053630. All sequencing data were first preprocessed to QC with fastp with flag --detect_adapter_for_pe. Hi-C contacts matrices were generated (.hic format) with Juicer v1.6 pipeline, with default parameters. .hic format (q30 filtered) has been converted to .mcool format with hicConvertFormat subprogram of HiCExplorer. Matrix balancing has been performed with balance subprogram of cooler with parameters --cis-only --max-iters 1000 --blacklist, where blacklist file is here. Accordingly, all of the 3D genome research code presented in this code repository operates on .cool/.mcool data format.

Dependencies

Below is the list of required Python packages, installing via Mamba is advised.

• python=3.11.0
• cooler=0.9.2
• cooltools=0.5.4
• powerlaw
• bioframe=0.4.1
• coolpuppy=1.1.0
• scikit-learn=1.3.0
• pytorch=2.0.1
• pytorch-geometric=2.3.1
• pytorch-lightning=2.0.6
• gseapy=1.0.6

Contact

Please email vogulcan@sabanciuniv.edu or raise an issue in the github repository with any questions about installation or usage.