Automatic prediction of non-coding RNA genes in prokaryotes based on compositional statistics

Although non-coding RNA (ncRNA) genes do not encode proteins, they play vital roles in cells by producing functionally important RNAs. In this paper, we present a novel method for predicting ncRNA genes based on compositional features extracted directly from gene sequences. Our method consists of two Support Vector Machine (SVM) models--Codon model which uses codon usage features derived from ncRNA genes and protein-coding genes and Kmer model which utilizes features of nucleotide and dinucleotide frequency extracted respectively from ncRNA genes and randomly chosen genome sequences. The 10-fold cross-validation accuracy for the two models is found to be 92% and 91%, respectively. Thus, we could make an automatic prediction of ncRNA genes in one genome without manual filtration of protein-coding genes. After applying our method in Sulfolobus solfataricus genome, 25 prediction results have been generated according to 25 cut-off pairs. We have also applied the approach in E. coli and found our results comparable to those of previous studies. In general, our method enables automatic identification of ncRNA genes in newly sequenced prokaryotic genomes.     

Seven clusters in genomic triplet distributions

In several recent papers new gene-detection algorithms were proposed for detecting protein-coding regions without requiring a learning dataset of already known genes. The fact that unsupervised gene-detection is possible is closely connected to the existence of a cluster structure in oligomer frequency distributions. In this paper we study the cluster structure of several genomes in the space of their triplet frequencies, using a pure data exploration strategy. Several complete genomic sequences were analyzed, using the visualization of tables of triplet frequencies in a sliding window. The distribution of 64-dimensional vectors of triplet frequencies displays a well-detectable cluster structure. The structure was found to consist of seven clusters, corresponding to protein-coding information in three possible phases in one of the two complementary strands and in the non-coding regions with high accuracy (higher than 90% on nucleotide level). Visualizing and understanding the structure allows to analyze effectively the performance of different gene-prediction tools. Since the method does not require extraction of ORFs, it can be applied even for unassembled genomes.     

A compression-based approach for coding sequences identification. I. Application to prokaryotic genomes.

Most of the gene prediction algorithms for prokaryotes are based on Hidden Markov Models or similar machine-learning approaches, which imply the optimization of a high number of parameters. The present paper presents a novel method for the classification of coding and non-coding regions in prokaryotic genomes, based on a suitably defined compression index of a DNA sequence. The main features of this new method are the non-parametric logic and the costruction of a dictionary of words extracted from the sequences. These dictionaries can be very useful to perform further analyses on the genomic sequences themselves. The proposed approach has been applied on some prokaryotic complete genomes, obtaining optimal scores of correctly recognized coding and non-coding regions. Several false-positive and false-negative cases have been investigated in detail, which have revealed that this approach can fail in the presence of highly structured coding regions (e.g., genes coding for modular proteins) or quasi-random non-coding regions (e.g., regions hosting non-functional fragments of copies of functional genes; regions hosting promoters or other protein-binding sequences). We perform an overall comparison with other gene-finder software, since at this step we are not interested in building another gene-finder system, but only in exploring the possibility of the suggested approach.     

Large multiple organism gene finding by collapsed Gibbs sampling.

The Gibbs sampling method has been widely used for sequence analysis after it was successfully applied to the problem of identifying regulatory motif sequences upstream of genes. Since then, numerous variants of the original idea have emerged: however, in all cases the application has been to finding short motifs in collections of short sequences (typically less than 100 nucleotides long). In this paper, we introduce a Gibbs sampling approach for identifying genes in multiple large genomic sequences up to hundreds of kilobases long. This approach leverages the evolutionary relationships between the sequences to improve the gene predictions, without explicitly aligning the sequences. We have applied our method to the analysis of genomic sequence from 14 genomic regions, totaling roughly 1.8 Mb of sequence in each organism. We show that our approach compares favorably with existing ab initio approaches to gene finding, including pairwise comparison based gene prediction methods which make explicit use of alignments. Furthermore, excellent performance can be obtained with as little as four organisms, and the method overcomes a number of difficulties of previous comparison based gene finding approaches: it is robust with respect to genomic rearrangements, can work with draft sequence, and is fast (linear in the number and length of the sequences). It can also be seamlessly integrated with Gibbs sampling motif detection methods.     

Identification of protein-coding sequences using the hybridization of 18S rRNA and mRNA during translation

We introduce a new approach in this article to distinguish protein-coding sequences from non-coding sequences utilizing a period-3, free energy signal that arises from the interactions of the 3′-terminal nucleotides of the 18S rRNA with mRNA. We extracted the special features of the amplitude and the phase of the period-3 signal in protein-coding regions, which is not found in non-coding regions, and used them to distinguish protein-coding sequences from non-coding sequences. We tested on all the experimental genes from Saccharomyces cerevisiae and Schizosaccharomyces pombe. The identification was consistent with the corresponding information from GenBank, and produced better performance compared to existing methods that use a period-3 signal. The primary tests on some fly, mouse and human genes suggests that our method is applicable to higher eukaryotic genes. The tests on pseudogenes indicated that most pseudogenes have no period-3 signal. Some exploration of the 3′-tail of 18S rRNA and pattern analysis of protein-coding sequences supported further our assumption that the 3′-tail of 18S rRNA has a role of synchronization throughout translation elongation process. This, in turn, can be utilized for the identification of protein-coding sequences.     

Identification of 96 single nucleotide polymorphisms in eight genes involved in iron metabolism: efficiency of bioinformatic extraction compared with a systematic sequencing approach

Single nucleotide polymorphisms (SNPs) can significantly contribute to the characterization of the genes predisposing to iron overloads or deficiencies. We report an SNP survey of coding and non-coding regions of eight genes involved in iron metabolism, by two successive methods. First, we made use of the public domain sequence data, by using assembled expressed sequence tags, non-redundant sequences, and SNP database screening. We extracted 77 potential SNPs of which only 31 could be further validated by sequencing DNA from 44 unrelated multi-ethnic individuals. Our results indicate that a bioinformatic approach may be effective only in those cases where candidate SNPs are extracted from two different data sources or in cases of experimentally confirmed SNPs. Second, additional systematic sequencing of DNA from 24 unrelated Breton subjects increased the number of SNPs over a total length of 86 kb to 96. The average distance between the SNPs and minor allele frequencies were higher than reported by others authors; this discrepancy may reflect the nature of the genes studied and the ethnic homogeneity of our test population.     

Predicting internal exons by oligonucleotide composition and discriminant analysis of spliceable open reading frames.

A new method which predicts internal exon sequences in human DNA has been developed. The method is based on a splice site prediction algorithm that uses the linear discriminant function to combine information about significant triplet frequencies of various functional parts of splice site regions and preferences of oligonucleotides in protein coding and intron regions. The accuracy of our splice site recognition function is 97% for donor splice sites and 96% for acceptor splice sites. For exon prediction, we combine in a discriminant function the characteristics describing the 5'-intron region, donor splice site, coding region, acceptor splice site and 3'-intron region for each open reading frame flanked by GT and AG base pairs. The accuracy of precise internal exon recognition on a test set of 451 exon and 246693 pseudoexon sequences is 77% with a specificity of 79%. The recognition quality computed at the level of individual nucleotides is 89% for exon sequences and 98% for intron sequences. This corresponds to a correlation coefficient for exon prediction of 0.87. The precision of this approach is better than other methods and has been tested on a larger data set. We have also developed a means for predicting exon-exon junctions in cDNA sequences, which can be useful for selecting optimal PCR primers.     

Inferring an organism-specific optimal threshold for predicting protein coding regions in eukaryotes based on a bootstrapping algorithm

The accuracy of prediction methods based on power spectrum analysis depends on the threshold that is used to discriminate between protein coding and non-coding sequences in the genomes of eukaryotes. Because the structure of genes vary among different eukaryotes, it is difficult to determine the best prediction threshold for a eukaryote relying only on prior biological knowledge. To improve the accuracy of prediction methods based on power spectral analysis, we developed a novel method based on a bootstrap algorithm to infer organism-specific optimal thresholds for eukaryotes. As prior information, our method requires the input of only a few annotated protein coding regions from the organism being studied. Our results show that using the calculated optimal thresholds for our test datasets, the average prediction accuracy of our method is 81%, an increase of 19% over that obtained using the same empirical threshold P = 4 for all datasets. The proposed method is simple and convenient and easily applied to infer optimal thresholds that can be used to predict coding regions in the genomes of most organisms.     

Mining Bacillus subtilis chromosome heterogeneities using hidden Markov models

We present here the use of a new statistical segmentation method on the Bacillus subtilis chromosome sequence. Maximum likelihood parameter estimation of a hidden Markov model, based on the expectation-maximization algorithm, enables one to segment the DNA sequence according to its local composition. This approach is not based on sliding windows; it enables different compositional classes to be separated without prior knowledge of their content, size and localization. We compared these compositional classes, obtained from the sequence, with the annotated DNA physical map, sequence homologies and repeat regions. The first heterogeneity revealed discriminates between the two coding strands and the non-coding regions. Other main heterogeneities arise; some are related to horizontal gene transfer, some to t-enriched composition of hydrophobic protein coding strands, and others to the codon usage fitness of highly expressed genes. Concerning potential and established gene transfers, we found 9 of the 10 known prophages, plus 14 new regions of atypical composition. Some of them are surrounded by repeats, most of their genes have unknown function or possess homology to genes involved in secondary catabolism, metal and antibiotic resistance. Surprisingly, we notice that all of these detected regions are a + t-richer than the host genome, raising the question of their remote sources.     

In search of the small ones: improved prediction of short exons in vertebrates, plants, fungi and protists

MOTIVATION: Prediction of the coding potential for stretches of DNA is crucial in gene calling and genome annotation, where it is used to identify potential exons and to position their boundaries in conjunction with functional sites, such as splice sites and translation initiation sites. The ability to discriminate between coding and non-coding sequences relates to the structure of coding sequences, which are organized in codons, and by their biased usage. For statistical reasons, the longer the sequences, the easier it is to detect this codon bias. However, in many eukaryotic genomes, where genes harbour many introns, both introns and exons might be small and hard to distinguish based on coding potential. RESULTS: Here, we present novel approaches that specifically aim at a better detection of coding potential in short sequences. The methods use complementary sequence features, combined with identification of which features are relevant in discriminating between coding and non-coding sequences. These newly developed methods are evaluated on different species, representative of four major eukaryotic kingdoms, and extensively compared to state-of-the-art Markov models, which are often used for predicting coding potential. The main conclusions drawn from our analyses are that (1) combining complementary sequence features clearly outperforms current Markov models for coding potential prediction in short sequence fragments, (2) coding potential prediction benefits from length-specific models, and these models are not necessarily the same for different sequence lengths and (3) comparing the results across several species indicates that, although our combined method consistently performs extremely well, there are important differences across genomes. SUPPLEMENTARY DATA: http://bioinformatics.psb.ugent.be/.     

Evolutionary changes of nucleotide sequences of papova viruses BKV and SV40: They are possibly hybrids

Summary Complete nucleotide sequences were compared between papova viruses BKV and SV40 and the degrees of sequence divergences were compared between structurally and/or functionally different segments or genes in details. It was shown that the rate of synonymous substitution is not only very high but also approximately uniform among different genes in these viruses as in eukaryotic genes examined to date. While all the non-coding regions including the intron showed marked sequence preservation which is in sharp contrasted with the case of eukaryotic genes where the large bulk of non-coding regions evolve at a rate as rapidly as that of synonymous substitution. It is remarkable that a long continuous stretch of sequence including the putative VPX gene and a 5 half of VP2 gene showed strong homology between BKV and SV40. A close examination of the pattern of base substitutions revealed that this unusual homology was derived by recombination between the two viruses during their evolution. On the basis of the pattern of base substitutions and the bias in code word utilization, we also showed that the putative VPX gene actually could code for a functional polypeptide. In papova viruses, the 3 terminal sequence of VP2/3 gene overlaps with the 5 terminal sequence of VPI gene. The pattern of base substitutions in the overlapping segment was examined in detail in comparison with those in the non-overlapping portions of VP2/3 and VP1 genes. It was shown that the evolutionary mode of the overlapping genes is in good agreement with our previous prediction.     

Ab initio gene identification in metagenomic sequences

We describe an algorithm for gene identification in DNA sequences derived from shotgun sequencing of microbial communities. Accurate ab initio gene prediction in a short nucleotide sequence of anonymous origin is hampered by uncertainty in model parameters. While several machine learning approaches could be proposed to bypass this difficulty, one effective method is to estimate parameters from dependencies, formed in evolution, between frequencies of oligonucleotides in protein-coding regions and genome nucleotide composition. Original version of the method was proposed in 1999 and has been used since for (i) reconstructing codon frequency vector needed for gene finding in viral genomes and (ii) initializing parameters of self-training gene finding algorithms. With advent of new prokaryotic genomes en masse it became possible to enhance the original approach by using direct polynomial and logistic approximations of oligonucleotide frequencies, as well as by separating models for bacteria and archaea. These advances have increased the accuracy of model reconstruction and, subsequently, gene prediction. We describe the refined method and assess its accuracy on known prokaryotic genomes split into short sequences. Also, we show that as a result of application of the new method, several thousands of new genes could be added to existing annotations of several human and mouse gut metagenomes.     

Consensus folding of unaligned RNA sequences revisited.

As one of the earliest problems in computational biology, RNA secondary structure prediction (sometimes referred to as "RNA folding") problem has attracted attention again, thanks to the recent discoveries of many novel non-coding RNA molecules. The two common approaches to this problem are de novo prediction of RNA secondary structure based on energy minimization and the consensus folding approach (computing the common secondary structure for a set of unaligned RNA sequences). Consensus folding algorithms work well when the correct seed alignment is part of the input to the problem. However, seed alignment itself is a challenging problem for diverged RNA families. In this paper, we propose a novel framework to predict the common secondary structure for unaligned RNA sequences. By matching putative stacks in RNA sequences, we make use of both primary sequence information and thermodynamic stability for prediction at the same time. We show that our method can predict the correct common RNA secondary structures even when we are given only a limited number of unaligned RNA sequences, and it outperforms current algorithms in sensitivity and accuracy.     

Identification of a conserved sequence in the non-coding regions of many human genes.

We have analyzed a sequence of approximately 70 base pairs (bp) that shows a high degree of similarity to sequences present in the non-coding regions of a number of human and other mammalian genes. The sequence was discovered in a fragment of human genomic DNA adjacent to an integrated hepatitis B virus genome in cells derived from human hepatocellular carcinoma tissue. When one of the viral flanking sequences was compared to nucleotide sequences in GenBank, more than thirty human genes were identified that contained a similar sequence in their non-coding regions. The sequence element was usually found once or twice in a gene, either in an intron or in the 5' or 3' flanking regions. It did not share any similarities with known short interspersed nucleotide elements (SINEs) or presently known gene regulatory elements. This element was highly conserved at the same position within the corresponding human and mouse genes for myoglobin and N-myc, indicating evolutionary conservation and possible functional importance. Preliminary DNase I footprinting data suggested that the element or its adjacent sequences may bind nuclear factors to generate specific DNase I hypersensitive sites. The size, structure, and evolutionary conservation of this sequence indicates that it is distinct from other types of short interspersed repetitive elements. It is possible that the element may have a cis-acting functional role in the genome.     

Predicting disordered regions in proteins based on decision trees of reduced amino acid composition.

Intrinsically unstructured proteins (IUPs) are proteins lacking a fixed three dimensional structure or containing long disordered regions. IUPs play an important role in biology and disease. Identifying disordered regions in protein sequences can provide useful information on protein structure and function, and can assist high-throughput protein structure determination. In this paper we present a system for predicting disordered regions in proteins based on decision trees and reduced amino acid composition. Concise rules based on biochemical properties of amino acid side chains are generated for prediction. Coarser information extracted from the composition of amino acids can not only improve the prediction accuracy but also increase the learning efficiency. In cross-validation tests, with four groups of reduced amino acid composition, our system can achieve a recall of 80% at a 13% false positive rate for predicting disordered regions, and the overall accuracy can reach 83.4%. This prediction accuracy is comparable to most, and better than some, existing predictors. Advantages of our approach are high prediction accuracy for long disordered regions and efficiency for large-scale sequence analysis. Our software is freely available for academic use upon request.     

A method for measuring the non-random bias of a codon usage table

We describe a new statistical method for measuring bias in the codon usage table of a gene. The test is based on the multinomial and Poisson distributions. The method is used to scan DNA sequences and measure the strength of codon preference. For E. Coli we show that the strength of codon preference is related to levels of gene expression. The method can also be used to compare base triplet frequencies with those expected from the base composition. This second type of codon bias test is useful for distinguishing coding from non-coding regions.     

Exploration of multivariate analysis in microbial coding sequence modeling

ABSTRACT: BACKGROUND: Gene finding is a complicated procedure that encapsulates algorithms for coding sequence modeling, identification of promoter regions, issues concerning overlapping genes and more. In the present study we focus on coding sequence modeling algorithms; that is, algorithms for identification and prediction of the actual coding sequences from genomic DNA. In this respect, we promote a novel multivariate method known as Canonical Powered Partial Least Squares (CPPLS) as an alternative to the commonly used Interpolated Markov model (IMM). Comparisons between the methods were performed on DNA, codon and protein sequences with highly conserved genes taken from several species with different genomic properties. RESULTS: The multivariate CPPLS approach classified coding sequence substantially better than the commonly used IMM on the same set of sequences. We also found that the use of CPPLS with codon representation gave significantly better classification results than both IMM with protein (p < 0.001) and with DNA (p < 0.001). Further, although the mean performance was similar, the variation of CPPLS performance on codon representation was significantly smaller than for IMM (p < 0.001). CONCLUSIONS: The performance of coding sequence modeling can be substantially improved by using an algorithm based on the multivariate CPPLS method applied to codon or DNA frequencies.     

Identification of 142 single nucleotide polymorphisms in 41 candidate genes for rheumatoid arthritis in the Japanese population

Single-nucleotide polymorphisms (SNPs) can make an important contribution to our understanding of genetic backgrounds that may influence medical conditions and ethnic diversity. We undertook a systematic survey of genomic DNA for SNPs located not only in coding sequences but also in non-coding regions (e.g., introns and 5' flanking regions) of selected genes. Using DNA samples from 48 Japanese patients with rheumatoid arthritis (RA) as templates, we surveyed 41 genes that represent candidates for RA, screening a total of 104 kb of DNA (30 kb of coding sequences and 74 kb of non-coding DNA). Within this 104 kb of genomic sequences we identified 163 polymorphisms (1 per 638 bases on average), of which 142 were single-nucleotide substitutions and the remainder, insertions or deletions. Of the coding SNPs, 52% were non-synonymous substitutions, and non-conservative amino acid changes were observed in a quarter of those. Sixty-nine polymorphisms showed high frequencies for minor alleles (more than 15%) and 20 revealed low frequencies (<5%). Our results indicated a greater average distance between SNPs than others have reported, but this disparity may reflect the type of genes surveyed and/or the relative ethnic homogeneity of our test population.