REPTILE-package {REPTILE}

Regulatory Element Prediction

Description

Predicting DNA regulatory elements based on epigenomic signatures. This package is more of a set of building blocks than a direct solution. REPTILE regulatory prediction pipeline is built on this R package. Please check the url below for details:

https://github.com/yupenghe/REPTILE

Details

Accurate enhancer identification is critical for understanding the spatiotemporal transcriptional regulation during development as well as the functional impact of disease-related non-coding genetic variants. REPTILE is a algorithm to identify the precise location of enhancers by integrating histone modification data and base-resolution DNA methylation profiles.

REPTILE was designed based on three observations: 1) regions that are differentially methylated (or differentially methylated regions, DMRs) across diverse cell and tissue types strongly overlap with enhancers. 2) With base-resolution DNA methylation data, the boundaries of DMRs can be accurately defined, circumventing the difficulty of determining enhancer boundaries. 3) DMR size is often smaller (~500bp) than known enhancers, known negative regions (regions with no observable enhancer activity) and genomic windows used in enhancer prediction (~2kb), all of which we termed as "query regions". Together with the association between transcription factor binding and DNA methylation level, DMRs may serve as high-resolution enhancer candidates and capture the local epigenomic patterns that would otherwise be averaged/washed out in analysis focusing on the query regions.

Running REPTILE involves four major steps. First, to identify DMRs, we compared the methylomes of target sample (where putative enhancers will be generated) and several other samples with different cell/tissue types (as reference). In the next step, input files for REPTILE are prepared, which store the information of query regions, DMRs and the epigenomic data. Taking these inputs, REPTILE represents each DMR or query region as a feature vector, where each element corresponds to either intensity or intensity deviation of one epigenetic mark. Intensity deviation is defined as the intensity in target sample subtracted by the mean intensity in reference samples (i.e. reference epigenome) and it captures the tissue-specificity of each epigenetic mark. In the third step, based on the feature vectors of known enhancers and negative regions as well as the feature vectors of the DMRs within them, we trained an enhancer model, containing two random forest classifiers, which respectively predict enhancer activities of query regions and DMRs. In the last step, REPTILE uses the enhancer model to calculate enhancer confidence scores for DMRs and query regions, based on which the final predictions are made.

The two key concepts on REPTILE are:

• Query regions - known enhancers, known negative regions and genomic windows used for enhancer prediction

• DMRs - differentially methylated regions

In REPTILE, DMRs are used as high-resolution candidates to capture the fine epigenomic signatures in query regions.

Author(s)

Yupeng He

Maintainer: Yupeng He <[e-mail]>

References

He, Yupeng et al., REPTILE: Regulatory Element Prediction based on TIssue-specific Local Epigenetic marks, in preparation

Examples

library("REPTILE")
data("rsd")

## Training (needs a few minutes and ~1.8 Gb memory)
reptile.model <- reptile_train(rsd$training_data$region_epimark,
                               rsd$training_data$region_label,
                               rsd$training_data$DMR_epimark,
                               rsd$training_data$DMR_label,
                               ntree=50)

## Prediction
## - REPTILE
pred <- reptile_predict(reptile.model,
                        rsd$test_data$region_epimark,
                        rsd$test_data$DMR_epimark)
## - Random guessing
pred_guess = runif(length(pred$D))
names(pred_guess) = names(pred$D)

## Evaluation
res_reptile <- reptile_eval_prediction(pred$D,
                                       rsd$test_data$region_label)
res_guess <- reptile_eval_prediction(pred_guess,
                                     rsd$test_data$region_label)
## - Print AUROC and AUPR
cat(paste0("REPTILE\n",
           "  AUROC = ",round(res_reptile$AUROC,digit=3),
           "\n",
           "  AUPR  = ",round(res_reptile$AUPR,digit=3))
    ,"\n")
cat(paste0("Random guessing\n",
           "  AUROC = ",round(res_guess$AUROC,digit=3),
           "\n",
           "  AUPR  = ",round(res_guess$AUPR,digit=3))
   ,"\n")