Probabilistic methods of identifying genes in prokaryotic genomes: connections to the HMM theory

In this paper, we review developments in probabilistic methods of gene recognition in prokaryotic genomes with the emphasis on connections to the general theory of hidden Markov models (HMM). We show that the Bayesian method implemented in GeneMark, a frequently used gene-finding tool, can be augmented and reintroduced as a rigorous forward-backward (FB) algorithm for local posterior decoding described in the HMM theory. Another earlier developed method, prokaryotic GeneMark.hmm, uses a modification of the Viterbi algorithm for HMM with duration to identify the most likely global path through hidden functional states given the DNA sequence. GeneMark and GeneMark.hmm programs are worth using in concert for analysing prokaryotic DNA sequences that arguably do not follow any exact mathematical model. The new extension of GeneMark using the FB algorithm was implemented in the software program GeneMark.fba. Given the DNA sequence, this program determines an a posteriori probability for each nucleotide to belong to coding or non-coding region. Also, for any open reading frame (ORF), it assigns a score defined as a probabilistic measure of all paths through hidden states that traverse the ORF as a coding region. The prediction accuracy of GeneMark.fba determined in our tests was compared favourably to the accuracy of the initial (standard) GeneMark program. Comparison to the prokaryotic GeneMark.hmm has also demonstrated a certain, yet species-specific, degree of improvement in raw gene detection, ie detection of correct reading frame (and stop codon). The accuracy of exact gene prediction, which is concerned about precise prediction of gene start (which in a prokaryotic genome unambiguously defines the reading frame and stop codon, thus, the whole protein product), still remains more accurate in GeneMarkS, which uses more elaborate HMM to specifically address this task.     

GeneMark.hmm: new solutions for gene finding.

The number of completely sequenced bacterial genomes has been growing fast. There are computer methods available for finding genes but yet there is a need for more accurate algorithms. The GeneMark. hmm algorithm presented here was designed to improve the gene prediction quality in terms of finding exact gene boundaries. The idea was to embed the GeneMark models into naturally derived hidden Markov model framework with gene boundaries modeled as transitions between hidden states. We also used the specially derived ribosome binding site pattern to refine predictions of translation initiation codons. The algorithm was evaluated on several test sets including 10 complete bacterial genomes. It was shown that the new algorithm is significantly more accurate than GeneMark in exact gene prediction. Interestingly, the high gene finding accuracy was observed even in the case when Markov models of order zero, one and two were used. We present the analysis of false positive and false negative predictions with the caution that these categories are not precisely defined if the public database annotation is used as a control.     

<Emphasis Type="Italic">Ab initio</Emphasis> gene identification: Prokaryote genome annotation with GeneScan and GLIMMER

We compare the annotation of three complete genomes using theab initio methods of gene identification GeneScan and GLIMMER. The annotation given in GenBank, the standard against which these are compared, has been made using GeneMark. We find a number of novel genes which are predicted by both methods used here, as well as a number of genes that are predicted by GeneMark, but are not identified by either of the nonconsensus methods that we have used. The three organisms studied here are all prokaryotic species with fairly compact genomes. The Fourier measure forms the basis for an efficient non-consensus method for gene prediction, and the algorithm GeneScan exploits this measure. We have bench-marked this program as well as GLIMMER using 3 complete prokaryotic genomes. An effort has also been made to study the limitations of these techniques for complete genome analysis. GeneScan and GLIMMER are of comparable accuracy insofar as gene-identification is concerned, with sensitivities and specificities typically greater than 0.9. The number of false predictions (both positive and negative) is higher for GeneScan as compared to GLIMMER, but in a significant number of cases, similar results are provided by the two techniques. This suggests that there could be some as-yet unidentified additional genes in these three genomes, and also that some of the putative identifications made hitherto might require re-evaluation. All these cases are discussed in detail.     

A probabilistic method for identifying start codons in bacterial genomes.

As the pace of genome sequencing has accelerated, the need for highly accurate gene prediction systems has grown. Computational systems for identifying genes in prokaryotic genomes have sensitivities of 98-99% or higher (Delcher et al., Nucleic Acids Res., 27, 4636-4641, 1999). These accuracy figures are calculated by comparing the locations of verified stop codons to the predictions. Determining the accuracy of start codon prediction is more problematic, however, due to the relatively small number of start sites that have been confirmed by independent, non-computational methods. Nonetheless, the accuracy of gene finders at predicting the exact gene boundaries at both the 5' and 3' ends of genes is of critical importance for microbial genome annotation, especially in light of the important signaling information that is sometimes found on the 5' end of a protein coding region. In this paper we propose a probabilistic method to improve the accuracy of gene identification systems at finding precise translation start sites. The new system, RBSfinder, is tested on a validated set of genes from Escherichia coli, for which it improves the accuracy of start site locations predicted by computational gene finding systems from the range 67-77% to 90% correct.     

An unsupervised classification scheme for improving predictions of prokaryotic TIS

Background  

Although it is not difficult for state-of-the-art gene finders to identify coding regions in prokaryotic genomes, exact prediction of the corresponding translation initiation sites (TIS) is still a challenging problem. Recently a number of post-processing tools have been proposed for improving the annotation of prokaryotic TIS. However, inherent difficulties of these approaches arise from the considerable variation of TIS characteristics across different species. Therefore prior assumptions about the properties of prokaryotic gene starts may cause suboptimal predictions for newly sequenced genomes with TIS signals differing from those of well-investigated genomes.     

Accuracy improvement for identifying translation initiation sites in microbial genomes

MOTIVATION: At present the computational gene identification methods in microbial genomes have a high prediction accuracy of verified translation termination site (3' end), but a much lower accuracy of the translation initiation site (TIS, 5' end). The latter is important to the analysis and the understanding of the putative protein of a gene and the regulatory machinery of the translation. Improving the accuracy of prediction of TIS is one of the remaining open problems. RESULTS: In this paper, we develop a four-component statistical model to describe the TIS of prokaryotic genes. The model incorporates several features with biological meanings, including the correlation between translation termination site and TIS of genes, the sequence content around the start codon; the sequence content of the consensus signal related to ribosomal binding sites (RBSs), and the correlation between TIS and the upstream consensus signal. An entirely non-supervised training system is constructed, which takes as input a set of annotated coding open reading frames (ORFs) by any gene finder, and gives as output a set of organism-specific parameters (without any prior knowledge or empirical constants and formulas). The novel algorithm is tested on a set of reliable datasets of genes from Escherichia coli and Bacillus subtillis. MED-Start may correctly predict 95.4% of the start sites of 195 experimentally confirmed E.coli genes, 96.6% of 58 reliable B.subtillis genes. Moreover, the test results indicate that the algorithm gives higher accuracy for more reliable datasets, and is robust to the variation of gene length. MED-Start may be used as a postprocessor for a gene finder. After processing by our program, the improvement of gene start prediction of gene finder system is remarkable, e.g. the accuracy of TIS predicted by MED 1.0 increases from 61.7 to 91.5% for 854 E.coli verified genes, while that by GLIMMER 2.02 increases from 63.2 to 92.0% for the same dataset. These results show that our algorithm is one of the most accurate methods to identify TIS of prokaryotic genomes. AVAILABILITY: The program MED-Start can be accessed through the website of CTB at Peking University: http://ctb.pku.edu.cn/main/SheGroup/MED_Start.htm.     

原核基因翻译起始位点预测的新方法

翻译起始位点（TIS,即基因5’端）的精确定位是原核生物基因预测的一个关键问题,而基因组GC含量和翻译起始机制的多样性是影响当前TIS预测水平的重要因素．结合基因组结构的复杂信息（包括GC含量、TIS邻近序列及上游调控信号、序列编码潜能、操纵子结构等）,发展刻画翻译起始机制的数学统计模型,据此设计TIS预测的新算法MED．StartPlus．并将MED．StartPlus与同类方法RBSfinder、GS．Finder、MED-Start、TiCo和Hon-yaku等进行系统地比较和评价．测试针对两种数据集进行：当前14个已知的TIS被确认的基因数据集,以及300个物种中功能已知的基因数据集．测试结果表明,MED-StartPlus的预测精度在总体上超过同类方法．尤其是对高GC含量基因组以及具有复杂翻译起始机制的基因组,MED-StartPlus具有明显的优势．     

GS-Finder: a program to find bacterial gene start sites with a self-training method

In this paper, a self-training method is proposed to recognize translation start sites in bacterial genomes without a prior knowledge of rRNA in the genomes concerned. Many features with biological meanings are incorporated, including mononucleotide distribution patterns near the start codon, the start codon itself, the coding potential and the distance from the most-left start codon to the start codon. The proposed method correctly predicts 92% of the translation start sites of 195 experimentally confirmed Escherichia coli CDSs, 96% of 58 reliable Bacillus subtilis CDSs and 82% of 140 reliable Synechocystis CDSs. Moreover, the self-training method presented might also be used to relocate the translation start sites of putative CDSs of genomes, which are predicted by gene-finding programs. After post-processing by the method presented, the improvement of gene start prediction of some gene-finding programs is remarkable, e.g., the accuracy of gene start prediction of Glimmer 2.02 increases from 63 to 91% for 832 E. coli reliable CDSs. An open source computer program to implement the method, GS-Finder, is freely available for academic purposes from http://tubic.tju.edu.cn/GS-Finder/.     

Heuristic approach to deriving models for gene finding.

Computer methods of accurate gene finding in DNA sequences require models of protein coding and non-coding regions derived either from experimentally validated training sets or from large amounts of anonymous DNA sequence. Here we propose a new, heuristic method producing fairly accurate inhomogeneous Markov models of protein coding regions. The new method needs such a small amount of DNA sequence data that the model can be built 'on the fly' by a web server for any DNA sequence >400 nt. Tests on 10 complete bacterial genomes performed with the GeneMark.hmm program demonstrated the ability of the new models to detect 93.1% of annotated genes on average, while models built by traditional training predict an average of 93.9% of genes. Models built by the heuristic approach could be used to find genes in small fragments of anonymous prokaryotic genomes and in genomes of organelles, viruses, phages and plasmids, as well as in highly inhomogeneous genomes where adjustment of models to local DNA composition is needed. The heuristic method also gives an insight into the mechanism of codon usage pattern evolution.     

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.     

Evaluation of gene prediction software using a genomic data set: application to Arabidopsis thaliana sequences

MOTIVATION: The annotation of the Arabidopsis thaliana genome remains a problem in terms of time and quality. To improve the annotation process, we want to choose the most appropriate tools to use inside a computer-assisted annotation platform. We therefore need evaluation of prediction programs with Arabidopsis sequences containing multiple genes. RESULTS: We have developed AraSet, a data set of contigs of validated genes, enabling the evaluation of multi-gene models for the Arabidopsis genome. Besides conventional metrics to evaluate gene prediction at the site and the exon levels, new measures were introduced for the prediction at the protein sequence level as well as for the evaluation of gene models. This evaluation method is of general interest and could apply to any new gene prediction software and to any eukaryotic genome. The GeneMark.hmm program appears to be the most accurate software at all three levels for the Arabidopsis genomic sequences. Gene modeling could be further improved by combination of prediction software. AVAILABILITY: The AraSet sequence set, the Perl programs and complementary results and notes are available at http://sphinx.rug.ac.be:8080/biocomp/napav/. CONTACT: Pierre.Rouze@gengenp.rug.ac.be.     

ZCURVE: a new system for recognizing protein-coding genes in bacterial and archaeal genomes

A new system, ZCURVE 1.0, for finding protein- coding genes in bacterial and archaeal genomes has been proposed. The current algorithm, which is based on the Z curve representation of the DNA sequences, lays stress on the global statistical features of protein-coding genes by taking the frequencies of bases at three codon positions into account. In ZCURVE 1.0, since only 33 parameters are used to characterize the coding sequences, it gives better consideration to both typical and atypical cases, whereas in Markov-model-based methods, e.g. Glimmer 2.02, thousands of parameters are trained, which may result in less adaptability. To compare the performance of the new system with that of Glimmer 2.02, both systems were run, respectively, for 18 genomes not annotated by the Glimmer system. Comparisons were also performed for predicting some function-known genes by both systems. Consequently, the average accuracy of both systems is well matched; however, ZCURVE 1.0 has more accurate gene start prediction, lower additional prediction rate and higher accuracy for the prediction of horizontally transferred genes. It is shown that the joint applications of both systems greatly improve gene-finding results. For a typical genome, e.g. Escherichia coli, the system ZCURVE 1.0 takes approximately 2 min on a Pentium III 866 PC without any human intervention. The system ZCURVE 1.0 is freely available at: http://tubic. tju.edu.cn/Zcurve_B/.     

SCGPred： A Score-based Method for Gene Structure Prediction by Combining Multiple Sources of Evidence

Predicting protein-coding genes still remains a significant challenge. Although a variety of computational programs that use commonly machine learning methods have emerged, the accuracy of predictions remains a low level when implementing in large genomic sequences. Moreover, computational gene finding in newly se- quenced genomes is especially a difficult task due to the absence of a training set of abundant validated genes. Here we present a new gene-finding program, SCGPred, to improve the accuracy of prediction by combining multiple sources of evidence. SCGPred can perform both supervised method in previously well-studied genomes and unsupervised one in novel genomes. By testing with datasets composed of large DNA sequences from human and a novel genome of Ustilago maydi, SCGPred gains a significant improvement in comparison to the popular ab initio gene predictors. We also demonstrate that SCGPred can significantly improve prediction in novel genomes by combining several foreign gene finders with similarity alignments, which is superior to other unsupervised methods. Therefore, SCGPred can serve as an alternative gene-finding tool for newly sequenced eukaryotic genomes. The program is freely available at http：//bio.scu.edu.cn/SCGPred/.     

Effects of choice of DNA sequence model structure on gene identification accuracy

MOTIVATION: Markov chain models of DNA sequences have frequently been used in gene finding algorithms. Performance of the algorithm critically depends on the model structure and parameters. Still, the issue of choosing the model structure has not been studied with sufficient attention. RESULTS: We have assessed performance of several types of Markov chain models, both fixed order (FO) models and models with interpolation, within the framework of the GeneMark algorithm. The performance was measured in two ways: (i) the accuracy of detection of protein-coding potential in artificial DNA sequences and (ii) the accuracy of identifying genes in real prokaryotic genomes. We observed that the models built by deleted interpolation (DI) slightly outperformed other models in detecting protein-coding potential in artificial DNA sequences with GC content in the medium range and also in detecting genes in real genomes with medium GC content. For artificial and real genomic DNA with high or low GC content, we observed that the models built by DI were in some cases slightly outperformed by FO models.     

Gene re-annotation in genome of the extremophile Pyrobaculum aerophilum by using bioinformatics methods

In this paper, we re-annotated the genome of Pyrobaculum aerophilum str. IM2, particularly for hypothetical ORFs. The annotation process includes three parts. Firstly and most importantly, 23 new genes, which were missed in the original annotation, are found by combining similarity search and the ab initio gene finding approaches. Among these new genes, five have significant similarities with function-known genes and the rest have significant similarities with hypothetical ORFs contained in other genomes. Secondly, the coding potentials of the 1645 hypothetical ORFs are re-predicted by using 33 Z curve variables combined with Fisher linear discrimination method. With the accuracy being 99.68%, 25 originally annotated hypothetical ORFs are recognized as non-coding by our method. Thirdly, 80 hypothetical ORFs are assigned with potential functions by using similarity search with BLAST program. Re-annotation of the genome will benefit related researches on this hyperthermophilic crenarchaeon. Also, the re-annotation procedure could be taken as a reference for other archaeal genomes. Details of the revised annotation are freely available at http://cobi.uestc.edu.cn/resource/paero/     

Gene Re-annotation in Genome of the Extremophile Pyrobaculum Aerophilum by Using Bioinformatics Methods

Abstract

In this paper, we re-annotated the genome of Pyrobaculum aerophilum str. IM2, particularly for hypothetical ORFs. The annotation process includes three parts. Firstly and most importantly, 23 new genes, which were missed in the original annotation, are found by combining similarity search and the ab initio gene finding approaches. Among these new genes, five have significant similarities with function-known genes and the rest have significant similarities with hypothetical ORFs contained in other genomes. Secondly, the coding potentials of the 1645 hypothetical ORFs are re-predicted by using 33 Z curve variables combined with Fisher linear discrimination method. With the accuracy being 99.68%, 25 originally annotated hypothetical ORFs are recognized as non-coding by our method. Thirdly, 80 hypothetical ORFs are assigned with potential functions by using similarity search with BLAST program. Re-annotation of the genome will benefit related researches on this hyperthermophilic crenarchaeon. Also, the re-annotation procedure could be taken as a reference for other archaeal genomes. Details of the revised annotation are freely available at http://cobi.uestc.edu.cn/resource/paero/     

Reannotation of Shewanella oneidensis genome

As more and more complete bacterial genome sequences become available, the genome annotation of previously sequenced genomes may become quickly outdated. This is primarily due to the discovery and functional characterization of new genes. We have reannotated the recently published genome of Shewanella oneidensis with the following results: 51 new genes have been identified, and functional annotation has been added to the 97 genes, including 15 new and 82 existing ones with previously unassigned function. The identification of new genes was achieved by predicting the protein coding regions using the HMM-based program GeneMark.hmm. Subsequent comparison of the predicted gene products to the non-redundant protein database using BLAST and the COG (Clusters of Orthologous Groups) database using COGNITOR provided for the functional annotation.     

Systematic integration of experimental data and models in systems biology

Background

The quality of automated gene prediction in microbial organisms has improved steadily over the past decade, but there is still room for improvement. Increasing the number of correct identifications, both of genes and of the translation initiation sites for each gene, and reducing the overall number of false positives, are all desirable goals.

Results

With our years of experience in manually curating genomes for the Joint Genome Institute, we developed a new gene prediction algorithm called Prodigal (PROkaryotic DYnamic programming Gene-finding ALgorithm). With Prodigal, we focused specifically on the three goals of improved gene structure prediction, improved translation initiation site recognition, and reduced false positives. We compared the results of Prodigal to existing gene-finding methods to demonstrate that it met each of these objectives.

Conclusion

We built a fast, lightweight, open source gene prediction program called Prodigal http://compbio.ornl.gov/prodigal/. Prodigal achieved good results compared to existing methods, and we believe it will be a valuable asset to automated microbial annotation pipelines.     

FrameD: A flexible program for quality check and gene prediction in prokaryotic genomes and noisy matured eukaryotic sequences

We describe FrameD, a program that predicts coding regions in prokaryotic and matured eukaryotic sequences. Initially targeted at gene prediction in bacterial GC rich genomes, the gene model used in FrameD also allows to predict genes in the presence of frameshifts and partially undetermined sequences which makes it also very suitable for gene prediction and frameshift correction in unfinished sequences such as EST and EST cluster sequences. Like recent eukaryotic gene prediction programs, FrameD also includes the ability to take into account protein similarity information both in its prediction and its graphical output. Its performances are evaluated on different bacterial genomes. The web site (http://genopole.toulouse.inra.fr/bioinfo/FrameD/FD) allows direct prediction, sequence correction and translation and the ability to learn new models for new organisms.