An assessment of gene prediction accuracy in large DNA sequences

One of the first useful products from the human genome will be a set of predicted genes. Besides its intrinsic scientific interest, the accuracy and completeness of this data set is of considerable importance for human health and medicine. Though progress has been made on computational gene identification in terms of both methods and accuracy evaluation measures, most of the sequence sets in which the programs are tested are short genomic sequences, and there is concern that these accuracy measures may not extrapolate well to larger, more challenging data sets. Given the absence of experimentally verified large genomic data sets, we constructed a semiartificial test set comprising a number of short single-gene genomic sequences with randomly generated intergenic regions. This test set, which should still present an easier problem than real human genomic sequence, mimics the approximately 200kb long BACs being sequenced. In our experiments with these longer genomic sequences, the accuracy of GENSCAN, one of the most accurate ab initio gene prediction programs, dropped significantly, although its sensitivity remained high. Conversely, the accuracy of similarity-based programs, such as GENEWISE, PROCRUSTES, and BLASTX was not affected significantly by the presence of random intergenic sequence, but depended on the strength of the similarity to the protein homolog. As expected, the accuracy dropped if the models were built using more distant homologs, and we were able to quantitatively estimate this decline. However, the specificities of these techniques are still rather good even when the similarity is weak, which is a desirable characteristic for driving expensive follow-up experiments. Our experiments suggest that though gene prediction will improve with every new protein that is discovered and through improvements in the current set of tools, we still have a long way to go before we can decipher the precise exonic structure of every gene in the human genome using purely computational methodology.

Comparative gene prediction in human and mouse

The completion of the sequencing of the mouse genome promises to help predict human genes with greater accuracy. While current ab initio gene prediction programs are remarkably sensitive (i.e., they predict at least a fragment of most genes), their specificity is often low, predicting a large number of false-positive genes in the human genome. Sequence conservation at the protein level with the mouse genome can help eliminate some of those false positives. Here we describe SGP2, a gene prediction program that combines ab initio gene prediction with TBLASTX searches between two genome sequences to provide both sensitive and specific gene predictions. The accuracy of SGP2 when used to predict genes by comparing the human and mouse genomes is assessed on a number of data sets, including single-gene data sets, the highly curated human chromosome 22 predictions, and entire genome predictions from ENSEMBL. Results indicate that SGP2 outperforms purely ab initio gene prediction methods. Results also indicate that SGP2 works about as well with 3x shotgun data as it does with fully assembled genomes. SGP2 provides a high enough specificity that its predictions can be experimentally verified at a reasonable cost. SGP2 was used to generate a complete set of gene predictions on both the human and mouse by comparing the genomes of these two species. Our results suggest that another few thousand human and mouse genes currently not in ENSEMBL are worth verifying experimentally.

Prediction of many new exons and introns in Plasmodium falciparum chromosome 2

The current prediction or genes in the Plasmodium falciparum genome database relies upon a limited number of specially developed computer algorithms. We have re-annotated the sequence of chromosome 2 of P. falciparum by a computer-assisted manual analysis. which is described here. Of 161 newly predicted introns, we have experimentally confirmed 98. We regard 110 introns from the previously published analyses as probable, we delete 3, change 26 and add 135. We recognise 214 genes in chromosome 2. We have predicted introns in 121 genes. The increased complexity or gene structure on chromosome 2 is likely to be mirrored by the entire genome. (C) 2001 Elsevier Science B.V. All rights reserved.

Gene finding in the chicken genome.

BACKGROUND: Despite the continuous production of genome sequence for a number of organisms, reliable, comprehensive, and cost effective gene prediction remains problematic. This is particularly true for genomes for which there is not a large collection of known gene sequences, such as the recently published chicken genome. We used the chicken sequence to test comparative and homology-based gene-finding methods followed by experimental validation as an effective genome annotation method. RESULTS: We performed experimental evaluation by RT-PCR of three different computational gene finders, Ensembl, SGP2 and TWINSCAN, applied to the chicken genome. A Venn diagram was computed and each component of it was evaluated. The results showed that de novo comparative methods can identify up to about 700 chicken genes with no previous evidence of expression, and can correctly extend about 40% of homology-based predictions at the 5' end. CONCLUSIONS: De novo comparative gene prediction followed by experimental verification is effective at enhancing the annotation of the newly sequenced genomes provided by standard homology-based methods.

SGP-1: prediction and validation of homologous genes based on sequence alignments

Conventional methods of gene prediction rely on the recognition of DNA-sequence signals, the coding potential or the comparison of a genomic sequence with a cDNA, EST, or protein database. Reasons for limited accuracy in many circumstances are species-specific training and the incompleteness of reference databases. Lately, comparative genome analysis has attracted increasing attention. Several analysis tools that are based on human/mouse comparisons are already available. Here, we present a program for the prediction of protein-coding genes, termed SGP-1 (Syntenic Gene Prediction), which is based on the similarity of homologous genomic sequences. In contrast to most existing tools, the accuracy of SGP-1 depends little on species-specific properties such as codon usage or the nucleotide distribution. SGP-1 may therefore be applied to nonstandard model organisms in vertebrates as well as in plants, without the need for extensive parameter training. In addition to predicting genes in large-scale genomic sequences, the program may be useful to validate gene structure annotations from databases. To this end, SGP-1 output also contains comparisons between predicted and annotated gene structures in HTML format. The program can be accessed via a Web server at http://soft.ice.mpg.de/sgp-1. The source code, written in ANSI C, is available on request from the authors.

Improving gene annotation using peptide mass spectrometry

Annotation of protein-coding genes is a key goal of genome sequencing projects. In spite of tremendous recent advances in computational gene finding, comprehensive annotation remains a challenge. Peptide mass spectrometry is a powerful tool for researching the dynamic proteome and suggests an attractive approach to discover and validate protein-coding genes. We present algorithms to construct and efficiently search spectra against a genomic database, with no prior knowledge of encoded proteins. By searching a corpus of 18.5 million tandem mass spectra (MS/MS) from human proteomic samples, we validate 39,000 exons and 11,000 introns at the level of translation. We present translation-level evidence for novel or extended exons in 16 genes, confirm translation of 224 hypothetical proteins, and discover or confirm over 40 alternative splicing events. Polymorphisms are efficiently encoded in our database, allowing us to observe variant alleles for 308 coding SNPs. Finally, we demonstrate the use of mass spectrometry to improve automated gene prediction, adding 800 correct exons to our predictions using a simple rescoring strategy. Our results demonstrate that proteomic profiling should play a role in any genome sequencing project.

Gene finding in the chicken genome

Background: Despite the continuous production of genome sequence for a number of organisms,reliable, comprehensive, and cost effective gene prediction remains problematic. This is particularlytrue for genomes for which there is not a large collection of known gene sequences, such as therecently published chicken genome. We used the chicken sequence to test comparative andhomology-based gene-finding methods followed by experimental validation as an effective genomeannotation method.Results: We performed experimental evaluation by RT-PCR of three different computational genefinders, Ensembl, SGP2 and TWINSCAN, applied to the chicken genome. A Venn diagram wascomputed and each component of it was evaluated. The results showed that de novo comparativemethods can identify up to about 700 chicken genes with no previous evidence of expression, andcan correctly extend about 40% of homology-based predictions at the 5' end.Conclusions: De novo comparative gene prediction followed by experimental verification iseffective at enhancing the annotation of the newly sequenced genomes provided by standardhomology-based methods.

Disease-Gene Association Using Genetic Programming

As a result of mutation in genes, which is a simple change in our DNA, we will have undesirable phenotypes which are known as genetic diseases or disorders. These small changes, which happen frequently, can have extreme results. Understanding and identifying these changes and associating these mutated genes with genetic diseases can play an important role in our health, by making us able to find better diagnosis and therapeutic strategies for these genetic diseases. As a result of years of experiments, there is a vast amount of data regarding human genome and different genetic diseases that they still need to be processed properly to extract useful information. This work is an effort to analyze some useful datasets and to apply different techniques to associate genes with genetic diseases. Two genetic diseases were studied here: Parkinson’s disease and breast cancer. Using genetic programming, we analyzed the complex network around known disease genes of the aforementioned diseases, and based on that we generated a ranking for genes, based on their relevance to these diseases. In order to generate these rankings, centrality measures of all nodes in the complex network surrounding the known disease genes of the given genetic disease were calculated. Using genetic programming, all the nodes were assigned scores based on the similarity of their centrality measures to those of the known disease genes. Obtained results showed that this method is successful at finding these patterns in centrality measures and the highly ranked genes are worthy as good candidate disease genes for being studied. Using standard benchmark tests, we tested our approach against ENDEAVOUR and CIPHER - two well known disease gene ranking frameworks - and we obtained comparable results.

Identification of protein coding regions using the modified Gabor-wavelet transform

An important topic in genomic sequence analysis is the identification of protein coding regions. In this context, several coding DNA model-independent methods based on the occurrence of specific patterns of nucleotides at coding regions have been proposed. Nonetheless, these methods have not been completely suitable due to their dependence on an empirically predefined window length required for a local analysis of a DNA region. We introduce a method based on a modified Gabor-wavelet transform (MGWT) for the identification of protein coding regions. This novel transform is tuned to analyze periodic signal components and presents the advantage of being independent of the window length. We compared the performance of the MGWT with other methods by using eukaryote data sets. The results show that MGWT outperforms all assessed model-independent methods with respect to identification accuracy. These results indicate that the source of at least part of the identification errors produced by the previous methods is the fixed working scale. The new method not only avoids this source of errors but also makes a tool available for detailed exploration of the nucleotide occurrence.

Isolation and characterization of a cone snail protease with homology to CRISP proteins of the pathogenesis-related protein superfamily

The pathogenesis-related (PR) protein superfamily is widely distributed in the animal, plant, and fungal kingdoms and is implicated in human brain tumor growth and plant pathogenesis. The precise biological activity of PR proteins, however, has remained elusive. Here we report the characterization, cloning and structural homology modeling of Tex31 from the venom duct of Conus textile. Tex31 was isolated to >95% purity by activity-guided fractionation using a para-nitroanilide substrate based on the putative cleavage site residues found in the propeptide precursor of conotoxin TxVIA. Tex31 requires four residues including a leucine N-terminal of the cleavage site for efficient substrate processing. The sequence of Tex31 was determined using two degenerate PCR primers designed from N-terminal and tryptic digest Edman sequences. A BLAST search revealed that Tex31 was a member of the PR protein superfamily and most closely related to the CRISP family of mammalian proteins that have a cysteine-rich C-terminal tail. A homology model constructed from two PR proteins revealed that the likely catalytic residues in Tex31 fall within a structurally conserved domain found in PR proteins. Thus, it is possible that other PR proteins may also be substrate-specific proteases.

Nineteen additional unpredicted transcripts from human chromosome 21.

The identification of all human chromosome 21 (HC21) genes is a necessary step in understanding the molecular pathogenesis of trisomy 21 (Down syndrome). The first analysis of the sequence of 21q included 127 previously characterized genes and predicted an additional 98 novel anonymous genes. Recently we evaluated the quality of this annotation by characterizing a set of HC21 open reading frames (C21orfs) identified by mapping spliced expressed sequence tags (ESTs) and predicted genes (PREDs), identified only in silico. This study underscored the limitations of in silico-only gene prediction, as many PREDs were incorrectly predicted. To refine the HC21 annotation, we have developed a reliable algorithm to extract and stringently map sequences that contain bona fide 3' transcript ends to the genome. We then created a specific 21q graphical display allowing an integrated view of the data that incorporates new ESTs as well as features such as CpG islands, repeats, and gene predictions. Using these tools we identified 27 new putative genes. To validate these, we sequenced previously cloned cDNAs and carried out RT-PCR, 5'- and 3'-RACE procedures, and comparative mapping. These approaches substantiated 19 new transcripts, thus increasing the HC21 gene count by 9.5%. These transcripts were likely not previously identified because they are small and encode small proteins. We also identified four transcriptional units that are spliced but contain no obvious open reading frame. The HC21 data presented here further emphasize that current gene prediction algorithms miss a substantial number of transcripts that nevertheless can be identified using a combination of experimental approaches and multiple refined algorithms.

The Aedes aegypti larval transcriptome: a comparative perspective with emphasis on trypsins and the domain structure of peritrophins

The genome sequence of Aedes aegypti was recently reported. A significant amount of Expressed Sequence Tags (ESTs) were sequenced to aid in the gene prediction process. In the present work we describe an integrated analysis of the genomic and EST data, focusing on genes with preferential expression in larvae (LG), adults (AG) and in both stages (SG). A total of 913 genes (5.4% of the transcript complement) are LG, including ion transporters and cuticle proteins that are important for ion homeostasis and defense. From a starting set of 245 genes encoding the trypsin domain, we identified 66 putative LG, AG, and SG trypsins by manual curation. Phylogenetic analyses showed that AG trypsins are divergent from their larval counterparts (LG), grouping with blood-induced trypsins from Anopheles gambiae and Simulium vittatum. These results support the hypothesis that blood-feeding arose only once, in the ancestral Culicomorpha. Peritrophins are proteins that interlock chitin fibrils to form the peritrophic membrane (PM) that compartmentalizes the food in the midgut. These proteins are recognized by having chitin-binding domains with 6 conserved Cys and may also present mucin-like domains (regions expected to be highly O-glycosylated). PM may be formed by a ring of cells (type 2, seen in Ae. aegypti larvae and Drosophila melanogaster) or by most midgut cells (type 1, found in Ae. aegypti adult and Tribolium castaneum). LG and D. melanogaster peritrophins have more complex domain structures than AG and T. castaneum peritrophins. Furthermore, mucin-like domains of peritrophins from T. castaneum (feeding on rough food) are lengthier than those of adult Ae. aegypti (blood-feeding). This suggests, for the first time, that type 1 and type 2 PM may have variable molecular architectures determined by different peritrophins and/or ancillary proteins, which may be partly modulated by diet.

Predição de promotores de Bacillus subtilis usando técnicas de aprendizado de máquina

One of the most important goals of bioinformatics is the ability to identify genes in uncharacterized DNA sequences on world wide database. Gene expression on prokaryotes initiates when the RNA-polymerase enzyme interacts with DNA regions called promoters. In these regions are located the main regulatory elements of the transcription process. Despite the improvement of in vitro techniques for molecular biology analysis, characterizing and identifying a great number of promoters on a genome is a complex task. Nevertheless, the main drawback is the absence of a large set of promoters to identify conserved patterns among the species. Hence, a in silico method to predict them on any species is a challenge. Improved promoter prediction methods can be one step towards developing more reliable ab initio gene prediction methods. In this work, we present an empirical comparison of Machine Learning (ML) techniques such as Na¨ýve Bayes, Decision Trees, Support Vector Machines and Neural Networks, Voted Perceptron, PART, k-NN and and ensemble approaches (Bagging and Boosting) to the task of predicting Bacillus subtilis. In order to do so, we first built two data set of promoter and nonpromoter sequences for B. subtilis and a hybrid one. In order to evaluate of ML methods a cross-validation procedure is applied. Good results were obtained with methods of ML like SVM and Naïve Bayes using B. subtilis. However, we have not reached good results on hybrid database

SpliceDB: database of canonical and non-canonical mammalian splice sites

A database (SpliceDB) of known mammalian splice site sequences has been developed. We extracted 43 337 splice pairs from mammalian divisions of the gene-centered Infogene database, including sites from incomplete or alternatively spliced genes. Known EST sequences supported 22 815 of them. After discarding sequences with putative errors and ambiguous location of splice junctions the verified dataset includes 22 489 entries. Of these, 98.71% contain canonical GT–AG junctions (22 199 entries) and 0.56% have non-canonical GC–AG splice site pairs. The remainder (0.73%) occurs in a lot of small groups (with a maximum size of 0.05%). We especially studied non-canonical splice sites, which comprise 3.73% of GenBank annotated splice pairs. EST alignments allowed us to verify only the exonic part of splice sites. To check the conservative dinucleotides we compared sequences of human non-canonical splice sites with sequences from the high throughput genome sequencing project (HTG). Out of 171 human non-canonical and EST-supported splice pairs, 156 (91.23%) had a clear match in the human HTG. They can be classified after sequence analysis as: 79 GC–AG pairs (of which one was an error that corrected to GC–AG), 61 errors corrected to GT–AG canonical pairs, six AT–AC pairs (of which two were errors corrected to AT–AC), one case was produced from a non-existent intron, seven cases were found in HTG that were deposited to GenBank and finally there were only two other cases left of supported non-canonical splice pairs. The information about verified splice site sequences for canonical and non-canonical sites is presented in SpliceDB with the supporting evidence. We also built weight matrices for the major splice groups, which can be incorporated into gene prediction programs. SpliceDB is available at the computational genomic Web server of the Sanger Centre: http://genomic.sanger.ac.uk/spldb/SpliceDB.html and at http://www.softberry.com/spldb/SpliceDB.html.

A human protein containing multiple types of protease-inhibitory modules

By using sensitive homology-search and gene-finding programs, we have found that a genomic region from the tip of the short arm of human chromosome 16 (16p13.3) encodes a putative secreted protein consisting of a domain related to the whey acidic protein (WAP) domain, a domain homologous with follistatin modules of the Kazal-domain family (FS module), an immunoglobulin-related domain (Ig domain), two tandem domains related to Kunitz-type protease inhibitor modules (KU domains), and a domain belonging to the recently defined NTR-module family (NTR domain). The gene encoding these WAP, FS, Ig, KU, and NTR modules (hereafter referred to as the WFIKKN gene) is intron-depleted—its single 1,157-bp intron splits the WAP module. The validity of our gene prediction was confirmed by sequencing a WFIKKN cDNA cloned from a lung cDNA library. Studies on the tissue-expression pattern of the WFIKKN gene have shown that the gene is expressed primarily in pancreas, kidney, liver, placenta, and lung. As to the function of the WFIKKN protein, it is noteworthy that it contains FS, WAP, and KU modules, i.e., three different module types homologous with domains frequently involved in inhibition of serine proteases. The protein also contains an NTR module, a domain type implicated in inhibition of zinc metalloproteinases of the metzincin family. On the basis of its intriguing homologies, we suggest that the WFIKKN protein is a multivalent protease inhibitor that may control the action of multiple types of serine proteases as well as metalloproteinase(s).