Discover Regulatory DNA Elements Using Chromatin Signatures and Artificial Neural Network

Department of Internal Medicine, University of Iowa, 2294 CBRB, 285 Newton Road, Iowa City, IA 52242, USA.
Bioinformatics (Impact Factor: 4.98). 07/2010; 26(13):1579-86. DOI: 10.1093/bioinformatics/btq248
Source: PubMed


Recent large-scale chromatin states mapping efforts have revealed characteristic chromatin modification signatures for various types of functional DNA elements. Given the important influence of chromatin states on gene regulation and the rapid accumulation of genome-wide chromatin modification data, there is a pressing need for computational methods to analyze these data in order to identify functional DNA elements. However, existing computational tools do not exploit data transformation and feature extraction as a means to achieve a more accurate prediction.
We introduce a new computational framework for identifying functional DNA elements using chromatin signatures. The framework consists of a data transformation and a feature extraction step followed by a classification step using time-delay neural network. We implemented our framework in a software tool CSI-ANN (chromatin signature identification by artificial neural network). When applied to predict transcriptional enhancers in the ENCODE region, CSI-ANN achieved a 65.5% sensitivity and 66.3% positive predictive value, a 5.9% and 11.6% improvement, respectively, over the previously best approach.
CSI-ANN is implemented in Matlab. The source code is freely available at
Supplementary Materials are available at Bioinformatics online.

1 Follower
15 Reads
  • Source
    • "Meanwhile, indirect methods use correlational analysis of enhancer regions with some landmark DNA features for the inference of approximate locations—e.g., CpG island, chromatin or histone marks. Extensive studies on indirect methods are focused mainly in the generation and modelling of discriminative features from landmarks of supervised learning [2] [3]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Using Genetic Algorithm, this paper presents a modelling method to generate novel logical-based features from DNA sequences enriched with H3K4mel histone signatures. Current histone signature is mostly represented using k-mers content features incapable of representing all the possible complex interactions of various DNA segments. The main contributions are, among others: (a) demonstrating that there are complex interactions among sequence segments in the histone regions; (b) developing a parse tree representation of the logical complex features. The proposed novel feature is compared to the k-mers content features using datasets from the mouse (mm9) genome. Evaluation results show that the new feature improves the prediction performance as shown by f-measure for all datasets tested. Also, it is discovered that tree-based features generated from a single chromosome can be generalized to predict histone marks in other chromosomes not used in the training. These findings have a great impact on feature design considerations for histone signatures as well as other classifier design features.
    Bio-medical materials and engineering 09/2014; 24(6):3807-14. DOI:10.3233/BME-141210 · 1.09 Impact Factor
  • Source
    • "A marker radioisotope attaches to a specific radioligand, which we contains a chemical affinity for the specific tissue of interest. This marriage allows the combination of ligand and radioisotope to be carried and attached to the specific area of interest in the target organ [13]. Subsequently, due to the gamma-emission of the isotope [8], the ligand concentration is visualized by a gamma-camera. "
    [Show abstract] [Hide abstract]
    ABSTRACT: The recent introduction of high-resolution molecular imaging technology is considered by many experts as a major breakthrough that will potentially lead to a revolutionary paradigm shift in health care and revolutionize clinical practice. This paper explores the challenges and strengths of the current major imaging modalities, as well as the biophysics engineering their repertoire of capabilities. Advancements in the mechanical aspects of both PET and SPECT imaging will advance molecular imaging diagnostic capabilities and have a direct impact on clinical medicine and biomedical research practice. A better understanding of the strengths and limitations of functional imaging modalities in the context of their particular hardware and software mechanics will shed light onto how we can advance their diagnostic capabilities on a biological level. Herein, this paper demonstrates the fundamental biomechanical differences between PET and SPECT imaging, and how these fundamental differences translate into clinically relevant data acquisition for brain disorders.
    07/2013; 4:157. DOI:10.4172/2155-9619.1000157
  • Source
    • "A case concerning enhancer recognition is a time-delay neural network (TDNN) which combines 39 histone modifications [10]. In an independent test, 66.3% of the putative regions identified by this model overlapped with experimentally supported enhancers [10]. A SVM performs classification by seeking a hyperplane in high dimensional labeled feature space that optimally separates the data into two categories regarding the classification labels [66]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: As a class of cis-regulatory elements, enhancers were first identified as the genomic regions that are able to markedly increase the transcription of genes nearly 30 years ago. Enhancers can regulate gene expression in a cell-type specific and developmental stage specific manner. Although experimental technologies have been developed to identify enhancers genome-wide, the design principle of the regulatory elements and the way they rewire the transcriptional regulatory network tempo-spatially are far from clear. At present, developing predictive methods for enhancers, particularly for the cell-type specific activity of enhancers, is central to computational biology. In this review, we survey the current computational approaches for active enhancer prediction and discuss future directions.
    05/2013; 11(3). DOI:10.1016/j.gpb.2013.04.002
Show more


15 Reads
Available from