StackEPI: identification of cell line-specific enhancer–promoter interactions based on stacking ensemble learning

Abstract

Background Understanding the regulatory role of enhancer–promoter interactions (EPIs) on specific gene expression in cells contributes to the understanding of gene regulation, cell differentiation, etc., and its identification has been a challenging task. On the one hand, using traditional wet experimental methods to identify EPIs often means a lot of human labor and time costs. On the other hand, although the currently proposed computational methods have good recognition effects, they generally require a long training time. Results In this study, we studied the EPIs of six human cell lines and designed a cell line-specific EPIs prediction method based on a stacking ensemble learning strategy, which has better prediction performance and faster training speed, called StackEPI. Specifically, by combining different encoding schemes and machine learning methods, our prediction method can extract the cell line-specific effective information of enhancer and promoter gene sequences comprehensively and in many directions, and make accurate recognition of cell line-specific EPIs. Ultimately, the source code to implement StackEPI and experimental data involved in the experiment are available at https://github.com/20032303092/StackEPI.git. Conclusions The comparison results show that our model can deliver better performance on the problem of identifying cell line-specific EPIs and outperform other state-of-the-art models. In addition, our model also has a more efficient computation speed.

Enhancer/gene relationships: need for more reliable genome-wide reference sets

“…They first train a model to discriminate true vs. false E/G based on distinctive features from the 1D data using a reference dataset of known E/G (most often determined using a combination of 1D data for enhancer and promoter identification and 3D or genetic data for the relationship identification) and a dataset of unsupported E/G as a negative control. When provided with a new relationship associated to 1D data attributes, the model is used to decide whether the relationship is more likely to be a true E/G than a false E/G ([29, 2, 14, 30, 36, 6, 37, 13, 20, 4, 15, 11]) (Figure 3).…”

“…Accordingly we also had to modify the sheffield.correlation.py script: 1) we added an all equal function to prevent divisions by zero in the Calculate Correlation function and 2) we added code to create the gene summary file that the script was supposed to take as input. The complete process to run the Sheffield method on the BENGI set can be found on this page 11 Finally we ran the Run-Average-Rank.sh script that evaluates the Average-Rank method on a BENGI set. This script takes as input a string defining the BENGI set and the version of the BENGI set, and outputs a 7 column tabulated file including for each E/G of the BENGI set, 1 or 0 according to whether this E/G is true or false in the BENGI set, the average rank score, the distance score, the correlation score, the distance rank, the correlation rank and the average rank between the distance and the correlation.…”

“…This script takes as input a string defining the BENGI set and the version of the BENGI set, and outputs a 7 column tabulated file including for each E/G of the BENGI set, 1 or 0 according to whether this E/G is true or false in the BENGI set, the average rank score, the distance score, the correlation score, the distance rank, the correlation rank and the average rank between the distance and the correlation. Here we also had to modify the bash script, first because the distance script now outputs a 4 instead of a 2 column file, but second and more importantly because we spotted a mistake in the calculation of the correlation rank, where the inverse distance is taken 11 https://genoweb.toulouse.inra.fr/ thoellinger/notes/notes BENGI/dnase expression correlation/correlation method with code.html instead of the correlation (we had to replace $2 by $3 in the following portion of the code sort -k3,3gr | awk 'BEGIN{x=0;r=0} {if (x != $2) r +=1; print $0 "\t" r "\t" ($NF+r)/2; x=$2}' Once again we plotted the Precision-Recall curves using an R code of our own. The complete process to evaluate the Average-Rank method on the BENGI sets can be found here 12 .…”

EPI-Trans: an effective transformer-based deep learning model for enhancer promoter interaction prediction

“…The performance of the model as a whole is enhanced by the output module’s subsequent use of a CNN and dense layer combination to further enhance these important properties. Recently, Fan and Peng [ 27 ] introduced a technique known as StackEPI, which merges several feature representations and classical machine learning algorithms, employs a stacking ensemble approach, and performs the prediction process solely based on promoter and enhancer gene sequences.…”

Analysis of super-enhancer using machine learning and its application to medical biology

“…The regulation of transcription by enhancers has been studied since the 1980s [ 43 ]. Enhancers in genetics are considered short (50–1500 base pairs) DNA regions to which proteins (activators) bind to increase the likelihood of transcription of a particular gene [ 44 , 45 ]. Often, the proteins that bind to these enhancers are called transcription factors.…”