Background Prediction of protein structures is one of the fundamental challenges in biology today. To fully understand how well different prediction methods perform, it is necessary to use measures that evaluate their performance. Every two years, starting in 1994, the CASP (Critical Assessment of protein Structure Prediction) process has been organized to evaluate the ability of different predictors to blindly predict the structure of proteins. To capture different features of the models, several measures have been developed during the CASP processes. However, these measures have not been examined in detail before. In an attempt to develop fully automatic measures that can be used in CASP, as well as in other type of benchmarking experiments, we have compared twenty-one measures. These measures include the measures used in CASP3 and CASP2 as well as have measures introduced later. We have studied their ability to distinguish between the better and worse models submitted to CASP3 and the correlation between them. Results Using a small set of 1340 models for 23 different targets we show that most methods correlate with each other. Most pairs of measures show a correlation coefficient of about 0.5. The correlation is slightly higher for measures of similar types. We found that a significant problem when developing automatic measures is how to deal with proteins of different length. Also the comparisons between different measures is complicated as many measures are dependent on the size of the target. We show that the manual assessment can be reproduced to about 70% using automatic measures. Alignment independent measures, detects slightly more of the models with the correct fold, while alignment dependent measures agree better when selecting the best models for each target. Finally we show that using automatic measures would, to a large extent, reproduce the assessors ranking of the predictors at CASP3. Conclusions We show that given a sufficient number of targets the manual and automatic measures would have given almost identical results at CASP3. If the intent is to reproduce the type of scoring done by the manual assessor in in CASP3, the best approach might be to use a combination of alignment independent and alignment dependent measures, as used in several recent studies.

Background: Correlations between genome composition (in terms of GC content) and usage of particular codons and amino acids have been widely reported, but poorly explained. We show here that a simple model of processes acting at the nucleotide level explains codon usage across a large sample of species (311 bacteria, 28 archaea and 257 eukaryotes). The model quantitatively predicts responses (slope and intercept of the regression line on genome GC content) of individual codons and amino acids to genome composition. Results: Codons respond to genome composition on the basis of their GC content relative to their synonyms (explaining 71-87% of the variance in response among the different codons, depending on measure). Amino-acid responses are determined by the mean GC content of their codons (explaining 71-79% of the variance). Similar trends hold for genes within a genome. Position-dependent selection for error minimization explains why individual bases respond differently to directional mutation pressure. Conclusions: Our model suggests that GC content drives codon usage (rather than the converse). It unifies a large body of empirical evidence concerning relationships between GC content and amino-acid or codon usage in disparate systems. The relationship between GC content and codon and amino-acid usage is ahistorical; it is replicated independently in the three domains of living organisms, reinforcing the idea that genes and genomes at mutation/selection equilibrium reproduce a unique relationship between nucleic acid and protein composition. Thus, the model may be useful in predicting amino-acid or nucleotide sequences in poorly characterized taxa.

Background The recent draft assembly of the human genome provides a unified basis for describing genomic structure and function. The draft is sufficiently accurate to provide useful annotation, enabling direct observations of previously inferred biological phenomena. Results We report here a functionally annotated human gene index placed directly on the genome. The index is based on the integration of public transcript, protein, and mapping information, supplemented with computational prediction. We describe numerous global features of the genome and examine the relationship of various genetic maps with the assembly. In addition, initial sequence analysis reveals highly ordered chromosomal landscapes associated with paralogous gene clusters and distinct functional compartments. Finally, these annotation data were synthesized to produce observations of gene density and number that accord well with historical estimates. Such a global approach had previously been described only for chromosomes 21 and 22, which together account for 2.2% of the genome. Conclusions We estimate that the genome contains 65,000-75,000 transcriptional units, with exon sequences comprising 4%. The creation of a comprehensive gene index requires the synthesis of all available computational and experimental evidence.

Background Sequestration of transcription factors in the membrane is emerging as an important mechanism for the regulation of gene expression. A handful of membrane-spanning transcription factors has been previously identified whose access to the nucleus is regulated by proteolytic cleavage from the membrane. To investigate the existence of other transmembrane transcription factors, we analyzed computationally all proteins in SWISS-PROT/TrEMBL for the combined presence of a DNA-binding domain and a transmembrane segment. Results Using Pfam hidden Markov models and four transmembrane-prediction programs, we identified with high confidence 76 membrane-spanning transcription factors in SWISS-PROT/TrEMBL. Analysis of the distribution of two proteins predicted by our method, MTJ1 and DMRT2, confirmed their localization to intracellular membrane compartments. Furthermore, elimination of the predicted transmembrane segment led to nuclear localization for each of these proteins. Conclusions Our analysis uncovered a wealth of predicted membrane-spanning transcription factors that are structurally and taxonomically diverse, 56 of which lack experimental annotation. Seventy-five of the proteins are modular in structure, suggesting that a single proteolysis may be sufficient to liberate a DNA-binding domain from the membrane. This study provides grounds for investigations into the stimuli and mechanisms that release this intriguing class of transcription factors from membranes.

Background Sites in DNA that bind regulatory proteins can be detected computationally in various ways. Pattern discovery methods analyze collections of genes suspected to be co-regulated on the evidence, for example, of clustering of transcriptome data. Pattern searching methods use sequences with known binding sites to find other genes regulated by a given protein. Such computational methods are important strategies in the discovery and elaboration of regulatory networks and can provide the experimental biologist with a precise prediction of a binding site or identify a gene as a member of a set of co-regulated genes (a regulon). As more variations on such methods are published, however, thorough evaluation is necessary, as performance may differ depending on the conditions of use. Detailed evaluation also helps to improve and understand the behavior of the different methods and computational strategies. Results We used a collection of 86 regulons from Escherichia coli as datasets to evaluate two methods for pattern discovery and pattern searching: dyad analysis/dyad sweeping using the program Dyad-analysis, and multiple alignment using the programs Consensus/Patser. Clearly defined statistical parameters are used to evaluate the two methods in different situations. We placed particular emphasis on minimizing the rate of false positives. Conclusions As a general rule, sensors obtained from experimentally reported binding sites in DNA frequently locate true sites as the highest-scoring sequences within a given upstream region, especially using Consensus/Patser. Pattern discovery is still an unsolved problem, although in the cases where Dyad-analysis finds significant dyads (around 50%), these frequently correspond to true binding sites. With more robust methods, regulatory predictions could help identify the function of unknown genes.

Background Earlier methods for detecting major genes responsible for a quantitative trait rely critically upon a well-structured pedigree in which the segregation pattern of genes exactly follow Mendelian inheritance laws. However, for many outcrossing species, such pedigrees are not available and genes also display population properties. Results In this paper, a hierarchical statistical model is proposed to monitor the existence of a major gene based on its segregation and transmission across two successive generations. The model is implemented with an EM algorithm to provide maximum likelihood estimates for genetic parameters of the major locus. This new method is successfully applied to identify an additive gene having a large effect on stem height growth of aspen trees. The estimates of population genetic parameters for this major gene can be generalized to the original breeding population from which the parents were sampled. A simulation study is presented to evaluate finite sample properties of the model. Conclusions A hierarchical model was derived for detecting major genes affecting a quantitative trait based on progeny tests of outcrossing species. The new model takes into account the population genetic properties of genes and is expected to enhance the accuracy, precision and power of gene detection.

Background We have constructed Bayesian prior-based, amino-acid sequence profiles for the complete yeast mitochondrial proteome and used them to develop methods for identifying and characterizing the context of protein mutations that give rise to human mitochondrial diseases. (Bayesian priors are conditional probabilities that allow the estimation of the likelihood of an event - such as an amino-acid substitution - on the basis of prior occurrences of similar events.) Because these profiles can assemble sets of taxonomically very diverse homologs, they enable identification of the structurally and/or functionally most critical sites in the proteins on the basis of the degree of sequence conservation. These profiles can also find distant homologs with determined three-dimensional structures that aid in the interpretation of effects of missense mutations. Results This survey reports such an analysis for 15 missense mutations, one insertion and three deletions involved in Leber's hereditary optic neuropathy, Leigh syndrome, mitochondrial neurogastrointestinal encephalomyopathy, Mohr-Tranebjaerg syndrome, iron-storage disorders related to Friedreich's ataxia, and hereditary spastic paraplegia. We present structural correlations for seven of the mutations. Conclusions Of the 19 mutations analyzed, 14 involved changes in very highly conserved parts of the affected proteins. Five out of seven structural correlations provided reasonable explanations for the malfunctions. As additional genetic and structural data become available, this methodology can be extended. It has the potential for assisting in identifying new disease-related genes. Furthermore, profiles with structural homologs can generate mechanistic hypotheses concerning the underlying biochemical processes - and why they break down as a result of the mutations.

Background: Classic methods of identifying genes involved in neural function include the laborious process of behavioral screening of mutagenized flies and then rescreening candidate lines for pleiotropic effects due to developmental defects. To accelerate the molecular analysis of brain function in Drosophila we constructed a cDNA library exclusively from adult brains. Our goal was to begin to develop a catalog of transcripts expressed in the brain. These transcripts are expected to contain a higher proportion of clones that are involved in neuronal function. Results: The library contains approximately 6.75 million independent clones. From our initial characterization of 271 randomly chosen clones, we expect that approximately 11% of the clones in this library will identify transcribed sequences not found in expressed sequence tag databases. Furthermore, 15% of these 271 clones are not among the 13,601 predicted Drosophila genes. Conclusions: Our analysis of this unique Drosophila brain library suggests that the number of genes may be underestimated in this organism. This work complements the Drosophila genome project by providing information that facilitates more complete annotation of the genomic sequence. This library should be a useful resource that will help in determining how basic brain functions operate at the molecular level.

Background Classification of the electrocardiogram using Neural Networks has become a widely used method in recent years. The efficiency of these classifiers depends upon a number of factors including network training. Unfortunately, there is a shortage of evidence available to enable specific design choices to be made and as a consequence, many designs are made on the basis of trial and error. In this study we develop prediction models to indicate the point at which training should stop for Neural Network based Electrocardiogram classifiers in order to ensure maximum generalisation. Methods Two prediction models have been presented; one based on Neural Networks and the other on Genetic Programming. The inputs to the models were 5 variable training parameters and the output indicated the point at which training should stop. Training and testing of the models was based on the results from 44 previously developed bi-group Neural Network classifiers, discriminating between Anterior Myocardial Infarction and normal patients. Results Our results show that both approaches provide close fits to the training data; p = 0.627 and p = 0.304 for the Neural Network and Genetic Programming methods respectively. For unseen data, the Neural Network exhibited no significant differences between actual and predicted outputs (p = 0.306) while the Genetic Programming method showed a marginally significant difference (p = 0.047). Conclusions The approaches provide reverse engineering solutions to the development of Neural Network based Electrocardiogram classifiers. That is given the network design and architecture, an indication can be given as to when training should stop to obtain maximum network generalisation.

Randomized trials and observational studies, such as case-control studies, are often seen as opposing approaches. However, in many instances results obtained by different designs may complement each other. For instance, case-control studies on aetiology of disease may help to give the direction of future trials. In this commentary, the author discusses the purpose of randomization and observation, and under which conditions one design may be preferred to another. Randomization is useful to combat 'confounding by indication', and is therefore the design of choice for most therapeutic trials. When this confounding is not an issue, as in studies of genetic risk factors or side-effects, then case-control studies are preferred.