shunsunsun / orca_manuscript Goto Github PK

This repository contains the code and data required for reproducing the analyses in the Orca manuscript. For using Orca and training Orca models, please visit our main repository. For running Orca from our webserver, please visit orca.zhoulab.io.

Most of the analyses code are provided in jupyter notebook format. Each jupyter notebook contains a series of analyses and typically generates multiple plots for the same theme of analyses. For large-scale virtual screens, you can find scripts under the virtual_screen directory, and the Compartment_activity_screen.ipynb also contains code for running multiple compartment activity screen analyses.

The jupyter notebooks are largely grouped by topics:

• Compartment_activity_screen.ipynb : Analyses of sequence activities in chromatin compartment alterations
• Local_interaction_screens.ipynb : Identifying sequences that affect submegabase-scale genome interactions
• Local_interaction_multiplex_demo.ipynb : Multiplexed in silico mutagenesis demo example.
• Enhancerpolycomb_example.ipynb : Example predictions of enhancer-promoter interactions and Polycomb-mediated interactions.
• Model_performance.ipynb : Orca model prediction performance evaluations.
• Model_performance_256M.ipynb : Orca model prediction performance evaluations (32-256Mb).
• Model_performance_hctnoc.ipynb : Orca model prediction performance evaluations for the cohesin-depleted HCT119 model.
• StructuraVariants.ipynb : Prediction of structural variant effects on 3D genome interactions.

Other than Orca dependencies, you will also need jupyter, rpy2, and plotnine python packages which can be installed with Anaconda or pip. For R packages, we will use data.table, ggplot2, patchwork, ggridges, ggrastr, ggthemes, and limma.

You will need additional resource files for reproducing some of the analyses, and we have provided these files here(2.9G). Note that the jupyter notebooks use GPU to generate Orca predictions by default. You can generally switch to CPU by using use_cuda=False option, but they may be too slow for computationally intensive steps. You can also skip the computationally intensive steps by downloading our precomputed results files here(19.9G). We have provided code to load precomputed results in the jupyter notebooks.