Protein molecular function prediction by Bayesian phylogenomics

We present a statistical graphical model to infer specific molecular function for unannotated protein sequences using homology. Based on phylogenomic principles, SIFTER (Statistical Inference of Function Through Evolutionary Relationships) accurately predicts molecular function for members of a protein family given a reconciled phylogeny and available function annotations, even when the data are sparse or noisy. Our method produced specific and consistent molecular function predictions across 100 Pfam families in comparison to the Gene Ontology annotation database, BLAST, GOtcha, and Orthostrapper. We performed a more detailed exploration of functional predictions on the adenosine-5′-monophosphate/adenosine deaminase family and the lactate/malate dehydrogenase family, in the former case comparing the predictions against a gold standard set of published functional characterizations. Given function annotations for 3% of the proteins in the deaminase family, SIFTER achieves 96% accuracy in predicting molecular function for experimentally characterized proteins as reported in the literature. The accuracy of SIFTER on this dataset is a significant improvement over other currently available methods such as BLAST (75%), GeneQuiz (64%), GOtcha (89%), and Orthostrapper (11%). We also experimentally characterized the adenosine deaminase from Plasmodium falciparum, confirming SIFTER's prediction. The results illustrate the predictive power of exploiting a statistical model of function evolution in phylogenomic problems. A software implementation of SIFTER is available from the authors.     

A Threading-Based Method for the Prediction of DNA-Binding Proteins with Application to the Human Genome

Diverse mechanisms for DNA-protein recognition have been elucidated in numerous atomic complex structures from various protein families. These structural data provide an invaluable knowledge base not only for understanding DNA-protein interactions, but also for developing specialized methods that predict the DNA-binding function from protein structure. While such methods are useful, a major limitation is that they require an experimental structure of the target as input. To overcome this obstacle, we develop a threading-based method, DNA-Binding-Domain-Threader (DBD-Threader), for the prediction of DNA-binding domains and associated DNA-binding protein residues. Our method, which uses a template library composed of DNA-protein complex structures, requires only the target protein''s sequence. In our approach, fold similarity and DNA-binding propensity are employed as two functional discriminating properties. In benchmark tests on 179 DNA-binding and 3,797 non-DNA-binding proteins, using templates whose sequence identity is less than 30% to the target, DBD-Threader achieves a sensitivity/precision of 56%/86%. This performance is considerably better than the standard sequence comparison method PSI-BLAST and is comparable to DBD-Hunter, which requires an experimental structure as input. Moreover, for over 70% of predicted DNA-binding domains, the backbone Root Mean Square Deviations (RMSDs) of the top-ranked structural models are within 6.5 Å of their experimental structures, with their associated DNA-binding sites identified at satisfactory accuracy. Additionally, DBD-Threader correctly assigned the SCOP superfamily for most predicted domains. To demonstrate that DBD-Threader is useful for automatic function annotation on a large-scale, DBD-Threader was applied to 18,631 protein sequences from the human genome; 1,654 proteins are predicted to have DNA-binding function. Comparison with existing Gene Ontology (GO) annotations suggests that ∼30% of our predictions are new. Finally, we present some interesting predictions in detail. In particular, it is estimated that ∼20% of classic zinc finger domains play a functional role not related to direct DNA-binding.     

Emergence of protein fold families through rational design

Diverse proteins with similar structures are grouped into families of homologs and analogs, if their sequence similarity is higher or lower, respectively, than 20%–30%. It was suggested that protein homologs and analogs originate from a common ancestor and diverge in their distinct evolutionary time scales, emerging as a consequence of the physical properties of the protein sequence space. Although a number of studies have determined key signatures of protein family organization, the sequence-structure factors that differentiate the two evolution-related protein families remain unknown. Here, we stipulate that subtle structural changes, which appear due to accumulating mutations in the homologous families, lead to distinct packing of the protein core and, thus, novel compositions of core residues. The latter process leads to the formation of distinct families of homologs. We propose that such differentiation results in the formation of analogous families. To test our postulate, we developed a molecular modeling and design toolkit, Medusa, to computationally design protein sequences that correspond to the same fold family. We find that analogous proteins emerge when a backbone structure deviates only 1–2 Å root-mean-square deviation from the original structure. For close homologs, core residues are highly conserved. However, when the overall sequence similarity drops to ~25%–30%, the composition of core residues starts to diverge, thereby forming novel families of protein homologs. This direct observation of the formation of protein homologs within a specific fold family supports our hypothesis. The conservation of amino acids in designed sequences recapitulates that of the naturally occurring sequences, thereby validating our computational design methodology.     

Systematic subcellular localization of novel proteins identified by large-scale cDNA sequencing

As a first step towards a more comprehensive functional characterization of cDNAs than bioinformatic analysis, which can only make functional predictions for about half of the cDNAs sequenced, we have developed and tested a strategy that allows their systematic and fast subcellular localization. We have used a novel cloning technology to rapidly generate N- and C-terminal green fluorescent protein fusions of cDNAs to examine the intracellular localizations of >100 expressed fusion proteins in living cells. The entire analysis is suitable for automation, which will be important for scaling up throughput. For >80% of these new proteins a clear intracellular localization to known structures or organelles could be determined. For the cDNAs where bioinformatic analyses were able to predict possible identities, the localization was able to support these predictions in 75% of cases. For those cDNAs where no homologies could be predicted, the localization data represent the first information.     

In silico Structural Study of Random Amino Acid Sequence Proteins Not Present in Nature

The three‐dimensional structures of a set of ‘never born proteins’ (NBP, random amino acid sequence proteins with no significant homology with known proteins) were predicted using two methods: Rosetta and the one based on the ‘fuzzy‐oil‐drop’ (FOD) model. More than 3000 different random amino acid sequences have been generated, filtered against the non redundant protein sequence data base, to remove sequences with significant homology with known proteins, and subjected to three‐dimensional structure prediction. Comparison between Rosetta and FOD predictions allowed to select the ten top (highest structural similarity) and the ten bottom (the lowest structural similarity) structures from the ranking list organized according to the RMS‐D value. The selected structures were taken for detailed analysis to define the scale of structural accordance and discrepancy between the two methods. The structural similarity measurements revealed discrepancies between structures generated on the basis of the two methods. Their potential biological function appeared to be quite different as well. The ten bottom structures appeared to be ‘unfoldable’ for the FOD model. Some aspects of the general characteristics of the NBPs are also discussed. The calculations were performed on the EUChinaGRID grid platform to test the performance of this infrastructure for massive protein structure predictions.     

Structure and functional annotation of hypothetical proteins having putative Rubisco activase function from Vitis vinifera

Rubisco is a very large, complex and one of the most abundant proteins in the world and comprises up to 50% of all soluble protein in plants. The activity of Rubisco, the enzyme that catalyzes CO₂ assimilation in photosynthesis, is regulated by Rubisco activase (Rca). In the present study, we searched for hypothetical protein of Vitis vinifera which has putative Rubisco activase function. The Arabidopsis and tobacco Rubisco activase protein sequences were used as seed sequences to search against Vitis vinifera in UniprotKB database. The selected hypothetical proteins of Vitis vinifera were subjected to sequence, structural and functional annotation. Subcellular localization predictions suggested it to be cytoplasmic protein. Homology modelling was used to define the three-dimensional (3D) structure of selected hypothetical proteins of Vitis vinifera. Template search revealed that all the hypothetical proteins share more than 80% sequence identity with structure of green-type Rubisco activase from tobacco, indicating proteins are evolutionary conserved. The homology modelling was generated using SWISS-MODEL. Several quality assessment and validation parameters computed indicated that homology models are reliable. Further, functional annotation through PFAM, CATH, SUPERFAMILY, CDART suggested that selected hypothetical proteins of Vitis vinifera contain ATPase family associated with various cellular activities (AAA) and belong to the AAA+ super family of ring-shaped P-loop containing nucleoside triphosphate hydrolases. This study will lead to research in the optimization of the functionality of Rubisco which has large implication in the improvement of plant productivity and resource use efficiency.     

Towards a reliable objective function for multiple sequence alignments.

Multiple sequence alignment is a fundamental tool in a number of different domains in modern molecular biology, including functional and evolutionary studies of a protein family. Multiple alignments also play an essential role in the new integrated systems for genome annotation and analysis. Thus, the development of new multiple alignment scores and statistics is essential, in the spirit of the work dedicated to the evaluation of pairwise sequence alignments for database searching techniques. We present here norMD, a new objective scoring function for multiple sequence alignments. NorMD combines the advantages of the column-scoring techniques with the sensitivity of methods incorporating residue similarity scores. In addition, norMD incorporates ab initio sequence information, such as the number, length and similarity of the sequences to be aligned. The sensitivity and reliability of the norMD objective function is demonstrated using structural alignments in the SCOP and BAliBASE databases. The norMD scores are then applied to the multiple alignments of the complete sequences (MACS) detected by BlastP with E-value<10, for a set of 734 hypothetical proteins encoded by the Vibrio cholerae genome. Unrelated or badly aligned sequences were automatically removed from the MACS, leaving a high-quality multiple alignment which could be reliably exploited in a subsequent functional and/or structural annotation process. After removal of unreliable sequences, 176 (24 %) of the alignments contained at least one sequence with a functional annotation. 103 of these new matches were supported by significant hits to the Interpro domain and motif database.     

A kingdom-specific protein domain HMM library for improved annotation of fungal genomes

Background

Mimivirus isolated from A. polyphaga is the largest virus discovered so far. It is unique among all the viruses in having genes related to translation, DNA repair and replication which bear close homology to eukaryotic genes. Nevertheless, only a small fraction of the proteins (33%) encoded in this genome has been assigned a function. Furthermore, a large fraction of the unassigned protein sequences bear no sequence similarity to proteins from other genomes. These sequences are referred to as ORFans. Because of their lack of sequence similarity to other proteins, they can not be assigned putative functions using standard sequence comparison methods. As part of our genome-wide computational efforts aimed at characterizing Mimivirus ORFans, we have applied fold-recognition methods to predict the structure of these ORFans and further functions were derived based on conservation of functionally important residues in sequence-template alignments.

Results

Using fold recognition, we have identified highly confident computational 3D structural assignments for 21 Mimivirus ORFans. In addition, highly confident functional predictions for 6 of these ORFans were derived by analyzing the conservation of functional motifs between the predicted structures and proteins of known function. This analysis allowed us to classify these 6 previously unannotated ORFans into their specific protein families: carboxylesterase/thioesterase, metal-dependent deacetylase, P-loop kinases, 3-methyladenine DNA glycosylase, BTB domain and eukaryotic translation initiation factor eIF4E.

Conclusion

Using stringent fold recognition criteria we have assigned three-dimensional structures for 21 of the ORFans encoded in the Mimivirus genome. Further, based on the 3D models and an analysis of the conservation of functionally important residues and motifs, we were able to derive functional attributes for 6 of the ORFans. Our computational identification of important functional sites in these ORFans can be the basis for a subsequent experimental verification of our predictions. Further computational and experimental studies are required to elucidate the 3D structures and functions of the remaining Mimivirus ORFans.     

Characterization of Insertions of IS476 and Two Newly Identified Insertion Sequences, IS1478 and IS1479, in Xanthomonas campestris pv. campestris

Thirty-two plasmid insertion mutants were independently isolated from two strains of Xanthomonas campestris pv. campestris in Taiwan. Of the 32 mutants, 14 (44%), 8 (25%), and 4 (12%) mutants resulted from separate insertions of an IS3 family member, IS476, and two new insertion sequences (IS), IS1478 and IS1479. While IS1478 does not have significant sequence homology with any IS elements in the EMBL/GenBank/DDBJ database, IS1479 demonstrated 73% sequence homology with IS1051 in X. campestris pv. dieffenbachiae, 62% homology with IS52 in Pseudomonas syringae pv. glycinea, and 60% homology with IS5 in Escherichia coli. Based on the predicted transposase sequences as well as the terminal nucleotide sequences, IS1478 by itself constitutes a new subfamily of the widespread IS5 family, whereas IS1479, along with IS1051, IS52, and IS5, belongs to the IS5 subfamily of the IS5 family. All but one of the IS476 insertions had duplications of 4 bp at the target sites without sequence preference and were randomly distributed. An IS476 insertion carried a duplication of 952 bp at the target site. A model for generating these long direct repeats is proposed. Insertions of IS1478 and IS1479, on the other hand, were not random, and IS1478 and IS1479 each showed conservation of PyPuNTTA and PyTAPu sequences (Py is a pyrimidine, Pu is a purine, and N is any nucleotide) for duplications at the target sites. The results of Southern blot hybridization analysis indicated that multiple copies of IS476, IS1478, and IS1479 are present in the genomes of all seven X. campestris pv. campestris strains tested and several X. campestris pathovars.     

An Atlas of the Thioredoxin Fold Class Reveals the Complexity of Function-Enabling Adaptations

The group of proteins that contain a thioredoxin (Trx) fold is huge and diverse. Assessment of the variation in catalytic machinery of Trx fold proteins is essential in providing a foundation for understanding their functional diversity and predicting the function of the many uncharacterized members of the class. The proteins of the Trx fold class retain common features—including variations on a dithiol CxxC active site motif—that lead to delivery of function. We use protein similarity networks to guide an analysis of how structural and sequence motifs track with catalytic function and taxonomic categories for 4,082 representative sequences spanning the known superfamilies of the Trx fold. Domain structure in the fold class is varied and modular, with 2.8% of sequences containing more than one Trx fold domain. Most member proteins are bacterial. The fold class exhibits many modifications to the CxxC active site motif—only 56.8% of proteins have both cysteines, and no functional groupings have absolute conservation of the expected catalytic motif. Only a small fraction of Trx fold sequences have been functionally characterized. This work provides a global view of the complex distribution of domains and catalytic machinery throughout the fold class, showing that each superfamily contains remnants of the CxxC active site. The unifying context provided by this work can guide the comparison of members of different Trx fold superfamilies to gain insight about their structure-function relationships, illustrated here with the thioredoxins and peroxiredoxins.     

Sequence-based prediction of protein domains

Guessing the boundaries of structural domains has been an important and challenging problem in experimental and computational structural biology. Predictions were based on intuition, biochemical properties, statistics, sequence homology and other aspects of predicted protein structure. Here, we introduced CHOPnet, a de novo method that predicts structural domains in the absence of homology to known domains. Our method was based on neural networks and relied exclusively on information available for all proteins. Evaluating sustained performance through rigorous cross-validation on proteins of known structure, we correctly predicted the number of domains in 69% of all proteins. For 50% of the two-domain proteins the centre of the predicted boundary was closer than 20 residues to the boundary assigned from three-dimensional (3D) structures; this was about eight percentage points better than predictions by ‘equal split’. Our results appeared to compare favourably with those from previously published methods. CHOPnet may be useful to restrict the experimental testing of different fragments for structure determination in the context of structural genomics.     

Prediction of Protein–Protein Interaction Sites in Sequences and 3D Structures by Random Forests

Identifying interaction sites in proteins provides important clues to the function of a protein and is becoming increasingly relevant in topics such as systems biology and drug discovery. Although there are numerous papers on the prediction of interaction sites using information derived from structure, there are only a few case reports on the prediction of interaction residues based solely on protein sequence. Here, a sliding window approach is combined with the Random Forests method to predict protein interaction sites using (i) a combination of sequence- and structure-derived parameters and (ii) sequence information alone. For sequence-based prediction we achieved a precision of 84% with a 26% recall and an F-measure of 40%. When combined with structural information, the prediction performance increases to a precision of 76% and a recall of 38% with an F-measure of 51%. We also present an attempt to rationalize the sliding window size and demonstrate that a nine-residue window is the most suitable for predictor construction. Finally, we demonstrate the applicability of our prediction methods by modeling the Ras–Raf complex using predicted interaction sites as target binding interfaces. Our results suggest that it is possible to predict protein interaction sites with quite a high accuracy using only sequence information.     

Prediction of Protein–Protein Interaction Sites in Sequences and 3D Structures by Random Forests

Identifying interaction sites in proteins provides important clues to the function of a protein and is becoming increasingly relevant in topics such as systems biology and drug discovery. Although there are numerous papers on the prediction of interaction sites using information derived from structure, there are only a few case reports on the prediction of interaction residues based solely on protein sequence. Here, a sliding window approach is combined with the Random Forests method to predict protein interaction sites using (i) a combination of sequence- and structure-derived parameters and (ii) sequence information alone. For sequence-based prediction we achieved a precision of 84% with a 26% recall and an F-measure of 40%. When combined with structural information, the prediction performance increases to a precision of 76% and a recall of 38% with an F-measure of 51%. We also present an attempt to rationalize the sliding window size and demonstrate that a nine-residue window is the most suitable for predictor construction. Finally, we demonstrate the applicability of our prediction methods by modeling the Ras–Raf complex using predicted interaction sites as target binding interfaces. Our results suggest that it is possible to predict protein interaction sites with quite a high accuracy using only sequence information.     

Fold recognition and ab initio structure predictions using hidden markov models and β-strand pair potentials

Protein structure predictions were submitted for 9 of the target sequences in the competition that ran during 1994. Targets sequences were selected that had no known homology with any sequence of known structure and were members of a reasonably sized family of related but divergent sequences. The objective was either to recognize a compatible fold for the target sequence in the database of known structures or to predict ab initio its rough 3D topology. The main tools used were Hidden Markov models (HMM) for fold recognition, a β- strand pair potential to predict β-sheet topology, and the PHD server for secondary structure prediction. Compatible folds were correctly identified in a number of cases and the β-strand pair potential was shown to be a useful tool for ab initio topology prediction. © 1995 Wiley-Liss, Inc.     

Analysis of the Thermotoga maritima genome combining a variety of sequence similarity and genome context tools

The proliferation of genome sequence data has led to the development of a number of tools and strategies that facilitate computational analysis. These methods include the identification of motif patterns, membership of the query sequences in family databases, metabolic pathway involvement and gene proximity. We re-examined the completely sequenced genome of Thermotoga maritima by employing the combined use of the above methods. By analyzing all 1877 proteins encoded in this genome, we identified 193 cases of conflicting annotations (10%), of which 164 are new function predictions and 29 are amendments of previously proposed assignments. These results suggest that the combined use of existing computational tools can resolve inconclusive sequence similarities and significantly improve the prediction of protein function from genome sequence.     

The ubiquitin domain superfold: structure-based sequence alignments and characterization of binding epitopes

Ubiquitin-like domains are present, apart from ubiquitin-like proteins themselves, in many multidomain proteins involved in different signal transduction processes. The sequence conservation for all ubiquitin superfold family members is rather poor, even between subfamily members, leading to mistakes in sequence alignments using conventional sequence alignment methods. However, a correct alignment is essential, especially for in silico methods that predict binding partners on the basis of sequence and structure. In this study, using 3D-structural information we have generated and manually corrected sequence alignments for proteins of the five ubiquitin superfold subfamilies. On the basis of this alignment, we suggest domains for which structural information will be useful to allow homology modelling. In addition, we have analysed the energetic and electrostatic properties of ubiquitin-like domains in complex with various functional binding proteins using the protein design algorithm FoldX. On the basis of an in silico alanine-scanning mutagenesis, we provide a detailed binding epitope mapping of the hotspots of the ubiquitin domain fold, involved in the interaction with different domains and proteins. Finally, we provide a consensus fingerprint sequence that identifies all sequences described to belong to the ubiquitin superfold family. It is possible that the method that we describe may be applied to other domain families sharing a similar fold but having low levels of sequence homology.     

Genome-wide identification of human functional DNA using a neutral indel model

It has become clear that a large proportion of functional DNA in the human genome does not code for protein. Identification of this non-coding functional sequence using comparative approaches is proving difficult and has previously been thought to require deep sequencing of multiple vertebrates. Here we introduce a new model and comparative method that, instead of nucleotide substitutions, uses the evolutionary imprint of insertions and deletions (indels) to infer the past consequences of selection. The model predicts the distribution of indels under neutrality, and shows an excellent fit to human–mouse ancestral repeat data. Across the genome, many unusually long ungapped regions are detected that are unaccounted for by the neutral model, and which we predict to be highly enriched in functional DNA that has been subject to purifying selection with respect to indels. We use the model to determine the proportion under indel-purifying selection to be between 2.56% and 3.25% of human euchromatin. Since annotated protein-coding genes comprise only 1.2% of euchromatin, these results lend further weight to the proposition that more than half the functional complement of the human genome is non-protein-coding. The method is surprisingly powerful at identifying selected sequence using only two or three mammalian genomes. Applying the method to the human, mouse, and dog genomes, we identify 90 Mb of human sequence under indel-purifying selection, at a predicted 10% false-discovery rate and 75% sensitivity. As expected, most of the identified sequence represents unannotated material, while the recovered proportions of known protein-coding and microRNA genes closely match the predicted sensitivity of the method. The method's high sensitivity to functional sequence such as microRNAs suggest that as yet unannotated microRNA genes are enriched among the sequences identified. Futhermore, its independence of substitutions allowed us to identify sequence that has been subject to heterogeneous selection, that is, sequence subject to both positive selection with respect to substitutions and purifying selection with respect to indels. The ability to identify elements under heterogeneous selection enables, for the first time, the genome-wide investigation of positive selection on functional elements other than protein-coding genes.     

Human Adeno-Associated Virus Type 5 Is Only Distantly Related to Other Known Primate Helper-Dependent Parvoviruses

We have characterized 95% (4,404 nucleotides) of the genome of adeno-associated virus type 5 (AAV5), including part of the terminal repeats and the terminal resolution site. Our results show that AAV5 is different from all other described AAV serotypes at the nucleotide level and at the amino acid level. The sequence homology to AAV2, AAV3B, AAV4, and AAV6 at the nucleotide level is only between 54 and 56%. The positive strand contains two large open reading frames (ORFs). The left ORF encodes the nonstructural (Rep) proteins, and the right ORF encodes the structural (Cap) proteins. At the amino acid level the identities with the capsid proteins of other AAVs range between 51 and 59%, with a high degree of heterogeneity in regions which are considered to be on the exterior surface of the viral capsid. The overall identity for the nonstructural Rep proteins at the amino acid level is 54.4%. It is lowest at the C-terminal 128 amino acids (10%). There are only two instead of the common three putative Zn fingers in the Rep proteins. The Cap protein data suggest differences in capsid surfaces and raise the possibility of a host range distinct from those of other parvoviruses. This may have important implications for AAV vectors used in gene therapy.     

PipeAlign: A new toolkit for protein family analysis

PipeAlign is a protein family analysis tool integrating a five step process ranging from the search for sequence homologues in protein and 3D structure databases to the definition of the hierarchical relationships within and between subfamilies. The complete, automatic pipeline takes a single sequence or a set of sequences as input and constructs a high-quality, validated MACS (multiple alignment of complete sequences) in which sequences are clustered into potential functional subgroups. For the more experienced user, the PipeAlign server also provides numerous options to run only a part of the analysis, with the possibility to modify the default parameters of each software module. For example, the user can choose to enter an existing multiple sequence alignment for refinement, validation and subsequent clustering of the sequences. The aim is to provide an interactive workbench for the validation, integration and presentation of a protein family, not only at the sequence level, but also at the structural and functional levels. PipeAlign is available at http://igbmc.u-strasbg.fr/PipeAlign/.