Hongyang449/Ocelot

Ocelot: Improved Epigenome Imputation Reveals Asymmetric Predictive Relationships Across Histone Modifications

Ocelot is a machine learning approach to impute epigenomes across tissues and cell types. It ranked first in the ENCODE Imputation Challenge with high accuracy on held-out prospective data. Beyond high predictive performance, it offers a new way to investigate the cross-histone regulations based on large-scale epigenomics datasets. Please contact (hyangl@umich.edu or gyuanfan@umich.edu) if you have any questions or suggestions.

---

Installation

Git clone a copy of code:

git clone https://github.com/GuanLab/Ocelot.git

Required dependencies

• python (3.6.5)
• numpy (1.13.3). It comes pre-packaged in Anaconda.
• pyBigWig A package for quick access to and create of bigwig files.

conda install pybigwig -c bioconda

• lightgbm(2.3.0) A gradient boosting tree-based algorithm with fast training speed and high efficienty.

pip install lightgbm

• tensorflow (1.14.0) A popular deep learning package.

conda install tensorflow-gpu

• keras (2.2.5) A popular deep learning package using tensorflow backend.

conda install keras

Dataset

• The ENCODE Imputation Challenge dataset
• Ocelot imputation for the ENCODE3 histone mark dataset
• Ocelot imputation for the Roadmap histone mark dataset

Code of Ocelot and evaluation on the challenge data

• data_challenge
• code_challenge
  

Reproducing all these imputations and evaluations requires considerable time even with super computing resources, we therefore also provide the processed data, trained models and predictions together with the reproducible scripts.

• 0a. Ocelot - the challenge final submission or npy format
• 0b. Ensemble predictions without DNA
• 1a. Processed data
• 2a. Trained lightGBM and neural network models and predictions
• 2b. Trained lightGBM and neural network models without DNA and predictions
  

For benchmarking, predictions from Avocado and ChromImpute are also provided:

• 3a. Avocado predictions
• 4a. ChromImpute models and predictions

Mapping between letter, id and histone mark in challenge

For simplicity, we map the epigeneic marks to captital letters as follows:

letter	id	mark
C	M02	DNase-seq
D	M18	H3K36me3
E	M17	H3K27me3
F	M16	H3K27ac
G	M20	H3K4me1
H	M22	H3K4me3
I	M29	H3K9me3
J	M01	ATAC-seq

For example, in the "CDEH_I" design, we used four marks (C, D, E, H) as cell type-specific features to predict mark I.

Data processing and model building scripts for ENCODE3 imputation

• data_encode3
• code_encode3

Data processing and model building scripts for Roadmap imputation

• data_roadmap
• code_roadmap

Code for SHAP analysis

• code_shap

Code of Ocelot final submission to the ENCODE Imputation Challenge

• challenge solution