Consistent probabilistic outputs for protein function prediction

In predicting hierarchical protein function annotations, such as terms in the Gene Ontology (GO), the simplest approach makes predictions for each term independently. However, this approach has the unfortunate consequence that the predictor may assign to a single protein a set of terms that are inconsistent with one another; for example, the predictor may assign a specific GO term to a given protein ('purine nucleotide binding') but not assign the parent term ('nucleotide binding'). Such predictions are difficult to interpret. In this work, we focus on methods for calibrating and combining independent predictions to obtain a set of probabilistic predictions that are consistent with the topology of the ontology. We call this procedure 'reconciliation'. We begin with a baseline method for predicting GO terms from a collection of data types using an ensemble of discriminative classifiers. We apply the method to a previously described benchmark data set, and we demonstrate that the resulting predictions are frequently inconsistent with the topology of the GO. We then consider 11 distinct reconciliation methods: three heuristic methods; four variants of a Bayesian network; an extension of logistic regression to the structured case; and three novel projection methods - isotonic regression and two variants of a Kullback-Leibler projection method. We evaluate each method in three different modes - per term, per protein and joint - corresponding to three types of prediction tasks. Although the principal goal of reconciliation is interpretability, it is important to assess whether interpretability comes at a cost in terms of precision and recall. Indeed, we find that many apparently reasonable reconciliation methods yield reconciled probabilities with significantly lower precision than the original, unreconciled estimates. On the other hand, we find that isotonic regression usually performs better than the underlying, unreconciled method, and almost never performs worse; isotonic regression appears to be able to use the constraints from the GO network to its advantage. An exception to this rule is the high precision regime for joint evaluation, where Kullback-Leibler projection yields the best performance.     

A probabilistic fragment-based protein structure prediction algorithm

Conformational sampling is one of the bottlenecks in fragment-based protein structure prediction approaches. They generally start with a coarse-grained optimization where mainchain atoms and centroids of side chains are considered, followed by a fine-grained optimization with an all-atom representation of proteins. It is during this coarse-grained phase that fragment-based methods sample intensely the conformational space. If the native-like region is sampled more, the accuracy of the final all-atom predictions may be improved accordingly. In this work we present EdaFold, a new method for fragment-based protein structure prediction based on an Estimation of Distribution Algorithm. Fragment-based approaches build protein models by assembling short fragments from known protein structures. Whereas the probability mass functions over the fragment libraries are uniform in the usual case, we propose an algorithm that learns from previously generated decoys and steers the search toward native-like regions. A comparison with Rosetta AbInitio protocol shows that EdaFold is able to generate models with lower energies and to enhance the percentage of near-native coarse-grained decoys on a benchmark of [Formula: see text] proteins. The best coarse-grained models produced by both methods were refined into all-atom models and used in molecular replacement. All atom decoys produced out of EdaFold's decoy set reach high enough accuracy to solve the crystallographic phase problem by molecular replacement for some test proteins. EdaFold showed a higher success rate in molecular replacement when compared to Rosetta. Our study suggests that improving low resolution coarse-grained decoys allows computational methods to avoid subsequent sampling issues during all-atom refinement and to produce better all-atom models. EdaFold can be downloaded from http://www.riken.jp/zhangiru/software/.     

Comparison of probabilistic combination methods for protein secondary structure prediction

MOTIVATION: Protein secondary structure prediction is an important step towards understanding how proteins fold in three dimensions. Recent analysis by information theory indicates that the correlation between neighboring secondary structures are much stronger than that of neighboring amino acids. In this article, we focus on the combination problem for sequences, i.e. combining the scores or assignments from single or multiple prediction systems under the constraint of a whole sequence, as a target for improvement in protein secondary structure prediction. RESULTS: We apply several graphical chain models to solve the combination problem and show that they are consistently more effective than the traditional window-based methods. In particular, conditional random fields (CRFs) moderately improve the predictions for helices and, more importantly, for beta sheets, which are the major bottleneck for protein secondary structure prediction.     

A probabilistic model for secondary structure prediction from protein chemical shifts

Protein chemical shifts encode detailed structural information that is difficult and computationally costly to describe at a fundamental level. Statistical and machine learning approaches have been used to infer correlations between chemical shifts and secondary structure from experimental chemical shifts. These methods range from simple statistics such as the chemical shift index to complex methods using neural networks. Notwithstanding their higher accuracy, more complex approaches tend to obscure the relationship between secondary structure and chemical shift and often involve many parameters that need to be trained. We present hidden Markov models (HMMs) with Gaussian emission probabilities to model the dependence between protein chemical shifts and secondary structure. The continuous emission probabilities are modeled as conditional probabilities for a given amino acid and secondary structure type. Using these distributions as outputs of first‐ and second‐order HMMs, we achieve a prediction accuracy of 82.3%, which is competitive with existing methods for predicting secondary structure from protein chemical shifts. Incorporation of sequence‐based secondary structure prediction into our HMM improves the prediction accuracy to 84.0%. Our findings suggest that an HMM with correlated Gaussian distributions conditioned on the secondary structure provides an adequate generative model of chemical shifts. Proteins 2013; © 2012 Wiley Periodicals, Inc.     

Network-based auto-probit modeling for protein function prediction

Predicting the functional roles of proteins based on various genome-wide data, such as protein-protein association networks, has become a canonical problem in computational biology. Approaching this task as a binary classification problem, we develop a network-based extension of the spatial auto-probit model. In particular, we develop a hierarchical Bayesian probit-based framework for modeling binary network-indexed processes, with a latent multivariate conditional autoregressive Gaussian process. The latter allows for the easy incorporation of protein-protein association network topologies-either binary or weighted-in modeling protein functional similarity. We use this framework to predict protein functions, for functions defined as terms in the Gene Ontology (GO) database, a popular rigorous vocabulary for biological functionality. Furthermore, we show how a natural extension of this framework can be used to model and correct for the high percentage of false negative labels in training data derived from GO, a serious shortcoming endemic to biological databases of this type. Our method performance is evaluated and compared with standard algorithms on weighted yeast protein-protein association networks, extracted from a recently developed integrative database called Search Tool for the Retrieval of INteracting Genes/proteins (STRING). Results show that our basic method is competitive with these other methods, and that the extended method-incorporating the uncertainty in negative labels among the training data-can yield nontrivial improvements in predictive accuracy.     

Machine learning techniques for protein function prediction

Proteins play important roles in living organisms, and their function is directly linked with their structure. Due to the growing gap between the number of proteins being discovered and their functional characterization (in particular as a result of experimental limitations), reliable prediction of protein function through computational means has become crucial. This paper reviews the machine learning techniques used in the literature, following their evolution from simple algorithms such as logistic regression to more advanced methods like support vector machines and modern deep neural networks. Hyperparameter optimization methods adopted to boost prediction performance are presented. In parallel, the metamorphosis in the features used by these algorithms from classical physicochemical properties and amino acid composition, up to text-derived features from biomedical literature and learned feature representations using autoencoders, together with feature selection and dimensionality reduction techniques, are also reviewed. The success stories in the application of these techniques to both general and specific protein function prediction are discussed.     

Scoring function accuracy for membrane protein structure prediction

We perform a systematic examination of the ability of several different high-resolution, atomic-detail scoring functions to discriminate native conformations of loops in membrane proteins from non-native but physically reasonable, or "decoy," conformations. Decoys constructed from changing a loop conformation while keeping the remainder of the protein fixed are a challenging test of energy function accuracy. Nevertheless, the best of the energy functions we examined recognized the native structure as lowest in energy around half the time, and consistently chose it as a low-energy structure. This suggests that the best of present energy functions, even without a representation of the lipid bilayer, are of sufficient accuracy to give reasonable confidence in predictions of membrane protein structure. We also constructed homology models for each structure, using other known structures in the same protein family as templates. Homology models were constructed using several scoring functions and modeling programs, but with a comparable sampling effort for each procedure. Our results indicate that the quality of sequence alignment is probably the most important factor in model accuracy for sequence identity from 20-40%; one can expect a reasonably accurate model for membrane proteins when sequence identity is greater than 30%, in agreement with previous studies. Most errors are localized in loop regions, which tend to be found outside the lipid bilayer. For the most discriminative energy functions, it appears that errors are most likely due to lack of sufficient sampling, although it should be stressed that present energy functions are still far from perfectly reliable.     

面向蛋白质功能位点识别的机器学习平台构建

有关蛋白质功能的研究是解析生命奥秘的基础,机器学习技术在该领域已有广泛应用。利用支持向量机(support vectormachine,SVM)方法,构建一个预测蛋白质功能位点的通用平台。该平台先提取非同源蛋白质序列,再对这些序列进行特征编码(包括序列的基本信息、物化特征、结构信息及序列保守性特征等),以编码好的样本作为训练数据,利用SVM进行训练,得到敏感性、特异性、Matthew相关系数、准确率及ROC曲线等评价指标,反复测试,得到评价指标最优的SVM模型后,便可以用来预测蛋白质序列上的功能位点。该平台除了应用在预测蛋白质功能位点之外,还可以应用于疾病相关单核苷酸多态性(SNP)预测分析、预测蛋白质结构域分析、生物分子间的相互作用等。     

Using search engine technology for protein function prediction

Prediction of protein function is one of the most challenging problems in the post-genomic era. In this paper, we propose a novel algorithm Improved ProteinRank (IPR) for protein function prediction, which is based on the search engine technology and the preferential attachment criteria. In addition, an improved algorithm IPRW is developed from IPR to be used in the weighted protein?protein interaction (PPI) network. The proposed algorithms IPR and IPRW are applied to the PPI network of S.cerevisiae. The experimental results show that both IPR and IPRW outweigh the previous methods for the prediction of protein functions.     

A partially function-to-topic model for protein function prediction

Background

Proteins are a kind of macromolecules and the main component of a cell, and thus it is the most essential and versatile material of life. The research of protein functions is of great significance in decoding the secret of life. In recent years, researchers have introduced multi-label supervised topic model such as Labeled Latent Dirichlet Allocation (Labeled-LDA) into protein function prediction, which can obtain more accurate and explanatory prediction. However, the topic-label corresponding way of Labeled-LDA is associating each label (GO term) with a corresponding topic directly, which makes the latent topics to be completely degenerated, and ignores the differences between labels and latent topics.

Result

To achieve more accurate probabilistic modeling of function label, we propose a Partially Function-to-Topic Prediction (PFTP) model for introducing the local topics subset corresponding to each function label. Meanwhile, PFTP not only supports latent topics subset within a given function label but also a background topic corresponding to a ‘fake’ function label, which represents common semantic of protein function. Related definitions and the topic modeling process of PFTP are described in this paper. In a 5-fold cross validation experiment on yeast and human datasets, PFTP significantly outperforms five widely adopted methods for protein function prediction. Meanwhile, the impact of model parameters on prediction performance and the latent topics discovered by PFTP are also discussed in this paper.

Conclusion

All of the experimental results provide evidence that PFTP is effective and have potential value for predicting protein function. Based on its ability of discovering more-refined latent sub-structure of function label, we can anticipate that PFTP is a potential method to reveal a deeper biological explanation for protein functions.

     

Predicting protein function from protein/protein interaction data: a probabilistic approach

MOTIVATION:The development of experimental methods for genome scale analysis of molecular interaction networks has made possible new approaches to inferring protein function. This paper describes a method of assigning functions based on a probabilistic analysis of graph neighborhoods in a protein-protein interaction network. The method exploits the fact that graph neighbors are more likely to share functions than nodes which are not neighbors. A binomial model of local neighbor function labeling probability is combined with a Markov random field propagation algorithm to assign function probabilities for proteins in the network. RESULTS: We applied the method to a protein-protein interaction dataset for the yeast Saccharomyces cerevisiae using the Gene Ontology (GO) terms as function labels. The method reconstructed known GO term assignments with high precision, and produced putative GO assignments to 320 proteins that currently lack GO annotation, which represents about 10% of the unlabeled proteins in S. cerevisiae.     

Diffusion kernel-based logistic regression models for protein function prediction

Assigning functions to unknown proteins is one of the most important problems in proteomics. Several approaches have used protein-protein interaction data to predict protein functions. We previously developed a Markov random field (MRF) based method to infer a protein's functions using protein-protein interaction data and the functional annotations of its protein interaction partners. In the original model, only direct interactions were considered and each function was considered separately. In this study, we develop a new model which extends direct interactions to all neighboring proteins, and one function to multiple functions. The goal is to understand a protein's function based on information on all the neighboring proteins in the interaction network. We first developed a novel kernel logistic regression (KLR) method based on diffusion kernels for protein interaction networks. The diffusion kernels provide means to incorporate all neighbors of proteins in the network. Second, we identified a set of functions that are highly correlated with the function of interest, referred to as the correlated functions, using the chi-square test. Third, the correlated functions were incorporated into our new KLR model. Fourth, we extended our model by incorporating multiple biological data sources such as protein domains, protein complexes, and gene expressions by converting them into networks. We showed that the KLR approach of incorporating all protein neighbors significantly improved the accuracy of protein function predictions over the MRF model. The incorporation of multiple data sets also improved prediction accuracy. The prediction accuracy is comparable to another protein function classifier based on the support vector machine (SVM), using a diffusion kernel. The advantages of the KLR model include its simplicity as well as its ability to explore the contribution of neighbors to the functions of proteins of interest.     

Sequential linear neighborhood propagation for semi-supervised protein function prediction

Predicting protein function is one of the most challenging problems of the post-genomic era. The development of experimental methods for genome scale analysis of molecular interaction networks has provided new approaches to inferring protein function. In this paper we introduce a new graph-based semi-supervised classification algorithm Sequential Linear Neighborhood Propagation (SLNP), which addresses the problem of the classification of partially labeled protein interaction networks. The proposed SLNP first constructs a sequence of node sets according to their shortest distance to the labeled nodes, and then predicts the function of the unlabel proteins from the set closer to labeled one, using Linear Neighborhood Propagation. Its performance is assessed on the Saccharomyces cerevisiae PPI network data sets, with good results compared with three current state-of-the-art algorithms, especially in settings where only a small fraction of the proteins are labeled.     

A protocol for computer-based protein structure and function prediction

Genome sequencing projects have ciphered millions of protein sequence, which require knowledge of their structure and function to improve the understanding of their biological role. Although experimental methods can provide detailed information for a small fraction of these proteins, computational modeling is needed for the majority of protein molecules which are experimentally uncharacterized. The I-TASSER server is an on-line workbench for high-resolution modeling of protein structure and function. Given a protein sequence, a typical output from the I-TASSER server includes secondary structure prediction, predicted solvent accessibility of each residue, homologous template proteins detected by threading and structure alignments, up to five full-length tertiary structural models, and structure-based functional annotations for enzyme classification, Gene Ontology terms and protein-ligand binding sites. All the predictions are tagged with a confidence score which tells how accurate the predictions are without knowing the experimental data. To facilitate the special requests of end users, the server provides channels to accept user-specified inter-residue distance and contact maps to interactively change the I-TASSER modeling; it also allows users to specify any proteins as template, or to exclude any template proteins during the structure assembly simulations. The structural information could be collected by the users based on experimental evidences or biological insights with the purpose of improving the quality of I-TASSER predictions. The server was evaluated as the best programs for protein structure and function predictions in the recent community-wide CASP experiments. There are currently >20,000 registered scientists from over 100 countries who are using the on-line I-TASSER server.     

A novel neural response algorithm for protein function prediction

Background

Large amounts of data are being generated by high-throughput genome sequencing methods. But the rate of the experimental functional characterization falls far behind. To fill the gap between the number of sequences and their annotations, fast and accurate automated annotation methods are required. Many methods, such as GOblet, GOFigure, and Gotcha, are designed based on the BLAST search. Unfortunately, the sequence coverage of these methods is low as they cannot detect the remote homologues. Adding to this, the lack of annotation specificity advocates the need to improve automated protein function prediction.

Results

We designed a novel automated protein functional assignment method based on the neural response algorithm, which simulates the neuronal behavior of the visual cortex in the human brain. Firstly, we predict the most similar target protein for a given query protein and thereby assign its GO term to the query sequence. When assessed on test set, our method ranked the actual leaf GO term among the top 5 probable GO terms with accuracy of 86.93%.

Conclusions

The proposed algorithm is the first instance of neural response algorithm being used in the biological domain. The use of HMM profiles along with the secondary structure information to define the neural response gives our method an edge over other available methods on annotation accuracy. Results of the 5-fold cross validation and the comparison with PFP and FFPred servers indicate the prominent performance by our method. The program, the dataset, and help files are available at http://www.jjwanglab.org/NRProF/.

     

Future directions in protein function prediction

New directions in computational methods for the prediction of protein function are discussed. THEMATICS, a method for the location and characterization of the active sites of enzymes, is featured. THEMATICS, for Theoretical Microscopic Titration Curves, is based on well-established finite-difference Poisson-Boltzmann methods for computing the electric field function of a protein. THEMATICS requires only the structure of the subject protein and thus may be applied to proteins that bear no similarity in structure or sequence to any previously characterized protein. The unique features of catalytic sites in proteins are discussed. Discussion of the chemical basis for the predictive powers of THEMATICS is featured in this paper. Some results are given for three illustrative examples: HIV-1 protease, human apurinic/apyrimidinic endonuclease, and human adenosine kinase.