Utilizing sequence intrinsic composition to classify protein-coding and long non-coding transcripts.
Sun L, Luo H, Bu D, Zhao G, Yu K, Zhang C, Liu Y, Chen R, Zhao Y. · 2013
Read Utilizing sequence intrinsic composition to classify protein-coding and long non-coding transcripts. by Sun L, Luo H, Bu D, Zhao G, Yu K, Zhang C, Liu Y, Chen R, Zhao Y. (2013) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc3783192.txt.
Introduction
Liang Sun and eight co-authors presented a classification tool in Nucleic Acids Research in 2013 for telling protein-coding transcripts from long non-coding ones. Sequencing now produces floods of transcripts, above all in species with poor annotation, and older tools leaned on known proteins or evolution signals that can mislead.
Their tool, the Coding-Non-Coding Index, profiles patterns of adjoining nucleotide triplets in the sequence itself, with no need for prior annotation. It classified transcripts accurately across species, handled incomplete transcripts and sense-antisense pairs, and revealed evolutionary divergence between vertebrate and invertebrate genes, and between plants. Applied to orangutan data, it produced a catalog of that species' long non-coding RNAs. The authors offer the software freely online.
Catalog record
| CREATOR | Sun L, Luo H, Bu D, Zhao G, Yu K, Zhang C, Liu Y, Chen R, Zhao Y. |
|---|---|
| YEAR | 2013 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Utilizing sequence intrinsic composition to classify protein-coding and long non-coding transcripts By Liang Sun, Haitao Luo, Dechao Bu, Guoguang Zhao, Kuntao Yu, Changhai Zhang, Yuanning Liu, Runsheng Chen, Yi Zhao Published in: Nucleic Acids Research DOI: 10.1093/nar/gkt646 ABSTRACT It is a challenge to classify protein-coding or non-coding transcripts, especially those re-constructed from high-throughput sequencing data of poorly annotated species. This study developed and evaluated a powerful signature tool, Coding-Non-Coding Index (CNCI), by profiling adjoining nucleotide triplets to effectively distinguish protein-coding and non-coding sequences independent of known annotations. CNCI is effective for classifying incomplete transcripts and sense–antisense pairs. The implementation of CNCI offered highly accurate classification of transcripts assembled from whole-transcriptome sequencing data in a cross-species manner, that demonstrated gene evolutionary divergence between vertebrates, and invertebrates, or between plants, and provided a long non-coding RNA catalog of orangutan. CNCI software is available at http://www.bioinfo.org/software/cnci. INTRODUCTION The recent progress in the ENCODE project suggests that >70% of human genome sequences are transcribed into processed or primary RNAs (1,2). For other species, numerous novel transcripts are identified by advances in RNA sequencing techniques (RNA-seq) (3). It remains a challenge to identify sequence differences between protein-coding and non-coding transcripts. During the past 5 years, several tools, such as CPC and phyloCSF, have been developed to classify protein-coding or non-coding transcripts using information on open reading frame (ORF), known protein database, or evolutionary signatures (4–6). These approaches have been effective in identifying long coding transcripts for putative ORF or peptide hits. However, they are not suitable for identifying long non-coding RNAs (lncRNAs), which may contain short protein-like sub-sequences or long putative ORFs (6,7). Moreover, these available tools search for the best segment of evolutionary signatures but may also lead to significant false-positive and false-negative discoveries, because most of the functional non-coding RNAs have higher evolutionary conservation relative to introns, which suggested the conserved element exists in these non-coding sequences (8), and in contrast, many newly evolved proteins do not contain a conserved ORF (7). In addition, for poorly annotated species or those without whole-genome sequence, it is hard to define the transcript status using the existing tools. To overcome these challenges, we developed Coding-Non-Coding Index (CNCI) software, a powerful signature tool, by profiling adjoining nucleotide triplets (ANT), to effectively distinguish between protein-coding and non-coding sequences independent of known annotations. In comparison with the existing tools, CNCI showed better performance in many aspects, especially for classification of incomplete transcripts and sense–antisense transcript pairs. Notably, CNCI performed well uniformly on all the species of the vertebrates, but relatively poorly for invertebrates and plants, when using human data as training sets. Because CNCI can classify protein-coding and non-coding RNAs solely based on sequence intrinsic composition, it is potentially applicable to a variety of species without whole-genome sequence or with poorly annotated information. As an example of application for poorly annotated species, we tested CNCI on a published RNA-Seq data set from six organs of orangutan. As a result, CNCI annotated 7697 novel transcripts as lncRNAs, which contributed to the first comprehensive orangutan lncRNA catalog. MATERIALS AND METHODS Data description For human training sets, protein-coding genes were collected from RefSeq database and long non-coding genes were collected from Gencode (v11) (9). For mouse testing sets, both protein-coding and non-coding genes were collected from Ensembl (v65) (10) database. As other testing sets, gene annotation of other vertebrates or plants was downloaded from Ensembl (v69) and EnsemblPlants (v16) (10) databases, respectively. LincRNA catalog was obtained from human body map (11). Whole-transcriptome sequencing data of the six organs of orangutan were obtained from the study of David Brawand et al. (3) and downloaded from Gene Expression Omnibus under accession code GSE30352. All the data were summarized in Supplementary Table S1, and the length distributions of the human and mouse transcript collections are depicted in Supplementary Table S2. Calculation of usage frequency of ANT In this study, we began analyzing the usage frequency of ANT in coding domain sequence (CDS) and non-coding RNA sequences. There were 64*64 ANT, and we calculated usage frequency of each ANT as follows: Where X indicates the ANT, Sj(Xi) is the occurrence number of X in sequence i, XiN is Xi’s total occurrence number in the data set, T is all kinds of ANT’s total occurrence number in the data set and XiF is the usage frequency of the ANT. In human sequences, the usage frequency of ANT was calculated in 30 507 CDS and 18 566 long non-coding transcripts. In mouse sequences, the usage frequency was calculated in 25 316 CDS and 8696 long non-coding transcripts (Supplementary Figure S1). The log-ratio of the usage frequency of all kinds of ANT constituted a 64*64 ANT score matrix (Figure 1a and b). Figure 1.Illustration of ANT score matrix and CNCI framework. The score of each ANT is calculated for human (a) or mouse (b). The three black rows or columns represent three stop codons, including UAA, UAG and UGA (corresponding to ATT, ATC and ACT in cDNA sequence, respectively), which shows low frequency in protein-coding sequence. (c) The framework of CNCI. The top panel shows the process of a sequence in a testing set. For a given sequence, six MLCDS regions (represented by six lines) are identified from six reading frames (represented by six color arrow lines) using a sliding window and dynamic programming algorithm. Then, an MLCDS region with a maximal S-score is selected to incorporate into an SVM. The bottom panel shows the training and classification process. Reliable protein-coding and non-coding sequences are used as a training set, and five features are extracted to train SVM, which classifies the incorporating sequence into protein-coding or non-coding sequence. Utilization of the sliding window to scan each transcript It is an essential step for our approach to identify the most-like CDS (MLCDS) region of each transcript. We first used the sliding window to analyze each transcript by setting the size of the sliding window and the scan step as one ANT. To verify the size of the sliding window (represented by parameter N) to achieve the robust final classification result, we defined a series of N’s with different lengths (from 30 nt to 300 nt, with 30 nt as a step-forward interval). For each of the N sequences, a classifier was trained with human training data. The sensitivity–specificity curves of the classification were then calculated on the testing set (Supplementary Figure S2). In our classification model, N of 150 nt was found to be the robust, and thus we chose N = 150 nt, which is a proper size longer than small RNAs but shorter than lncRNAs, into the classification model. The window scanned each transcript six times to generate six reading frames. At the same time, CNCI calculated the sequence-score (S-score) of each window based on ANT score matrix; thus, a given transcript produced six discrete numerical arrays. In the process of sliding window, each transcript was converted into six arrays, and each array was composed of sliding window’s S-score of one kind of reading frame. S-score can reflect the coding ability of each sliding window, and our initial purpose was to find out sub-sequences (described as MLCDS earlier in the text) in each of the six reading frames