Protein Sequence Data Research Articles

Mask blast with a new chemical logic of amino acids for improved protein function prediction.

With the exponential increase in protein sequence data, there is an urgency to acquire a knowledge of function of the millions of sequences, using automated methods with high reliability. Conventional methods for annotating a protein sequence transfer the function of a homologous sequence with known functions based on evolutionary information. Here, we present a newer way of classifying amino acids based on chemical measures and demonstrate that, when integrated with mask BLAST, the chemical properties identified outperform current classifications of amino acids as well as evolutionary measures in function detection.

Generating functional protein variants with variational autoencoders.

The vast expansion of protein sequence databases provides an opportunity for new protein design approaches which seek to learn the sequence-function relationship directly from natural sequence variation. Deep generative models trained on protein sequence data have been shown to learn biologically meaningful representations helpful for a variety of downstream tasks, but their potential for direct use in the design of novel proteins remains largely unexplored. Here we show that variational autoencoders trained on a dataset of almost 70000 luciferase-like oxidoreductases can be used to generate novel, functional variants of the luxA bacterial luciferase. We propose separate VAE models to work with aligned sequence input (MSA VAE) and raw sequence input (AR-VAE), and offer evidence that while both are able to reproduce patterns of amino acid usage characteristic of the family, the MSA VAE is better able to capture long-distance dependencies reflecting the influence of 3D structure. To confirm the practical utility of the models, we used them to generate variants of luxA whose luminescence activity was validated experimentally. We further showed that conditional variants of both models could be used to increase the solubility of luxA without disrupting function. Altogether 6/12 of the variants generated using the unconditional AR-VAE and 9/11 generated using the unconditional MSA VAE retained measurable luminescence, together with all 23 of the less distant variants generated by conditional versions of the models; the most distant functional variant contained 35 differences relative to the nearest training set sequence. These results demonstrate the feasibility of using deep generative models to explore the space of possible protein sequences and generate useful variants, providing a method complementary to rational design and directed evolution approaches.

Deep Protein Subcellular Localization Predictor Enhanced with Transfer Learning of GO Annotation

Since the large‐scale protein sequence data is available, applying deep neural networks to mine better features from the sequences becomes possible. Eukaryotic protein subcellular localization prediction which makes a contribution in many biology process, has used protein sequences in many automatic predicting methods. Moreover, gene ontology (GO) annotation has been shown to be helpful in improving the prediction accuracy of subcellular localization. However, experimentally annotated proteins are not always available. On the other hand, experimentally annotated proteins are available for certain species such as human, mouse, Arabidopsis thaliana, etc. It is highly motivated to perform deep learning of GO annotations on the available experimentally annotated proteins and to transfer it to subcellular localization prediction on other species. In this paper, we propose a deep protein subcellular localization predictor, consisting of a linear classifier and a deep feature extractor of convolution neural network (CNN). The deep CNN feature extractor is first shared and pre‐trained in a deep GO annotation predictor, and then is transferred to the subcellular localization predictor with fine‐tuning using protein localization samples. In this way, we have a deep protein subcellular localization predictor enhanced with transfer learning of GO annotation. The proposed method has good performances on the Swiss‐Prot datasets, when transfer learning using the protein samples both within and out species. Moreover, it outperforms the state‐of‐the‐art traditional methods on benchmark datasets. © 2021 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC.

An Investigation of Alternatives to Transform Protein Sequence Databases to a Columnar Index Schema

Mass spectrometers enable identifying proteins in biological samples leading to biomarkers for biological process parameters and diseases. However, bioinformatic evaluation of the mass spectrometer data needs a standardized workflow and system that stores the protein sequences. Due to its standardization and maturity, relational systems are a great fit for storing protein sequences. Hence, in this work, we present a schema for distributed column-based database management systems using a column-oriented index to store sequence data. In order to achieve a high storage performance, it was necessary to choose a well-performing strategy for transforming the protein sequence data from the FASTA format to the new schema. Therefore, we applied an in-memory map, HDDmap, database engine, and extended radix tree and evaluated their performance. The results show that our proposed extended radix tree performs best regarding memory consumption and runtime. Hence, the radix tree is a suitable data structure for transforming protein sequences into the indexed schema.

A Simple Protein Evolutionary Classification Method Based on the Mutual Relations Between Protein Sequences

Background: Protein is a kind of important organics in life. It is varied with its sequences, structures and functions. Protein evolutionary classification is one of the popular research topics in computational bioinformatics. Many studies have used protein sequence information to classify the evolutionary relationships of proteins. As the amount of protein sequence data increases, efficient computational tools are needed to make efficient protein evolutionary classifications with high accuracies in the big data paradigm. Methods: In this study, we propose a new simple and efficient computational approach based on the normalized mutual information rates to compute the relationship between protein sequences, we then use the “distances” defined on the relationships to perform the evolutionary classifications of proteins. The new method is computational efficient, model-free and unsupervised, which does not require training data when performing classifications. Result: Simulation studies on various examples demonstrate the efficiency of the new method. We use precision-recall curves to compare the efficiency of our new method with traditional methods, results show that the new method outperforms the traditional methods in most of the cases when performing evolutionary classifications. Conclusion: The new method is simple and proved to be efficient in protein evolutionary classifications, which is useful in future evolutionary analysis particularly in the big data paradigm.

Sequence representation approaches for sequence-based protein prediction tasks that use deep learning.

Deep learning has been increasingly used in bioinformatics, especially in sequence-based protein prediction tasks, as large amounts of biological data are available and deep learning techniques have been developed rapidly in recent years. For sequence-based protein prediction tasks, the selection of a suitable model architecture is essential, whereas sequence data representation is a major factor in controlling model performance. Here, we summarized all the main approaches that are used to represent protein sequence data (amino acid sequence encoding or embedding), which include end-to-end embedding methods, non-contextual embedding methods and embedding methods that use transfer learning and others that are applied for some specific tasks (such as protein sequence embedding based on extracted features for protein structure predictions and graph convolutional network-based embedding for drug discovery tasks). We have also reviewed the architectures of various types of embedding models theoretically and the development of these types of sequence embedding approaches to facilitate researchers and users in selecting the model that best suits their requirements.

Homology modeling and molecular docking simulation of some novel imidazo[1,2-a]pyridine-3-carboxamide (IPA) series as inhibitors of Mycobacterium tuberculosis

BackgroundTuberculosis (TB) remains a serious global health challenge that is caused by Mycobacterium tuberculosis and has killed numerous people. This necessitated the urgent need for the hunt and development of more potent drugs against the fast-emerging extensively drug-resistant (XDR) and multiple-drug-resistant (MDR) M. tuberculosis strains. Mycobacterium tuberculosis cytochrome b subunit of the cytochrome bc1 complex (QcrB) was recognized as a potential drug target in M. tuberculosis (25618/H37Rv) for imidazo[1,2-a]pyridine-3-carboxamides whose crystal strucuture is not yet reported in the Protein Data Bank (PDB). The concept of homology modeling as a powerful and useful computational method can be applied, since the M. tuberculosis QcrB protein sequence data are available. ResultsThe homology model of QcrB protein in M. tuberculosis was built from the X-ray structure of QcrB in M. smegmatis as a template using the Swiss-Model online workspace. The modeled protein was assessed, validated, and prepared for the molecular docking simulation of 35 ligands of N-(2-phenoxy)ethyl imidazo[1,2-a] pyridine-3-carboxamide (IPA) to analyze their theoretical binding affinities and modes. The docking results showed that the binding affinity values ranged from − 6.5 to − 10.1 kcal/mol which confirms their resilience potency when compared with 6.0kcal/mol of isoniazid standard drug. However, ligands 2, 7, 22, 26, and 35 scored higher binding affinity values of − 9.60, − 9.80, − 10.10, − 10.00, and − 10.00 kcal/mol, and are respectively considered as the best ligands among others with better binding modes in the active site of the modeled QcrB protein. ConclusionThe information derived in this research revealed some potential hits and paved a route for structure-based drug discovery of new hypothetical imidazo pyridine amide analogs as anti-tubercular drug candidates.

Predicting bacteriophage hosts based on sequences of annotated receptor-binding proteins

Nowadays, bacteriophages are increasingly considered as an alternative treatment for a variety of bacterial infections in cases where classical antibiotics have become ineffective. However, characterizing the host specificity of phages remains a labor- and time-intensive process. In order to alleviate this burden, we have developed a new machine-learning-based pipeline to predict bacteriophage hosts based on annotated receptor-binding protein (RBP) sequence data. We focus on predicting bacterial hosts from the ESKAPE group, Escherichia coli, Salmonella enterica and Clostridium difficile. We compare the performance of our predictive model with that of the widely used Basic Local Alignment Search Tool (BLAST). Our best-performing predictive model reaches Precision-Recall Area Under the Curve (PR-AUC) scores between 73.6 and 93.8% for different levels of sequence similarity in the collected data. Our model reaches a performance comparable to that of BLASTp when sequence similarity in the data is high and starts outperforming BLASTp when sequence similarity drops below 75%. Therefore, our machine learning methods can be especially useful in settings in which sequence similarity to other known sequences is low. Predicting the hosts of novel metagenomic RBP sequences could extend our toolbox to tune the host spectrum of phages or phage tail-like bacteriocins by swapping RBPs.

Distinguishing Enzymes and Non-enzymes Based on Structural Information with an Alignment Free Approach

Background: Knowledge of protein functions is very crucial for the understanding of biological processes. Experimental methods for protein function prediction are powerless to treat the growing amount of protein sequence and structure data. Objective: To develop some computational techniques for the protein function prediction. Method: Based on the residue interaction network features and the motion mode information, an SVM model was constructed and used as the predictor. The role of these features was analyzed and some interesting results were obtained. Results: An alignment-free method for the classification of enzyme and non-enzyme is developed in this work. There is not any single feature that occupies a dominant position in the prediction process. The topological and the information-theoretic residue interaction network features have a better performance. The combination of the fast mode and the slow mode can get a better explanation for the classification result. Conclusion: The method proposed in this paper can act as a classifier for the enzymes and nonenzymes.

Performance of Regression Models as a Function of Experiment Noise.

Background:A challenge in developing machine learning regression models is that it is difficult to know whether maximal performance has been reached on the test dataset, or whether further model improvement is possible. In biology, this problem is particularly pronounced as sample labels (response variables) are typically obtained through experiments and therefore have experiment noise associated with them. Such label noise puts a fundamental limit to the metrics of performance attainable by regression models on the test dataset.Results:We address this challenge by deriving an expected upper bound for the coefficient of determination (R2) for regression models when tested on the holdout dataset. This upper bound depends only on the noise associated with the response variable in a dataset as well as its variance. The upper bound estimate was validated via Monte Carlo simulations and then used as a tool to bootstrap performance of regression models trained on biological datasets, including protein sequence data, transcriptomic data, and genomic data.Conclusions:The new method for estimating upper bounds for model performance on test data should aid researchers in developing ML regression models that reach their maximum potential. Although we study biological datasets in this work, the new upper bound estimates will hold true for regression models from any research field or application area where response variables have associated noise.

DGraph Clusters Flaviviruses and β-Coronaviruses According to Their Hosts, Disease Type, and Human Cell Receptors.

Motivation:There is a need for rapid and easy-to-use, alignment-free methods to cluster large groups of protein sequence data. Commonly used phylogenetic trees based on alignments can be used to visualize only a limited number of protein sequences. DGraph, introduced here, is an application developed to generate 2-dimensional (2D) maps based on similarity scores for sequences. The program automatically calculates and graphically displays property distance (PD) scores based on physico-chemical property (PCP) similarities from an unaligned list of FASTA files. Such “PD-graphs” show the interrelatedness of the sequences, whereby clusters can reveal deeper connectivities.Results:Property distance graphs generated for flavivirus (FV), enterovirus (EV), and coronavirus (CoV) sequences from complete polyproteins or individual proteins are consistent with biological data on vector types, hosts, cellular receptors, and disease phenotypes. Property distance graphs separate the tick- from the mosquito-borne FV, cluster viruses that infect bats, camels, seabirds, and humans separately. The clusters correlate with disease phenotype. The PD method segregates the β-CoV spike proteins of severe acute respiratory syndrome (SARS), severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), and Middle East respiratory syndrome (MERS) sequences from other human pathogenic CoV, with clustering consistent with cellular receptor usage. The graphs also suggest evolutionary relationships that may be difficult to determine with conventional bootstrapping methods that require postulating an ancestral sequence.

A computational approach for predicting drug–target interactions from protein sequence and drug substructure fingerprint information

Identification of drug–target interactions (DTIs) is critical for discovering potential target protein candidates for new drugs. However, traditional experimental methods have limitations in discovering DTIs. They are time-consuming, tedious, and expensive, and often suffer from high false-positive rates and false-negative rates. Therefore, using computational methods to predict DTIs has received extensive attention from many researchers in recent years. To address this issue, in this paper, an effective prediction model is presented which is based on the information of drug molecular structure data and protein sequence data. It performs prediction with the following procedures. First, we transform the sequences of each target into a position-specific scoring matrix (PSSM), such that the features can retain biological evolutionary information. We then use a feature vector of molecular substructure fingerprints to describe the chemical structure information of the drug compounds. Second, the Legendre moments algorithm is used to extract new features from the PSSM. Finally, a classification algorithm called rotation forest is used to perform prediction, we tested its prediction performance on four golden standard data sets: enzymes, G-protein-coupled receptors, ion channels, and nuclear receptors. As a result, the proposed method achieves average accuracies of 0.9026, 0.8260, 0.8703, and 0.7444 on these four data sets using five-fold cross-validation. We also compare the proposed method with the support vector machine and other existing approaches. The proposed model is proved to be superior to comparative methods, showing that it is feasible, effective, and robust for predicting potential DTI.

A novel positive single-stranded RNA virus from the crustacean parasite, Probopyrinella latreuticola (Peracarida: Isopoda: Bopyridae)

A positive, single-stranded RNA virus is identified from the transcriptome of Probopyrinella latreuticola Gissler, 1882; a bopyrid isopod parasite of the Sargassum shrimp, Latreutes fucorum Fabricius, 1789. The viral sequence is 13,098 bp in length (including polyA), encoding four open reading frames (ORF). ORF-1 encodes a polyprotein, with three computationally discernible functional domains: viral methyltransferase; viral helicase; and RNA-directed RNA polymerase. The remaining ORFs encode a transmembrane protein, a capsid protein and a protein of undetermined function. The raw transcriptomic data reveal a low level of background single nucleotide mutations within the data. Comparison of the protein sequence data and synteny with other viral isolates reveals that the greatest protein similarity (<39%) is shared with the Negevirus group, a group that exclusively infects insects.Phylogenetic assessment of the individual polyprotein domains revealed a mixed prediction of phylogenetic origins, suggesting with low confidence that the novel +ssRNA virus could be present in multiple places throughout the individual gene trees. A concatenated approach strongly suggested that this new virus is an early diverging isolate, branching before the Negevirus and Cilevirus groups. Alongside the new isolate are other marine viruses, also present toward the base of the tree.The isopod virosphere, with the addition of this novel virus, is discussed relative to viral genomics/systematics. A great diversity of nege-like viruses appears to be present in marine invertebrate hosts, which require greater efforts for discovery and identification.

HumDLoc: Human Protein Subcellular Localization Prediction Using Deep Neural Network.

Aims To develop a tool that can annotate subcellular localization of human proteins.Background With the progression of high throughput human proteomics projects, an enormous amount of protein sequence data has been discovered in the recent past. All these raw sequence data require precise mapping and annotation for their respective biological role and functional attributes. The functional characteristics of protein molecules are highly dependent on the subcellular localization/compartment. Therefore, a fully automated and reliable protein subcellular localization prediction system would be very useful for current proteomic research.Objective To develop a machine learning-based predictive model that can annotate the subcellular localization of human proteins with high accuracy and precision.Methods In this study, we used the PSI-CD-HIT homology criterion and utilized the sequence-based features of protein sequences to develop a powerful subcellular localization predictive model. The dataset used to train the HumDLoc model was extracted from a reliable data source, Uniprot knowledge base, which helps the model to generalize on the unseen dataset.Results The proposed model, HumDLoc, was compared with two of the most widely used techniques: CELLO and DeepLoc, and other machine learning-based tools. The result demonstrated promising predictive performance of HumDLoc model based on various machine learning parameters such as accuracy (≥97.00%), precision (≥0.86), recall (≥0.89), MCC score (≥0.86), ROC curve (0.98 square unit), and precision-recall curve (0.93 square unit).Conclusion In conclusion, HumDLoc was able to outperform several alternative tools for correctly predicting subcellular localization of human proteins. The HumDLoc has been hosted as a web-based tool at https://bioserver.iiita.ac.in/HumDLoc/.

Self-evoluting framework of deep convolutional neural network for multilocus protein subcellular localization.

In the present paper, deep convolutional neural network (DCNN) is applied to multilocus protein subcellular localization as it is more suitable for multi-class classification. There are two main problems with this application. First, the appropriate features for correlation between multiple sites are hard to find. Second, the classifier structure is difficult to determine as it is greatly affected by the distribution of classified data. To solve these problems, a self-evoluting framework using DCNNs for multilocus protein subcellular localization is proposed. It has three characteristics that the previous algorithms do not. The first is that it combines the ant colony algorithm with the DCNN to form a self-evoluting algorithm for multilocus protein subcellular localization. The second is that it randomly groups subcellular sites using a limited random k-labelsets multi-label classification method. It also solves complex problems in a divide-and-conquer approach and proposes a flexible expansion model. The third is that it realizes the random selection feature extraction method in the positioning process and avoids the defects in individual feature extraction methods. The algorithm in the present paper is tested on the human database, and the overall correct rate is 67.17%, which is higher than that for the stacked self-encoder (SAE), support vector machine (SVM), random forest classifier (RF), or single deep convolutional neural network.Graphical abstract The algorithm mentioned in the present paper mainly includes four parts. They are protein sequence data preprocessing, integrated DCNN model construction, finding optimal DCNN combination by ant colony optimization, and protein subcellular localization for sequences. These parts are sequential relationships and the data obtained in the previous part is the basis for the latter part of the function. In the part of data preprocessing, the limited RAkEL multi-label classification method is used to randomly group subcellular sites. At the same time, the feature fusion of protein sequences is carried out by using multiple feature extraction methods. Each combination including features and sites information corresponds to a DCNN model. In the part of finding optimal DCNN combination by ant colony optimization, the main purpose is to find the best combination of DCNN models through the global optimization ability of the ant colony algorithm. The positioning of sequences is mainly to obtain multilocus subcellular localization by the optimal model combination.

Protein Sequence Selection Method That Enables Full Consensus Design of Artificial l-Threonine 3-Dehydrogenases with Unique Enzymatic Properties.

Exponentially increasing protein sequence data enables artificial enzyme design using sequence-based protein design methods, including full-consensus protein design (FCD). The success of artificial enzyme design is strongly dependent on the nature of the sequences used. Hence, sequences must be selected from databases and curated libraries prepared to enable a successful design by FCD. In this study, we proposed a selection approach regarding several key residues as sequence motifs. We used l-threonine 3-dehydrogenase (TDH) as a model to test the validity of this approach. In the classification, four residues (143, 174, 188, and 214) were used as key residues. We classified thousands of TDH homologous sequences into five groups containing hundreds of sequences. Utilizing sequences in the libraries, we designed five artificial TDHs by FCD. Among the five, we successfully expressed four in soluble form. Biochemical analysis of artificial TDHs indicated that their enzymatic properties vary; half of the maximum measured enzyme activity (t1/2) and activation energies were distributed from 53 to 65 °C and from 38 to 125 kJ/mol, respectively. The artificial TDHs had unique kinetic parameters, distinct from one another. Structural analysis indicates that consensus mutations are mainly introduced in the secondary or outer shell. The functional diversity of the artificial TDHs is due to the accumulation of mutations that affect their physicochemical properties. Taken together, our findings indicate that our proposed approach can help generate artificial enzymes with unique enzymatic properties.

Assessment of different screening methods for selecting palaeontological bone samples for peptide sequencing

Ancient proteomics is being applied to samples dating further and further back in time, with many palaeontological specimens providing protein sequence data for phylogenetic analysis as well as protein degradation studies. However, fossils are a precious material and proteomic analysis is destructive and costly. In this paper we consider three different techniques (ATR-FTIR, MALDI-ToF MS and chiral AA analysis) to screen fossil material for potential protein preservation, aiming to maximise the proteomic information recovered and saving costly time consuming analyses which may produce low quality results. It was found that splitting factor and C/P indices from ATR-FTIR were not a reliable indicator of protein survival as they are confounded by secondary mineralisation of the fossil material. Both MALDI-ToF MS and chiral AA analysis results were able to successfully identify samples with surviving proteins, and it is suggested that one or both of these analyses be used for screening palaeontological specimens. SignificanceThis study has shown both chiral amino acid analysis and MALDI-ToF MS are reliable screening methods for predicting protein survival in fossils. Both these methods are quick, cheap, minimally destructive (1 mg and 15 mg respectively) and can provide crucial additional information about the endogeneity of the surviving proteins. It is hoped that the use of these screening methods will encourage the examination of a wide range of palaeontological specimens for potential proteomic analysis. This in turn will give us a better understanding of protein survival far back in time and under different environmental conditions.

Engineering a Histone Reader Protein by Combining Directed Evolution, Sequencing, and Neural Network Based Ordinal Regression.

Directed evolution is a powerful approach for engineering proteins with enhanced affinity or specificity for a ligand of interest but typically requires many rounds of screening/library mutagenesis to obtain mutants with desired properties. Furthermore, mutant libraries generally only cover a small fraction of the available sequence space. Here, for the first time, we use ordinal regression to model protein sequence data generated through successive rounds of sorting and amplification of a protein-ligand system. We show that the ordinal regression model trained on only two sorts successfully predicts chromodomain CBX1 mutants that would have stronger binding affinity with the H3K9me3 peptide. Furthermore, we can extract the predictive features using contextual regression, a method to interpret nonlinear models, which successfully guides identification of strong binders not even present in the original library. We have demonstrated the power of this approach by experimentally confirming that we were able to achieve the same improvement in binding affinity previously achieved through a more laborious directed evolution process. This study presents an approach that reduces the number of rounds of selection required to isolate strong binders and facilitates the identification of strong binders not present in the original library.

A simple strategy to enhance the speed of protein secondary structure prediction without sacrificing accuracy.

The secondary structure prediction of proteins is a classic topic of computational structural biology with a variety of applications. During the past decade, the accuracy of prediction achieved by state-of-the-art algorithms has been >80%; meanwhile, the time cost of prediction increased rapidly because of the exponential growth of fundamental protein sequence data. Based on literature studies and preliminary observations on the relationships between the size/homology of the fundamental protein dataset and the speed/accuracy of predictions, we raised two hypotheses that might be helpful to determine the main influence factors of the efficiency of secondary structure prediction. Experimental results of size and homology reductions of the fundamental protein dataset supported those hypotheses. They revealed that shrinking the size of the dataset could substantially cut down the time cost of prediction with a slight decrease of accuracy, which could be increased on the contrary by homology reduction of the dataset. Moreover, the Shannon information entropy could be applied to explain how accuracy was influenced by the size and homology of the dataset. Based on these findings, we proposed that a proper combination of size and homology reductions of the protein dataset could speed up the secondary structure prediction while preserving the high accuracy of state-of-the-art algorithms. Testing the proposed strategy with the fundamental protein dataset of the year 2018 provided by the Universal Protein Resource, the speed of prediction was enhanced over 20 folds while all accuracy measures remained equivalently high. These findings are supposed helpful for improving the efficiency of researches and applications depending on the secondary structure prediction of proteins. To make future implementations of the proposed strategy easy, we have established a database of size and homology reduced protein datasets at http://10.life.nctu.edu.tw/UniRefNR.

Homology Modeling of Distant Lipocalin Homologs Using a Structure-based Fingerprint as a Constraint for Sequence Alignment

The incessant expansion of protein sequence data presents a huge challenge in the identification of their structures which relies on laborious and time-consuming experimental approaches. Homology modeling of protein structure is highly accepted as an alternative to experimental approach, but its efficacy tends to decline when the sequence homology between the target and template sequences is low due to the limitation of their sequence alignment. Here, we demonstrated that employing a structure-based fingerprint of a homologous protein set as a constraint for the alignment of distant template and target sequences could improve the efficiency of homology modeling by using lipocalin proteins as a model system. Nine distantly related kernel lipocalins were structurally aligned, which was used to design a structure-based fingerprint consisting of commonly conserved hydrophobic and hydrophilic residues. The fingerprint was introduced as constraint in the alignment of template and target lipocalins, and the homology modeling of target lipocalin was performed based on the aligned sequences. Although the efficiency was marginal, the approach showed a consistent improvement compared to the homology modeling of lipocalin structures without the constraint for sequence alignment.