README.md

In yasin-uzun/MAPLE.1.0:

MAPLE: Methylome Association by Predictive Linkage to Expression

An R package for predicting gene activity level for single cell DNA Methylation data

MAPLE is a supervised learning algorithm that is developed to predict the gene activity levels of individual cells from single cell DNA Methylation. It is implemented in R and uses CpG methylation sites in promoter regions to infer the activity levels of the genes.

Installation

To install MAPLE, type the following commands in R command prompt:

library(devtools)
install_github("yasin-uzun/MAPLE.1.0")

Annotation Data

To run MethylPredict, you need some genome annotation files. You can download those annotation from here:

Annotation Data

Built Models

We trained multiple models using different multi-omics training data. You can download them from here:

Trained Models

Example Data

We provide an example dataset o test MethylPredict. This data is a subset of CpG files from the study of Luo et al [1] (GSE97179). The original data has over 3000 cells. In here, we just provide data 100 cells to give a quick run:

Example Dataset

Running

Assuming you have downloaded the test data and you have the CpG cov files (in Bismark format), you can run MethylPredict as follows:

library(MAPLE)

#Set directory names
annot_dir = 'data/annot/'
model_dir = 'data/models/Clark/'
cov_dir = 'data/example/cov_files/'

#Set input files
annot_file = paste0(annot_dir,'/gencode.mm10.vM22.genes.bed')
cpg_content_file =  paste0(annot_dir,'/regions.genes.tss_ud_5K.cpg_ratio.bin_size_500.mm10.rds')

#Compute binned data
#IMPORTANT NOTE: If you have multiple types of methylation (CpG, nonCpG), you need to use
#the prefix CpG or CpH in the file names and use the specific methylation_type argument for filtering.
#If you just have one methylation type (eg. CpG only), you can specify the methylation_type argument empty as below
binned_list = compute_binned_met_counts(cov_dir = cov_dir, annot_file = annot_file, methylation_type = "")

#Compute meta cells
meta_object = compute_meta_cells(df_met =  binned_list[["df_binned_met"]],
                                 df_demet =  binned_list[["df_binned_demet"]])

#Generate features                               
fr_list = get_fr_list(meta_data = meta_object, cpg_content_file = cpg_content_file)

#Load CNN model and predict
cnn_model_file = paste0(model_dir, '/cnn_model.hd5')
predict_cnn = cnn_predict(fr_list, cnn_model_file)

#Load Elastic model and predict
elastic_model_file = paste0(model_dir, '/elastic_model.rds')
predict_elastic = elastic_predict(fr_list, elastic_model_file)

#Load RF model and predict
rf_model_file = paste0(model_dir, '/rf_model.rds')
predict_rf = rf_predict(fr_list, rf_model_file)

#Compute Ensemble prediction
prediction_list = list(predict_cnn, predict_elastic, predict_rf)
predict_ensem = ensemble_predict(prediction_list)

#Convert gene activity predictions into matrix format (genesxcells)
gene_activity_matrix = convert_preds_to_matrix(predict_ensem)

Citation

If you use our software, please cite our paper [2] .

Contact

For comments and questions, please contact Yasin Uzun (uzuny at email chop edu)

References

[1] Luo, Chongyuan, Christopher L. Keown, Laurie Kurihara, Jingtian Zhou, Yupeng He, Junhao Li, Rosa Castanon, et al. 2017. “Single-Cell Methylomes Identify Neuronal Subtypes and Regulatory Elements in Mammalian Cortex.” Science 357 (6351): 600–604

[2] Yasin Uzun, Hao Wu, Kai Tan. Predictive modeling of single-cell DNA methylome data enhances integration with transcriptome data. Genome Research. 31 (1): 101–109. 2021.

yasin-uzun/MAPLE.1.0 documentation built on Oct. 9, 2024, 12:49 p.m.