FlexTaxD: flexible modification of taxonomy databases for improved sequence classification.

SUMMARY: The Flexible Taxonomy Database framework provides a method for modification and merging official and custom taxonomic databases to create improved databases. Using such databases will increase accuracy and precision of existing methods to classify sequence reads. AVAILABILITY AND IMPLEMENTATION: Source code is freely available at https://github.com/FOI-Bioinformatics/flextaxd and installable through Bioconda.

Ultrasensitive Detection of Bacillus anthracis by Real-Time PCR Targeting a Polymorphism in Multi-Copy 16S rRNA Genes and Their Transcripts.

The anthrax pathogen Bacillus anthracis poses a significant threat to human health. Identification of B. anthracis is challenging because of the bacterium's close genetic relationship to other Bacillus cereus group species. Thus, molecular detection is founded on species-specific PCR targeting single-copy genes. Here, we validated a previously recognized multi-copy target, a species-specific single nucleotide polymorphism (SNP) present in 2-5 copies in every B. anthracis genome analyzed. For this, a hydrolysis probe-based real-time PCR assay was developed and rigorously tested. The assay was specific as only B. anthracis DNA yielded positive results, was linear over 9 log10 units, and was sensitive with a limit of detection (LoD) of 2.9 copies/reaction. Though not exhibiting a lower LoD than established single-copy PCR targets (dhp61 or PL3), the higher copy number of the B. anthracis-specific 16S rRNA gene alleles afforded ≤2 unit lower threshold (Ct) values. To push the detection limit even further, the assay was adapted for reverse transcription PCR on 16S rRNA transcripts. This RT-PCR assay was also linear over 9 log10 units and was sensitive with an LoD of 6.3 copies/reaction. In a dilution series of experiments, the 16S RT-PCR assay achieved a thousand-fold higher sensitivity than the DNA-targeting assays. For molecular diagnostics, we recommend a real-time RT-PCR assay variant in which both DNA and RNA serve as templates (thus, no requirement for DNase treatment). This can at least provide results equaling the DNA-based implementation if no RNA is present but is superior even at the lowest residual rRNA concentrations.

Whole genome sequencing and phylogenetic classification of Tunisian SARS-CoV-2 strains from patients of the Military Hospital in Tunis.

In the present work, two complete genome sequences of SARS-CoV-2 were obtained from nasal swab samples of Tunisian SARS-CoV-2 PCR-positive patients using nanopore sequencing. The virus genomes of two of the patients examined, a Tunisian soldier returning from a mission in Morocco and a member of another Tunisian family, showed significant differences in analyses of the total genome and single nucleotide polymorphisms (SNPs). Phylogenetic relationships with known SARS-CoV-2 genomes in the African region, some European and Middle Eastern countries and initial epidemiological conclusions indicate that the introduction of SARS-CoV-2 into Tunisia from two independent sources was travel-related.

A headache with surprising outcome: first case of brucellosis caused by Brucella suis biovar 1 in Germany.

In July 2018, brucellosis was diagnosed in a German patient without a travel history to regions endemic for Brucella. Microbiological analysis, including whole-genome sequencing, revealed Brucella suis biovar 1 as the etiologic agent. Core-genome-based multilocus sequence-typing analysis placed the isolate in close proximity to strains originating from Argentina. Notably, despite a strong IgM response, the patient did not develop Brucella-specific IgG antibodies during infection. Here, we describe the clinical course of infection, the extensive epidemiological investigations, and discuss possible routes of transmission.

eggNOG 4.5: a hierarchical orthology framework with improved functional annotations for eukaryotic, prokaryotic and viral sequences.

eggNOG is a public resource that provides Orthologous Groups (OGs) of proteins at different taxonomic levels, each with integrated and summarized functional annotations. Developments since the latest public release include changes to the algorithm for creating OGs across taxonomic levels, making nested groups hierarchically consistent. This allows for a better propagation of functional terms across nested OGs and led to the novel annotation of 95 890 previously uncharacterized OGs, increasing overall annotation coverage from 67% to 72%. The functional annotations of OGs have been expanded to also provide Gene Ontology terms, KEGG pathways and SMART/Pfam domains for each group. Moreover, eggNOG now provides pairwise orthology relationships within OGs based on analysis of phylogenetic trees. We have also incorporated a framework for quickly mapping novel sequences to OGs based on precomputed HMM profiles. Finally, eggNOG version 4.5 incorporates a novel data set spanning 2605 viral OGs, covering 5228 proteins from 352 viral proteomes. All data are accessible for bulk downloading, as a web-service, and through a completely redesigned web interface. The new access points provide faster searches and a number of new browsing and visualization capabilities, facilitating the needs of both experts and less experienced users. eggNOG v4.5 is available at http://eggnog.embl.de.

HoPaCI-DB: host-Pseudomonas and Coxiella interaction database.

Bacterial infectious diseases are the result of multifactorial processes affected by the interplay between virulence factors and host targets. The host-Pseudomonas and Coxiella interaction database (HoPaCI-DB) is a publicly available manually curated integrative database (http://mips.helmholtz-muenchen.de/HoPaCI/) of host-pathogen interaction data from Pseudomonas aeruginosa and Coxiella burnetii. The resource provides structured information on 3585 experimentally validated interactions between molecules, bioprocesses and cellular structures extracted from the scientific literature. Systematic annotation and interactive graphical representation of disease networks make HoPaCI-DB a versatile knowledge base for biologists and network biology approaches.

Molecular analysis of Coxiella burnetii in Germany reveals evolution of unique clonal clusters.

The causative agent of Q fever, Coxiella burnetii, is a query agent occurring naturally all over the world. We studied 104 German Coxiella burnetii strains/DNA samples obtained between 1969 and 2011 using a 14 microsatellite marker Multiple-locus variable-number of tandem repeat (VNTR) analysis (MLVA) technique. We were able to divide our collection into 32 different genotypes clustered into four major groups (A-D). Two of these (A and C) formed predominant clonal complexes that covered 97% of all studied samples. Group C consisted exclusively of cattle-associated isolates/DNA specimens, while group A comprised all other affected species including all sheep-derived strains/DNA samples. Within this second cluster, two major genotypes (A1, A2) were identified. Genotype A2 occurred in strains isolated from ewes in northern and central Germany, whereas genotype A1 was found in most areas of Germany. MLVA analysis of C. burnetii strains from neighbouring countries revealed a close relationship to German strains. We thus hypothesize that there is a western and central European cluster of C. burnetii. We identified predominant genotypes related to relevant host species and geographic regions which is in line with findings of the Dutch Q fever outbreak (2007-2010). Furthermore three of our analyzed German strains are closely related to the Dutch outbreak clone. These findings support the theory of predominant genotypes in the context of regional outbreaks. Our results show that a combination of 8 MLVA markers provides the highest discriminatory power for attributing C. burnetii isolates to genotypes. For future epidemiological studies we propose the use of three MLVA markers for easy and rapid classification of C. burnetii into 4 main clusters.

Sequence-structure relationships in yeast mRNAs.

It is generally accepted that functionally important RNA structure is more conserved than sequence due to compensatory mutations that may alter the sequence without disrupting the structure. For small RNA molecules sequence-structure relationships are relatively well understood. However, structural bioinformatics of mRNAs is still in its infancy due to a virtual absence of experimental data. This report presents the first quantitative assessment of sequence-structure divergence in the coding regions of mRNA molecules based on recently published transcriptome-wide experimental determination of their base paring patterns. Structural resemblance in paralogous mRNA pairs quickly drops as sequence identity decreases from 100% to 85-90%. Structures of mRNAs sharing sequence identity below roughly 85% are essentially uncorrelated. This outcome is in dramatic contrast to small functional non-coding RNAs where sequence and structure divergence are correlated at very low levels of sequence similarity. The fact that very similar mRNA sequences can have vastly different secondary structures may imply that the particular global shape of base paired elements in coding regions does not play a major role in modulating gene expression and translation efficiency. Apparently, the need to maintain stable three-dimensional structures of encoded proteins places a much higher evolutionary pressure on mRNA sequences than on their RNA structures.

FGDB: revisiting the genome annotation of the plant pathogen Fusarium graminearum.

The MIPS Fusarium graminearum Genome Database (FGDB) was established as a comprehensive genome database on one of the most devastating fungal plant pathogens of wheat, barley and maize. The current version of FGDB v3.1 provides information on the full manually revised gene set based on the Broad Institute assembly FG3 genome sequence. The results of gene prediction tools were integrated with the help of comparative data on related species to result in a set of 13.718 annotated protein coding genes. This rigorous approach involved adding or modifying gene models and represents a coding sequence gold standard for the genus Fusarium. The gene loci improvements results in 2461 genes which either are new or have different structures compared to the Broad Institute assembly 3 gene set. Moreover the database serves as a convenient entry point to explore expression data results and to obtain information on the Affymetrix GeneChip probe sets. The resource is accessible on http://mips.gsf.de/genre/proj/FGDB/.

MIPS: curated databases and comprehensive secondary data resources in 2010.

The Munich Information Center for Protein Sequences (MIPS at the Helmholtz Center for Environmental Health, Neuherberg, Germany) has many years of experience in providing annotated collections of biological data. Selected data sets of high relevance, such as model genomes, are subjected to careful manual curation, while the bulk of high-throughput data is annotated by automatic means. High-quality reference resources developed in the past and still actively maintained include Saccharomyces cerevisiae, Neurospora crassa and Arabidopsis thaliana genome databases as well as several protein interaction data sets (MPACT, MPPI and CORUM). More recent projects are PhenomiR, the database on microRNA-related phenotypes, and MIPS PlantsDB for integrative and comparative plant genome research. The interlinked resources SIMAP and PEDANT provide homology relationships as well as up-to-date and consistent annotation for 38,000,000 protein sequences. PPLIPS and CCancer are versatile tools for proteomics and functional genomics interfacing to a database of compilations from gene lists extracted from literature. A novel literature-mining tool, EXCERBT, gives access to structured information on classified relations between genes, proteins, phenotypes and diseases extracted from Medline abstracts by semantic analysis. All databases described here, as well as the detailed descriptions of our projects can be accessed through the MIPS WWW server (http://mips.helmholtz-muenchen.de).

In vitro evaluation of the effect of mutations in primer binding sites on detection of SARS-CoV-2 by RT-qPCR.

A number of RT-qPCR assays for the detection of SARS-CoV-2 have been published and are listed by the WHO as recommended assays. Furthermore, numerous commercial assays with undisclosed primer and probe sequences are on the market. As the SARS-CoV-2 pandemic progresses, the virus accrues mutations, which in some cases - as seen with the B.1.1.7 variant - can outperform and push back other strains of SARS-CoV-2. If mutations occur in primer or probe binding sites, this can impact RT-qPCR results and impede SARS-CoV-2 diagnostics. Here we tested the effect of primer mismatches on RT-qPCR performance in vitro using synthetic mismatch in vitro transcripts. The effects of the mismatches ranged from a shift in ct values from -0.13 to +7.61. Crucially, we found that a mismatch in the forward primer has a more detrimental effect for PCR performance than a mismatch in the reverse primer. Furthermore, we compared the performance of the original Charité RdRP primer set, which has several ambiguities, with a primer version without ambiguities and found that without ambiguities the ct values are ca. 3 ct lower. Finally, we investigated the shift in ct values observed with the Seegene Allplex kit with the B.1.1.7 SARS-CoV-2 variant and found a three-nucleotide mismatch in the forward primer of the N target.

Machine Learning Algorithms for Classification of MALDI-TOF MS Spectra from Phylogenetically Closely Related Species Brucella melitensis, Brucella abortus and Brucella suis.

(1) Background: MALDI-TOF mass spectrometry (MS) is the gold standard for microbial fingerprinting, however, for phylogenetically closely related species, the resolution power drops down to the genus level. In this study, we analyzed MALDI-TOF spectra from 44 strains of B. melitensis, B. suis and B. abortus to identify the optimal classification method within popular supervised and unsupervised machine learning (ML) algorithms. (2) Methods: A consensus feature selection strategy was applied to pinpoint from among the 500 MS features those that yielded the best ML model and that may play a role in species differentiation. Unsupervised k-means and hierarchical agglomerative clustering were evaluated using the silhouette coefficient, while the supervised classifiers Random Forest, Support Vector Machine, Neural Network, and Multinomial Logistic Regression were explored in a fine-tuning manner using nested k-fold cross validation (CV) with a feature reduction step between the two CV loops. (3) Results: Sixteen differentially expressed peaks were identified and used to feed ML classifiers. Unsupervised and optimized supervised models displayed excellent predictive performances with 100% accuracy. The suitability of the consensus feature selection strategy for learning system accuracy was shown. (4) Conclusion: A meaningful ML approach is here introduced, to enhance Brucella spp. classification using MALDI-TOF MS data.

In-Depth Analysis of Bacillus anthracis 16S rRNA Genes and Transcripts Reveals Intra- and Intergenomic Diversity and Facilitates Anthrax Detection.

Analysis of 16S rRNA (rRNA) genes provides a central means of taxonomic classification of bacterial species. Based on presumed sequence identity among species of the Bacillus cereus sensu lato group, the 16S rRNA genes of B. anthracis have been considered unsuitable for diagnosis of the anthrax pathogen. With the recent identification of a single nucleotide polymorphism in some 16S rRNA gene copies, specific identification of B. anthracis becomes feasible. Here, we designed and evaluated a set of in situ, in vitro, and in silico assays to assess the unknown 16S state of B. anthracis from different perspectives. Using a combination of digital PCR, fluorescence in situ hybridization, long-read genome sequencing, and bioinformatics, we were able to detect and quantify a unique 16S rRNA gene allele of B. anthracis (16S-BA-allele). This allele was found in all available B. anthracis genomes and may facilitate differentiation of the pathogen from any close relative. Bioinformatics analysis of 959 B. anthracis SRA data sets inferred that abundances and genomic arrangements of the 16S-BA-allele and the entire rRNA operon copy numbers differ considerably between strains. Expression ratios of 16S-BA-alleles were proportional to the respective genomic allele copy numbers. The findings and experimental tools presented here provide detailed insights into the intra- and intergenomic diversity of 16S rRNA genes and may pave the way for improved identification of B. anthracis and other pathogens with diverse rRNA operons. IMPORTANCE For severe infectious diseases, precise pathogen detection is crucial for antibiotic therapy and patient survival. Identification of Bacillus anthracis, the causative agent of the zoonosis anthrax, can be challenging when querying specific nucleotide sequences such as in small subunit rRNA (16S rRNA) genes, which are commonly used for typing of bacteria. This study analyzed on a broad genomic scale a cryptic and hitherto underappreciated allelic variant of the bacterium's 16S rRNA genes and their transcripts using a set of in situ, in vitro, and in silico assays and found significant intra- and intergenomic heterogeneity in the distribution of the allele and overall rRNA operon copy numbers. This allelic variation was uniquely species specific, which enabled sensitive pathogen detection on both DNA and transcript levels. The methodology used here is likely also applicable to other pathogens that are otherwise difficult to discriminate from their less harmful relatives.

PEDANT covers all complete RefSeq genomes.

The PEDANT genome database provides exhaustive annotation of nearly 3000 publicly available eukaryotic, eubacterial, archaeal and viral genomes with more than 4.5 million proteins by a broad set of bioinformatics algorithms. In particular, all completely sequenced genomes from the NCBI's Reference Sequence collection (RefSeq) are covered. The PEDANT processing pipeline has been sped up by an order of magnitude through the utilization of precalculated similarity information stored in the similarity matrix of proteins (SIMAP) database, making it possible to process newly sequenced genomes immediately as they become available. PEDANT is freely accessible to academic users at http://pedant.gsf.de. For programmatic access Web Services are available at http://pedant.gsf.de/webservices.jsp.

Catching SARS-CoV-2 by Sequence Hybridization: a Comparative Analysis.

Controlling and monitoring the still ongoing severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic regarding geographical distribution, evolution, and emergence of new mutations of the SARS-CoV-2 virus is only possible due to continuous next-generation sequencing (NGS) and sharing sequence data worldwide. Efficient sequencing strategies enable the retrieval of increasing numbers of high-quality, full-length genomes and are, hence, indispensable. Two opposed enrichment methods, tiling multiplex PCR and sequence hybridization by bait capture, have been established for SARS-CoV-2 sequencing and are both frequently used, depending on the quality of the patient sample and the question at hand. Here, we focused on the evaluation of the sequence hybridization method by studying five commercially available sequence capture bait panels with regard to sensitivity and capture efficiency. We discovered the SARS-CoV-2-specific panel of Twist Bioscience to be the most efficient panel, followed by two respiratory panels from Twist Bioscience and Illumina, respectively. Our results provide on the one hand a decision basis for the sequencing community including a computation for using the full capacity of the flow cell and on the other hand potential improvements for the manufacturers. IMPORTANCE Sequencing the genomes of the circulating SARS-CoV-2 strains is the only way to monitor the viral spread and evolution of the virus. Two different approaches, namely, tiling multiplex PCR and sequence hybridization by bait capture, are commonly used to fulfill this task. This study describes for the first time a combined approach of droplet digital PCR (ddPCR) and NGS to evaluate five commercially available sequence capture panels targeting SARS-CoV-2. In doing so, we were able to determine the most sensitive and efficient capture panel, distinguish the mode of action of the various bait panels, and compute the number of read pairs needed to recover a high-quality full-length genome. By calculating the minimum number of read pairs needed, we are providing optimized flow cell loading conditions for all sequencing laboratories worldwide that are striving for maximizing sequencing output and simultaneously minimizing time, costs, and sequencing resources.

CoxBase: an Online Platform for Epidemiological Surveillance, Visualization, Analysis, and Typing of Coxiella burnetii Genomic Sequences.

Q (query) fever is an infectious zoonotic disease caused by the Gram-negative bacterium Coxiella burnetii. Although the disease has been studied for decades, it still represents a threat due to sporadic outbreaks across farms in Europe. The absence of a central platform for Coxiella typing data management is an important epidemiological gap that is relevant in the case of an outbreak. To fill this gap, we have designed and implemented an online, open-source, web-based platform called CoxBase (https://coxbase.q-gaps.de). This platform includes a database that holds genotyping information on more than 400 Coxiella isolates alongside metadata that annotate them. We have also implemented features for in silico genotyping of completely or minimally assembled Coxiella sequences using five different typing methods, querying of existing isolates, visualization of isolate geodata via aggregation on a world map, and submission of new isolates. We tested our in silico typing method on 50 Coxiella genomes downloaded from the RefSeq database, and we successfully genotyped all genomes except for cases where the sequence quality was poor. We identified new spacer sequences using our implementation of the multispacer sequence typing (MST) in silico typing method and established adaA gene phenotypes for all 50 genomes as well as their plasmid types. IMPORTANCE Q fever is a zoonotic disease that is a source of active epidemiological concern due to its persistent threat to public health. In this project, we have identified areas in the field of Coxiella research, especially regarding public health and genomic analysis, where there is an inadequacy of resources to monitor, organize, and analyze genomic data from C. burnetii. Subsequently, we have created an open, web-based platform that contains epidemiological information, genome typing functions comprising all the available Coxiella typing methods, and tools for isolate data discovery and visualization that could help address the above-mentioned challenges. This is the first platform to combine all disparate genotyping systems for Coxiella burnetii as well as metadata assets with tools for genomic comparison and analyses. This platform is a valuable resource for laboratory researchers as well as research epidemiologists interested in investigating the relatedness or dissimilarity among C. burnetii strains.

Comparison of Coxiella burnetii Excretion between Sheep and Goats Naturally Infected with One Cattle-Associated Genotype.

The main reservoir of Coxiella (C.) burnetii are ruminants. They shed the pathogen through birth products, vaginal mucus, faeces and milk. A direct comparison of C. burnetii excretions between naturally infected sheep and goats was performed on the same farm to investigate species-specific differences. The animals were vaccinated with an inactivated C. burnetii phase I vaccine at the beginning of the study period for public health reasons. Vaginal and rectal swabs along with milk specimens were taken monthly during the lambing period and once again at the next lambing season. To estimate the environmental contamination of the animals' housings, nasal swabs from every animal were taken simultaneously. Moreover, dust samples from the windowsills and straw beddings were collected. All samples were examined by qPCR targeting the IS1111 gene and the MLVA/VNTR typing method was performed. Whole genome sequencing was applied to determine the number of IS1111 copies followed by a calculation of C. burnetii genome equivalents of each sample. The cattle-associated genotype C7 was detected containing 29 IS1111 copies. Overall, goats seem to shed more C. burnetii through vaginal mucus and in particular shed more and for longer via the rectal route than sheep. This is supported by the larger quantities of C. burnetii DNA detected in caprine nasal swabs and environmental samples compared to the ovine ones. Transmission of C. burnetii from cattle to small ruminants must also be considered.

First Phylogenetic Analysis of Malian SARS-CoV-2 Sequences Provides Molecular Insights into the Genomic Diversity of the Sahel Region.

We are currently facing a pandemic of COVID-19, caused by a spillover from an animal-originating coronavirus to humans occurring in the Wuhan region of China in December 2019. From China, the virus has spread to 188 countries and regions worldwide, reaching the Sahel region on March 2, 2020. Since whole genome sequencing (WGS) data is very crucial to understand the spreading dynamics of the ongoing pandemic, but only limited sequencing data is available from the Sahel region to date, we have focused our efforts on generating the first Malian sequencing data available. Screening 217 Malian patient samples for the presence of SARS-CoV-2 resulted in 38 positive isolates, from which 21 whole genome sequences were generated. Our analysis shows that both the early A (19B) and the later observed B (20A/C) clade are present in Mali, indicating multiple and independent introductions of SARS-CoV-2 to the Sahel region.

Investigation of a COVID-19 outbreak in Germany resulting from a single travel-associated primary case: a case series.

BACKGROUND: In December, 2019, the newly identified severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) emerged in Wuhan, China, causing COVID-19, a respiratory disease presenting with fever, cough, and often pneumonia. WHO has set the strategic objective to interrupt spread of SARS-CoV-2 worldwide. An outbreak in Bavaria, Germany, starting at the end of January, 2020, provided the opportunity to study transmission events, incubation period, and secondary attack rates. METHODS: A case was defined as a person with SARS-CoV-2 infection confirmed by RT-PCR. Case interviews were done to describe timing of onset and nature of symptoms and to identify and classify contacts as high risk (had cumulative face-to-face contact with a confirmed case for ≥15 min, direct contact with secretions or body fluids of a patient with confirmed COVID-19, or, in the case of health-care workers, had worked within 2 m of a patient with confirmed COVID-19 without personal protective equipment) or low risk (all other contacts). High-risk contacts were ordered to stay at home in quarantine for 14 days and were actively followed up and monitored for symptoms, and low-risk contacts were tested upon self-reporting of symptoms. We defined fever and cough as specific symptoms, and defined a prodromal phase as the presence of non-specific symptoms for at least 1 day before the onset of specific symptoms. Whole genome sequencing was used to confirm epidemiological links and clarify transmission events where contact histories were ambiguous; integration with epidemiological data enabled precise reconstruction of exposure events and incubation periods. Secondary attack rates were calculated as the number of cases divided by the number of contacts, using Fisher's exact test for the 95% CIs. FINDINGS: Patient 0 was a Chinese resident who visited Germany for professional reasons. 16 subsequent cases, often with mild and non-specific symptoms, emerged in four transmission generations. Signature mutations in the viral genome occurred upon foundation of generation 2, as well as in one case pertaining to generation 4. The median incubation period was 4·0 days (IQR 2·3-4·3) and the median serial interval was 4·0 days (3·0-5·0). Transmission events were likely to have occurred presymptomatically for one case (possibly five more), at the day of symptom onset for four cases (possibly five more), and the remainder after the day of symptom onset or unknown. One or two cases resulted from contact with a case during the prodromal phase. Secondary attack rates were 75·0% (95% CI 19·0-99·0; three of four people) among members of a household cluster in common isolation, 10·0% (1·2-32·0; two of 20) among household contacts only together until isolation of the patient, and 5·1% (2·6-8·9; 11 of 217) among non-household, high-risk contacts. INTERPRETATION: Although patients in our study presented with predominately mild, non-specific symptoms, infectiousness before or on the day of symptom onset was substantial. Additionally, the incubation period was often very short and false-negative tests occurred. These results suggest that although the outbreak was controlled, successful long-term and global containment of COVID-19 could be difficult to achieve. FUNDING: All authors are employed and all expenses covered by governmental, federal state, or other publicly funded institutions.

Beyond the 'best' match: machine learning annotation of protein sequences by integration of different sources of information.

MOTIVATION: Accurate automatic assignment of protein functions remains a challenge for genome annotation. We have developed and compared the automatic annotation of four bacterial genomes employing a 5-fold cross-validation procedure and several machine learning methods. RESULTS: The analyzed genomes were manually annotated with FunCat categories in MIPS providing a gold standard. Features describing a pair of sequences rather than each sequence alone were used. The descriptors were derived from sequence alignment scores, InterPro domains, synteny information, sequence length and calculated protein properties. Following training we scored all pairs from the validation sets, selected a pair with the highest predicted score and annotated the target protein with functional categories of the prototype protein. The data integration using machine-learning methods provided significantly higher annotation accuracy compared to the use of individual descriptors alone. The neural network approach showed the best performance. The descriptors derived from the InterPro domains and sequence similarity provided the highest contribution to the method performance. The predicted annotation scores allow differentiation of reliable versus non-reliable annotations. The developed approach was applied to annotate the protein sequences from 180 complete bacterial genomes. AVAILABILITY: The FUNcat Annotation Tool (FUNAT) is available on-line as Web Services at http://mips.gsf.de/proj/funat.