SARS-CoV-2 Spike (Spike) binds to human angiotensin-converting enzyme 2 (ACE2) and the strength of this interaction could influence parameters relating to virulence. To explore whether population variants in ACE2 influence Spike binding and hence infection, we selected 10 ACE2 variants based on affinity predictions and prevalence in gnomAD and measured their affinities and kinetics for Spike receptor binding domain through surface plasmon resonance (SPR) at 37°C. We discovered variants that reduce and enhance binding, including three ACE2 variants that strongly inhibited (p.Glu37Lys, ΔΔG = -1.33 ± 0.15 kcal mol-1 and p.Gly352Val, predicted ΔΔG = -1.17 kcal mol-1) or abolished (p.Asp355Asn) binding. We also identified two variants with distinct population distributions that enhanced affinity for Spike. ACE2 p.Ser19Pro (ΔΔG = 0.59 ± 0.08 kcal mol-1) is predominant in the gnomAD African cohort (AF = 0.003) whilst p.Lys26Arg (ΔΔG = 0.26 ± 0.09 kcal mol-1) is predominant in the Ashkenazi Jewish (AF = 0.01) and European non-Finnish (AF = 0.006) cohorts. We compared ACE2 variant affinities to published SARS-CoV-2 pseudotype infectivity data and confirmed that ACE2 variants with reduced affinity for Spike can protect cells from infection. The effect of variants with enhanced Spike affinity remains unclear, but we propose a mechanism whereby these alleles could cause greater viral spreading across tissues and cell types, as is consistent with emerging understanding regarding the interplay between receptor affinity and cell-surface abundance. Finally, we compared mCSM-PPI2 ΔΔG predictions against our SPR data to assess the utility of predictions in this system. We found that predictions of decreased binding were well-correlated with experiment and could be improved by calibration, but disappointingly, predictions of highly enhanced binding were unreliable. Recalibrated predictions for all possible ACE2 missense variants at the Spike interface were calculated and used to estimate the overall burden of ACE2 variants on Covid-19.

Ankyrin protein repeats bind to a wide range of substrates and are one of the most common protein motifs in nature. Here, we collate a high-quality alignment of 7,407 ankyrin repeats and examine for the first time, the distribution of human population variants from large-scale sequencing of healthy individuals across this family. Population variants are not randomly distributed across the genome but are constrained by gene essentiality and function. Accordingly, we interpret the population variants in context with evolutionary constraint and structural features including secondary structure, accessibility and protein-protein interactions across 383 three-dimensional structures of ankyrin repeats. We find five positions that are highly conserved across homologues and also depleted in missense variants within the human population. These positions are significantly enriched in intra-domain contacts and so likely to be key for repeat packing. In contrast, a group of evolutionarily divergent positions are found to be depleted in missense variants in human and significantly enriched in protein-protein interactions. Our analysis also suggests the domain has three, not two surfaces, each with different patterns of enrichment in protein-substrate interactions and missense variants. Our findings will be of interest to those studying or engineering ankyrin-repeat containing proteins as well as those interpreting the significance of disease variants.

MOTIVATION: RNA-seq experiments are usually carried out in three or fewer replicates. In order to work well with so few samples, differential gene expression (DGE) tools typically assume the form of the underlying gene expression distribution. In this paper, the statistical properties of gene expression from RNA-seq are investigated in the complex eukaryote, Arabidopsis thaliana, extending and generalizing the results of previous work in the simple eukaryote Saccharomyces cerevisiae. RESULTS: We show that, consistent with the results in S.cerevisiae, more gene expression measurements in A.thaliana are consistent with being drawn from an underlying negative binomial distribution than either a log-normal distribution or a normal distribution, and that the size and complexity of the A.thaliana transcriptome does not influence the false positive rate performance of nine widely used DGE tools tested here. We therefore recommend the use of DGE tools that are based on the negative binomial distribution. AVAILABILITY AND IMPLEMENTATION: The raw data for the 17 WT Arabidopsis thaliana datasets is available from the European Nucleotide Archive (E-MTAB-5446). The processed and aligned data can be visualized in context using IGB (Freese et al., 2016), or downloaded directly, using our publicly available IGB quickload server at https://compbio.lifesci.dundee.ac.uk/arabidopsisQuickload/public_quickload/ under 'RNAseq>Froussios2019'. All scripts and commands are available from github at https://github.com/bartongroup/KF_arabidopsis-GRNA. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Summary: JABAWS 2.2 is a computational framework that simplifies the deployment of web services for Bioinformatics. In addition to the five multiple sequence alignment (MSA) algorithms in JABAWS 1.0, JABAWS 2.2 includes three additional MSA programs (Clustal Omega, MSAprobs, GLprobs), four protein disorder prediction methods (DisEMBL, IUPred, Ronn, GlobPlot), 18 measures of protein conservation as implemented in AACon, and RNA secondary structure prediction by the RNAalifold program. JABAWS 2.2 can be deployed on a variety of in-house or hosted systems. JABAWS 2.2 web services may be accessed from the Jalview multiple sequence analysis workbench (Version 2.8 and later), as well as directly via the JABAWS command line interface (CLI) client. JABAWS 2.2 can be deployed on a local virtual server as a Virtual Appliance (VA) or simply as a Web Application Archive (WAR) for private use. Improvements in JABAWS 2.2 also include simplified installation and a range of utility tools for usage statistics collection, and web services querying and monitoring. The JABAWS CLI client has been updated to support all the new services and allow integration of JABAWS 2.2 services into conventional scripts. A public JABAWS 2 server has been in production since December 2011 and served over 800 000 analyses for users worldwide. Availability and implementation: JABAWS 2.2 is made freely available under the Apache 2 license and can be obtained from: http://www.compbio.dundee.ac.uk/jabaws. Contact: [email protected].

MOTIVATION: The current generation of DNA sequencing technologies produce a large amount of data quickly. All of these data need to pass some form of quality control (QC) processing and checking before they can be used for any analysis. The large number of samples that are run through Illumina sequencing machines makes the process of QC an onerous and time-consuming task that requires multiple pieces of information from several sources. RESULTS: AlmostSignificant is an open-source platform for aggregating multiple sources of quality metrics as well as run and sample meta-data associated with DNA sequencing runs from Illumina sequencing machines. AlmostSignificant is a graphical platform to streamline the QC of DNA sequencing data, to store these data for future reference together with extra meta-data associated with the sequencing runs not typically retained. This simplifies the challenge of monitoring the volume of data produced by Illumina sequencers. AlmostSignificant has been used to track the quality of over 80 sequencing runs covering over 2500 samples produced over the last three years. AVAILABILITY AND IMPLEMENTATION: The code and documentation for AlmostSignificant is freely available at https://github.com/bartongroup/AlmostSignificant CONTACTS: [email protected] or [email protected] information: Supplementary data are available at Bioinformatics online.

MOTIVATION: High-throughput RNA sequencing (RNA-seq) is now the standard method to determine differential gene expression. Identifying differentially expressed genes crucially depends on estimates of read-count variability. These estimates are typically based on statistical models such as the negative binomial distribution, which is employed by the tools edgeR, DESeq and cuffdiff. Until now, the validity of these models has usually been tested on either low-replicate RNA-seq data or simulations. RESULTS: A 48-replicate RNA-seq experiment in yeast was performed and data tested against theoretical models. The observed gene read counts were consistent with both log-normal and negative binomial distributions, while the mean-variance relation followed the line of constant dispersion parameter of ∼0.01. The high-replicate data also allowed for strict quality control and screening of 'bad' replicates, which can drastically affect the gene read-count distribution. AVAILABILITY AND IMPLEMENTATION: RNA-seq data have been submitted to ENA archive with project ID PRJEB5348. CONTACT: [email protected].

MOTIVATION: The 14-3-3 family of phosphoprotein-binding proteins regulates many cellular processes by docking onto pairs of phosphorylated Ser and Thr residues in a constellation of intracellular targets. Therefore, there is a pressing need to develop new prediction methods that use an updated set of 14-3-3-binding motifs for the identification of new 14-3-3 targets and to prioritize the downstream analysis of >2000 potential interactors identified in high-throughput experiments. RESULTS: Here, a comprehensive set of 14-3-3-binding targets from the literature was used to develop 14-3-3-binding phosphosite predictors. Position-specific scoring matrix, support vector machines (SVM) and artificial neural network (ANN) classification methods were trained to discriminate experimentally determined 14-3-3-binding motifs from non-binding phosphopeptides. ANN, position-specific scoring matrix and SVM methods showed best performance for a motif window spanning from -6 to +4 around the binding phosphosite, achieving Matthews correlation coefficient of up to 0.60. Blind prediction showed that all three methods outperform two popular 14-3-3-binding site predictors, Scansite and ELM. The new methods were used for prediction of 14-3-3-binding phosphosites in the human proteome. Experimental analysis of high-scoring predictions in the FAM122A and FAM122B proteins confirms the predictions and suggests the new 14-3-3-predictors will be generally useful. AVAILABILITY AND IMPLEMENTATION: A standalone prediction web server is available at http://www.compbio.dundee.ac.uk/1433pred. Human candidate 14-3-3-binding phosphosites were integrated in ANIA: ANnotation and Integrated Analysis of the 14-3-3 interactome database.

SUMMARY: JABAWS is a web services framework that simplifies the deployment of web services for bioinformatics. JABAWS:MSA provides services for five multiple sequence alignment (MSA) methods (Probcons, T-coffee, Muscle, Mafft and ClustalW), and is the system employed by the Jalview multiple sequence analysis workbench since version 2.6. A fully functional, easy to set up server is provided as a Virtual Appliance (VA), which can be run on most operating systems that support a virtualization environment such as VMware or Oracle VirtualBox. JABAWS is also distributed as a Web Application aRchive (WAR) and can be configured to run on a single computer and/or a cluster managed by Grid Engine, LSF or other queuing systems that support DRMAA. JABAWS:MSA provides clients full access to each application's parameters, allows administrators to specify named parameter preset combinations and execution limits for each application through simple configuration files. The JABAWS command-line client allows integration of JABAWS services into conventional scripts. AVAILABILITY AND IMPLEMENTATION: JABAWS is made freely available under the Apache 2 license and can be obtained from: http://www.compbio.dundee.ac.uk/jabaws.

BACKGROUND: Nucleolar localization sequences (NoLSs) are short targeting sequences responsible for the localization of proteins to the nucleolus. Given the large number of proteins experimentally detected in the nucleolus and the central role of this subnuclear compartment in the cell, NoLSs are likely to be important regulatory elements controlling cellular traffic. Although many proteins have been reported to contain NoLSs, the systematic characterization of this group of targeting motifs has only recently been carried out. RESULTS: Here, we describe NoD, a web server and a command line program that predicts the presence of NoLSs in proteins. Using the web server, users can submit protein sequences through the NoD input form and are provided with a graphical output of the NoLS score as a function of protein position. While the web server is most convenient for making prediction for just a few proteins, the command line version of NoD can return predictions for complete proteomes. NoD is based on our recently described human-trained artificial neural network predictor. Through stringent independent testing of the predictor using available experimentally validated NoLS-containing eukaryotic and viral proteins, the NoD sensitivity and positive predictive value were estimated to be 71% and 79% respectively. CONCLUSIONS: NoD is the first tool to provide predictions of nucleolar localization sequences in diverse eukaryotes and viruses. NoD can be run interactively online at http://www.compbio.dundee.ac.uk/nod or downloaded to use locally.

In the scientific literature and large public databases there are currently only ~39000 human protein-protein interactions that have been experimentally confirmed out of a potential 330,000,000 (assuming 1 protein per gene). To bridge the gap, computational methods are required to guide further experimental endeavours. The PIPs framework [ 1 ] uses a naïve Bayesian method that combines the predictive capabilities of numerous features to calculate the likelihood of interaction between two proteins. Features considered by the predictor include co-expression, orthology, domain co-occurrence, post translational modification and a new feature analysing semantic similarity of Gene Ontology terms. The predictor now includes two modules that make predictions based on the topology of the predicted protein-protein interaction network. We predict 318800 interaction predictions of which 310732 (96.3%) are not present within other publically available databases. Several of the predictions have been experimentally validated by external groups. The PIPs website ( http://www.compbio.dundee.ac.uk/pips ) [ 2 ] is an easy to use system to explore the predictions that have been made. Searches can be initiated by querying with a protein identifier (IPI, RefSeq or UniProt) or via a keyword search. All predicted protein-protein interactions are returned ranked by their likelihood of interaction. The website allows the user to analyse the evidence used to calculate the likelihood of interaction and provides links through to external databases and publications to retrieve the source data. The set of predictions that have been made in this work increase the coverage of the human interactome and help guide future research.

UNLABELLED: Jalview Version 2 is a system for interactive WYSIWYG editing, analysis and annotation of multiple sequence alignments. Core features include keyboard and mouse-based editing, multiple views and alignment overviews, and linked structure display with Jmol. Jalview 2 is available in two forms: a lightweight Java applet for use in web applications, and a powerful desktop application that employs web services for sequence alignment, secondary structure prediction and the retrieval of alignments, sequences, annotation and structures from public databases and any DAS 1.53 compliant sequence or annotation server. AVAILABILITY: The Jalview 2 Desktop application and JalviewLite applet are made freely available under the GPL, and can be downloaded from www.jalview.org.

MicroRNAs (miRNAs) and small nucleolar RNAs (snoRNAs) are two classes of small non-coding regulatory RNAs, which have been much investigated in recent years. While their respective functions in the cell are distinct, they share interesting genomic similarities, and recent sequencing projects have identified processed forms of snoRNAs that resemble miRNAs. Here, we investigate a possible evolutionary relationship between miRNAs and box H/ACA snoRNAs. A comparison of the genomic locations of reported miRNAs and snoRNAs reveals an overlap of specific members of these classes. To test the hypothesis that some miRNAs might have evolved from snoRNA encoding genomic regions, reported miRNA-encoding regions were scanned for the presence of box H/ACA snoRNA features. Twenty miRNA precursors show significant similarity to H/ACA snoRNAs as predicted by snoGPS. These include molecules predicted to target known ribosomal RNA pseudouridylation sites in vivo for which no guide snoRNA has yet been reported. The predicted folded structures of these twenty H/ACA snoRNA-like miRNA precursors reveal molecules which resemble the structures of known box H/ACA snoRNAs. The genomic regions surrounding these predicted snoRNA-like miRNAs are often similar to regions around snoRNA retroposons, including the presence of transposable elements, target site duplications and poly (A) tails. We further show that the precursors of five H/ACA snoRNA-like miRNAs (miR-151, miR-605, mir-664, miR-215 and miR-140) bind to dyskerin, a specific protein component of functional box H/ACA small nucleolar ribonucleoprotein complexes suggesting that these molecules have retained some H/ACA snoRNA functionality. The detection of small RNA molecules that share features of miRNAs and snoRNAs suggest that these classes of RNA may have an evolutionary relationship.

The ability to rank proteins by their likely success in crystallization is useful in current Structural Biology efforts and in particular in high-throughput Structural Genomics initiatives. We present ParCrys, a Parzen Window approach to estimate a protein's propensity to produce diffraction-quality crystals. The Protein Data Bank (PDB) provided training data whilst the databases TargetDB and PepcDB were used to define feature selection data as well as test data independent of feature selection and training. ParCrys outperforms the OB-Score, SECRET and CRYSTALP on the data examined, with accuracy and Matthews correlation coefficient values of 79.1% and 0.582, respectively (74.0% and 0.227, respectively, on data with a 'real-world' ratio of positive:negative examples). ParCrys predictions and associated data are available from www.compbio.dundee.ac.uk/parcrys.

BACKGROUND: Amino acids responsible for structure, core function or specificity may be inferred from multiple protein sequence alignments where a limited set of residue types are tolerated. The rise in available protein sequences continues to increase the power of techniques based on this principle. RESULTS: A new algorithm, SMERFS, for predicting protein functional sites from multiple sequences alignments was compared to 14 conservation measures and to the MINER algorithm. Validation was performed on an automatically generated dataset of 1457 families derived from the protein interactions database SNAPPI-DB, and a smaller manually curated set of 148 families. The best performing measure overall was Williamson property entropy, with ROC0.1 scores of 0.0087 and 0.0114 for domain and small molecule contact prediction, respectively. The Lancet method performed worse than random on protein-protein interaction site prediction (ROC0.1 score of 0.0008). The SMERFS algorithm gave similar accuracy to the phylogenetic tree-based MINER algorithm but was superior to Williamson in prediction of non-catalytic transient complex interfaces. SMERFS predicts sites that are significantly more solvent accessible compared to Williamson. CONCLUSION: Williamson property entropy is the the best performing of 14 conservation measures examined. The difference in performance of SMERFS relative to Williamson in manually defined complexes was dependent on complex type. The best choice of analysis method is therefore dependent on the system of interest. Additional computation employed by Miner in calculation of phylogenetic trees did not produce improved results over SMERFS. SMERFS performance was improved by use of windows over alignment columns, illustrating the necessity of considering the local environment of positions when assessing their functional significance.

BACKGROUND: Although the prediction of protein-protein interactions has been extensively investigated for yeast, few such datasets exist for the far larger proteome in human. Furthermore, it has recently been estimated that the overall average false positive rate of available computational and high-throughput experimental interaction datasets is as high as 90%. RESULTS: The prediction of human protein-protein interactions was investigated by combining orthogonal protein features within a probabilistic framework. The features include co-expression, orthology to known interacting proteins and the full-Bayesian combination of subcellular localization, co-occurrence of domains and post-translational modifications. A novel scoring function for local network topology was also investigated. This topology feature greatly enhanced the predictions and together with the full-Bayes combined features, made the largest contribution to the predictions. Using a conservative threshold, our most accurate predictor identifies 37606 human interactions, 32892 (80%) of which are not present in other publicly available large human interaction datasets, thus substantially increasing the coverage of the human interaction map. A subset of the 32892 novel predicted interactions have been independently validated. Comparison of the prediction dataset to other available human interaction datasets estimates the false positive rate of the new method to be below 80% which is competitive with other methods. Since the new method scores and ranks all human protein pairs, smaller subsets of higher quality can be generated thus leading to even lower false positive prediction rates. CONCLUSION: The set of interactions predicted in this work increases the coverage of the human interaction map and will help determine the highest confidence human interactions.

BACKGROUND: Percentage Identity (PID) is frequently quoted in discussion of sequence alignments since it appears simple and easy to understand. However, although there are several different ways to calculate percentage identity and each may yield a different result for the same alignment, the method of calculation is rarely reported. Accordingly, quantification of the variation in PID caused by the different calculations would help in interpreting PID values in the literature. In this study, the variation in PID was quantified systematically on a reference set of 1028 alignments generated by comparison of the protein three-dimensional structures. Since the alignment algorithm may also affect the range of PID, this study also considered the effect of algorithm, and the combination of algorithm and PID method. RESULTS: The maximum variation in PID due to the calculation method was 11.5% while the effect of alignment algorithm on PID was up to 14.6% across three popular alignment methods. The combined effect of alignment algorithm and PID calculation gave a variation of up to 22% on the test data, with an average of 5.3% +/- 2.8% for sequence pairs with < 30% identity. In order to see which PID method was most highly correlated with structural similarity, four different PID calculations were compared to similarity scores (Sc) from the comparison of the corresponding protein three-dimensional structures. The highest correlation coefficient for a PID calculation was 0.80. In contrast, the more sophisticated Z-score calculated by reference to randomized sequences gave a correlation coefficient of 0.84. CONCLUSION: Although it is well known amongst expert sequence analysts that PID is a poor score for discriminating between protein sequences, the apparent simplicity of the percentage identity score encourages its widespread use in establishing cutoffs for structural similarity. This paper illustrates that not only is PID a poor measure of sequence similarity when compared to the Z-score, but that there is also a large uncertainty in reported PID values. Since better alternatives to PID exist to quantify sequence similarity, these should be quoted where possible in preference to PID. The findings presented here should prove helpful to those new to sequence analysis, and in warning those who seek to interpret the value of a PID reported in the literature.

BACKGROUND: In the post-genomic era, systems-level studies are being performed that seek to explain complex biological systems by integrating diverse resources from fields such as genomics, proteomics or transcriptomics. New information management systems are now needed for the collection, validation and analysis of the vast amount of heterogeneous data available. Multiple alignments of complete sequences provide an ideal environment for the integration of this information in the context of the protein family. RESULTS: MACSIMS is a multiple alignment-based information management program that combines the advantages of both knowledge-based and ab initio sequence analysis methods. Structural and functional information is retrieved automatically from the public databases. In the multiple alignment, homologous regions are identified and the retrieved data is evaluated and propagated from known to unknown sequences with these reliable regions. In a large-scale evaluation, the specificity of the propagated sequence features is estimated to be >99%, i.e. very few false positive predictions are made. MACSIMS is then used to characterise mutations in a test set of 100 proteins that are known to be involved in human genetic diseases. The number of sequence features associated with these proteins was increased by 60%, compared to the features available in the public databases. An XML format output file allows automatic parsing of the MACSIM results, while a graphical display using the JalView program allows manual analysis. CONCLUSION: MACSIMS is a new information management system that incorporates detailed analyses of protein families at the structural, functional and evolutionary levels. MACSIMS thus provides a unique environment that facilitates knowledge extraction and the presentation of the most pertinent information to the biologist. A web server and the source code are available at http://bips.u-strasbg.fr/MACSIMS/.

Abstract Summary: Multiple sequence alignment remains a crucial method for understanding the function of groups of related nucleic acid and protein sequences. However, it is known that automatic multiple sequence alignments can often be improved by manual editing. Therefore, tools are needed to view and edit multiple sequence alignments. Due to growth in the sequence databases, multiple sequence alignments can often be large and difficult to view efficiently. The Jalview Java alignment editor is presented here, which enables fast viewing and editing of large multiple sequence alignments. Availability: The Jar file and source code for Jalview is freely available via the World Wide Web at http://www.jalview.org. A Jalview mailing list is also available by e-mailing [email protected] with subscribe Jalview in the body of the mail.

BACKGROUND: The function of a novel gene product is typically predicted by transitive assignment of annotation from similar sequences. We describe a novel method, GOtcha, for predicting gene product function by annotation with Gene Ontology (GO) terms. GOtcha predicts GO term associations with term-specific probability (P-score) measures of confidence. Term-specific probabilities are a novel feature of GOtcha and allow the identification of conflicts or uncertainty in annotation. RESULTS: The GOtcha method was applied to the recently sequenced genome for Plasmodium falciparum and six other genomes. GOtcha was compared quantitatively for retrieval of assigned GO terms against direct transitive assignment from the highest scoring annotated BLAST search hit (TOPBLAST). GOtcha exploits information deep into the 'twilight zone' of similarity search matches, making use of much information that is otherwise discarded by more simplistic approaches. At a P-score cutoff of 50%, GOtcha provided 60% better recovery of annotation terms and 20% higher selectivity than annotation with TOPBLAST at an E-value cutoff of 10(-4). CONCLUSIONS: The GOtcha method is a useful tool for genome annotators. It has identified both errors and omissions in the original Plasmodium falciparum annotation and is being adopted by many other genome sequencing projects.

SUMMARY: An algorithm and software are described that provide a fast method to produce a novel, function-oriented visualization of the results of a sequence database search. Text mining of sequence annotations allows position specific plots of potential functional similarity to be compared in a simple compact representation. AVAILABILITY: The application can be accessed via a web server at http://www.compbio.dundee.ac.uk. The RHIMS software may be obtained by request to the authors.

MOTIVATION: Sequence alignment methods that compare two sequences (pairwise methods) are important tools for the detection of biological sequence relationships. In genome annotation, multiple methods are often run and agreement between methods taken as confirmation. In this paper, we assess the advantages of combining search methods by comparing seven pairwise alignment methods, including three local dynamic programming algorithms (PRSS, SSEARCH and SCANPS), two global dynamic programming algorithms (GSRCH and AMPS) and two heuristic approximations (BLAST and FASTA), individually and by pairwise intersection and union of their result lists at equal p-value cut-offs. RESULTS: When applied singly, the dynamic programming methods SCANPS and SSEARCH gave significantly better coverage (p=0.01) compared to AMPS, GSRCH, PRSS, BLAST and FASTA. Results ranked by BLAST p-values gave significantly better coverage compared to ranking by BLAST e-values. Of 56 combinations of eight methods considered, 19 gave significant increases in coverage at low error compared to the parent methods at an equal p-value cutoff. The union of results by BLAST (p-value) and FASTA at an equal p-value cutoff gave significantly better coverage than either method individually. The best overall performance was obtained from the intersection of the results from SSEARCH and the GSRCH62 global alignment method. At an error level of five false positives, this combination found 444 true positives, a significant 12.4% increase over SSEARCH applied alone.

BACKGROUND: The alignment of two or more protein sequences provides a powerful guide in the prediction of the protein structure and in identifying key functional residues, however, the utility of any prediction is completely dependent on the accuracy of the alignment. In this paper we describe a suite of reference alignments derived from the comparison of protein three-dimensional structures together with evaluation measures and software that allow automatically generated alignments to be benchmarked. We test the OXBench benchmark suite on alignments generated by the AMPS multiple alignment method, then apply the suite to compare eight different multiple alignment algorithms. The benchmark shows the current state-of-the art for alignment accuracy and provides a baseline against which new alignment algorithms may be judged. RESULTS: The simple hierarchical multiple alignment algorithm, AMPS, performed as well as or better than more modern methods such as CLUSTALW once the PAM250 pair-score matrix was replaced by a BLOSUM series matrix. AMPS gave an accuracy in Structurally Conserved Regions (SCRs) of 89.9% over a set of 672 alignments. The T-COFFEE method on a data set of families with <8 sequences gave 91.4% accuracy, significantly better than CLUSTALW (88.9%) and all other methods considered here. The complete suite is available from http://www.compbio.dundee.ac.uk. CONCLUSIONS: The OXBench suite of reference alignments, evaluation software and results database provide a convenient method to assess progress in sequence alignment techniques. Evaluation measures that were dependent on comparison to a reference alignment were found to give good discrimination between methods. The STAMP Sc Score which is independent of a reference alignment also gave good discrimination. Application of OXBench in this paper shows that with the exception of T-COFFEE, the majority of the improvement in alignment accuracy seen since 1985 stems from improved pair-score matrices rather than algorithmic refinements. The maximum theoretical alignment accuracy obtained by pooling results over all methods was 94.5% with 52.5% accuracy for alignments in the 0-10 percentage identity range. This suggests that further improvements in accuracy will be possible in the future.

Abstract Summary: The 3Dee database is a repository of protein structural domains. It stores alternative domain definitions for the same protein, organises domains into sequence and structural hierarchies, contains non-redundant set(s) of sequences and structures, multiple structure alignments for families of domains, and allows previous versions of the database to be regenerated. Availability: 3Dee is accessible on the World Wide Web at the URL http://barton.ebi.ac.uk/servers/3Dee.html Contact: [email protected]; http://barton.ebi.ac.uk * To whom correspondence should be addressed. Present address: EMBL-European Bioinformatics Institute, Wellcome Trust Genome Campus, Hinxton, Cambridge CB10 1SD, UK.

MOTIVATION: The global alignment of protein sequence pairs is often used in the classification and analysis of full-length sequences. The calculation of a Z-score for the comparison gives a length and composition corrected measure of the similarity between the sequences. However, the Z-score alone, does not indicate the likely biological significance of the similarity. In this paper, all pairs of domains from 250 sequences belonging to different SCOP folds were aligned and Z-scores calculated. The distribution of Z-scores was fitted with a peak distribution from which the probability of obtaining a given Z-score from the global alignment of two protein sequences of unrelated fold was calculated. A similar analysis was applied to subsequence pairs found by the Smith-Waterman algorithm. These analyses allow the probability that two protein sequences share the same fold to be estimated by global sequence alignment. RESULTS: The relationship between Z-score and probability varied little over the matrix/gap penalty combinations examined. However, an average shift of +4.7 was observed for Z-scores derived from global alignment of locally-aligned subsequences compared to global alignment of the full-length sequences. This shift was shown to be the result of pre-selection by local alignment, rather than any structural similarity in the subsequences. The search ability of both methods was benchmarked against the SCOP superfamily classification and showed that global alignment Z-scores generated from the entire sequence are as effective as SSEARCH at low error rates and more effective at higher error rates. However, global alignment Z-scores generated from the best locally-aligned subsequence were significantly less effective than SSEARCH. The method of estimating statistical significance described here was shown to give similar values to SSEARCH and BLAST, providing confidence in the significance estimation. AVAILABILITY: Software to apply the statistics to global alignments is available from http://barton.ebi.ac.uk. CONTACT: [email protected]

MOTIVATION: An automatic sequence searching method (ProtEST) is described which constructs multiple protein sequence alignments from protein sequences and translated expressed sequence tags (ESTs). ProtEST is more effective than a simple TBLASTN search of the query against the EST database, as the sequences are automatically clustered, assembled, made non-redundant, checked for sequence errors, translated into protein and then aligned and displayed. RESULTS: A ProtEST search found a non-redundant, translated, error- and length-corrected EST sequence for > 58% of sequences when single sequences from 1407 Pfam-A seed alignments were used as the probe. The average family size of the resulting alignments of translated EST sequences contained > 10 sequences. In a cross-validated test of protein secondary structure prediction, alignments from the new procedure led to an improvement of 3.4% average Q3 prediction accuracy over single sequences. AVAILABILITY: The ProtEST method is available as an Internet World Wide Web service http://barton.ebi.ac.uk/servers/protest.html+ ++ The Wise2 package for protein and genomic comparisons and the ProtESTWise script can be found at http://www.sanger.ac.uk/Software/Wise2 CONTACT: [email protected]

UNLABELLED: An interactive protein secondary structure prediction Internet server is presented. The server allows a single sequence or multiple alignment to be submitted, and returns predictions from six secondary structure prediction algorithms that exploit evolutionary information from multiple sequences. A consensus prediction is also returned which improves the average Q3 accuracy of prediction by 1% to 72.9%. The server simplifies the use of current prediction algorithms and allows conservation patterns important to structure and function to be identified. AVAILABILITY: http://barton.ebi.ac.uk/servers/jpred.h tml CONTACT: [email protected]

An efficient algorithm is described to locate locally optimal alignments between two sequences allowing for insertions and deletions. The algorithm is based on that of Smith and Waterman which returns the single best local alignment. However, the algorithm described here permits all non-intersecting locally optimal alignments to be determined in a single pass through the comparison matrix. The algorithm simplifies the location of repeats, multiple domains and shuffled motifs, and is fast enough to be used on a conventional workstation to scan large sequence databanks.

An algorithm is described for the systematic characterization of the physico-chemical properties seen at each position in a multiple protein sequence alignment. The new algorithm allows questions important in the design of mutagenesis experiments to be quickly answered since positions in the alignment that show unusual or interesting residue substitution patterns may be rapidly identified. The strategy is based on a flexible set-based description of amino acid properties, which is used to define the conservation between any group of amino acids. Sequences in the alignment are gathered into subgroups on the basis of sequence similarity, functional, evolutionary or other criteria. All pairs of subgroups are then compared to highlight positions that confer the unique features of each subgroup. The algorithm is encoded in the computer program AMAS (Analysis of Multiply Aligned Sequences) which provides a textual summary of the analysis and an annotated (boxed, shaded and/or coloured) multiple sequence alignment. The algorithm is illustrated by application to an alignment of 67 SH2 domains where patterns of conserved hydrophobic residues that constitute the protein core are highlighted. The analysis of charge conservation across annexin domains identifies the locations at which conserved charges change sign. The algorithm simplifies the analysis of multiple sequence data by condensing the mass of information present, and thus allows the rapid identification of substitutions of structural and functional importance.

The programme pscan has been developed to distribute protein databank scans over a network of computers that share a common file system. pscan may be used in conjunction with most conventional sequence comparison programmes with few modifications. In test runs using the Smith-Waterman dynamic programming algorithm, the time required to scan a 6858 sequence databank using a query sequence 740 residues long was reduced from approximately 50 min for a single processor, to approximately 11 minutes for five processors. Accordingly, pscan provides a low-cost, portable alternative to dedicated parallel processing computers.