MOTIVATION: Protein thermostability is of great interest, both in theory and in practice. RESULTS: This study compared orthologous proteins with different cellular thermostability. A large number of physicochemical properties of protein were calculated and used to develop a series of machine learning models for predicting cellular thermostability differences between orthologous proteins. Most of the important features in these models are also highly correlated to relative cellular thermostability. A comparison between the present study with previous comparison of orthologous proteins from thermophilic and mesophilic organisms found that most highly correlated features are consistent in these studies, suggesting they may be important to protein thermostability. AVAILABILITY AND IMPLEMENTATION: Data freely available for download at https://github.com/fangj3/cellular-protein-thermostability-dataset.

A number of machine learning (ML)-based algorithms have been proposed for predicting mutation-induced stability changes in proteins. In this critical review, we used hypothetical reverse mutations to evaluate the performance of five representative algorithms and found all of them suffer from the problem of overfitting. This approach is based on the fact that if a wild-type protein is more stable than a mutant protein, then the same mutant is less stable than the wild-type protein. We analyzed the underlying issues and suggest that the main causes of the overfitting problem include that the numbers of training cases were too small, and the features used in the models were not sufficiently informative for the task. We make recommendations on how to avoid overfitting in this important research area and improve the reliability and robustness of ML-based algorithms in general.

Motivation: Complex diseases such as cancers often involve multiple types of genomic and/or epigenomic abnormalities. Rapid accumulation of multiple types of omics data demands methods for integrating the multidimensional data in order to elucidate complex relationships among different types of genomic and epigenomic abnormalities. Results: In the present study, we propose a tightly integrated approach based on tensor decomposition. Multiple types of data, including mRNA, methylation, copy number variations and somatic mutations, are merged into a high-order tensor which is used to develop predictive models for overall survival. The weight tensors of the models are constrained using CANDECOMP/PARAFAC (CP) tensor decomposition and learned using support tensor machine regression (STR) and ridge tensor regression (RTR). The results demonstrate that the tensor decomposition based approaches can achieve better performance than the models based individual data type and the concatenation approach. Supplementary information: Supplementary data are available at Bioinformatics online.

Spectral clustering has become a popular subspace clustering algorithm in machine learning and data mining, which aims at finding a low-dimensional representation by utilizing the spectrum of a Laplacian matrix. It is a key to construct a discriminative and reliable affinity matrix for spectral clustering to achieve impressive clustering quality. As the real word data increase with higher dimension of features and larger number of data samples, it is a challenge to construct a good affinity matrix. Recently, sparse representation based spectral clustering (SRSC) has proven its efficiency for clustering and lead to promising clustering results in high-dimensional data. SRSC constructs affinity matrix by using sparse representation coefficient vectors. However, it is very time consuming. Additionally, the dimension of the sparse coefficient vector is equal to the number of samples, which may make the affinity matrix not discriminative enough. Therefore, it is inefficient to apply SRSC in clustering large scale datasets. To remedy these issues, we propose a new spectral clustering algorithm which constructs affinity matrix via Nonnegative Matrix Factorization (NMF) coefficient vectors. We call our algorithm as NMF based spectral clustering (NMFSC). The dimension of NMF coefficient vector is independent on the number of the samples and significantly smaller than that of sparse coefficient vector. Therefore, the affinity matrix can be constructed via NMF coefficient vector with much lower computational cost. The experimental results on several public gene expression profiling (GEP) datasets demonstrate the advantage of NMF coefficient over sparse representation coefficient and suggest that NMFSC is promising in clustering high-dimensional data.

MOTIVATION: Previous studies have demonstrated that machine learning based molecular cancer classification using gene expression profiling (GEP) data is promising for the clinic diagnosis and treatment of cancer. Novel classification methods with high efficiency and prediction accuracy are still needed to deal with high dimensionality and small sample size of typical GEP data. Recently the sparse representation (SR) method has been successfully applied to the cancer classification. Nevertheless, its efficiency needs to be improved when analyzing large-scale GEP data. RESULTS: In this paper we present the meta-sample-based regularized robust coding classification (MRRCC), a novel effective cancer classification technique that combines the idea of meta-sample-based cluster method with regularized robust coding (RRC) method. It assumes that the coding residual and the coding coefficient are respectively independent and identically distributed. Similar to meta-sample-based SR classification (MSRC), MRRCC extracts a set of meta-samples from the training samples, and then encodes a testing sample as the sparse linear combination of these meta-samples. The representation fidelity is measured by the l2-norm or l1-norm of the coding residual. CONCLUSIONS: Extensive experiments on publicly available GEP datasets demonstrate that the proposed method is more efficient while its prediction accuracy is equivalent to existing MSRC-based methods and better than other state-of-the-art dimension reduction based methods.

SUMMARY: Nuclear receptors (NRs) are a class of transcription factors playing important roles in various biological processes. An NR often impacts numerous genes and different NRs share overlapped target networks. To fulfil the need for a database incorporating binding sites of different NRs at various conditions for easy comparison and visualization to improve our understanding of NR binding mechanisms, we have developed NURBS, a database for experimental and predicted nuclear receptor binding sites of mouse (NURBS). NURBS currently contains binding sites across the whole-mouse genome of 8 NRs identified in 40 chromatin immunoprecipitation with massively parallel DNA sequencing experiments. All datasets are processed using a widely used procedure and same statistical criteria to ensure the binding sites derived from different datasets are comparable. NURBS also provides predicted binding sites using NR-HMM, a Hidden Markov Model (HMM) model. AVAILABILITY: The GBrowse-based user interface of NURBS is freely accessible at http://shark.abl.ku.edu/nurbs/. NR-HMM and all results can be downloaded for free at the website. CONTACT: [email protected]

BACKGROUND: Protein aggregation is a significant problem in the biopharmaceutical industry (protein drug stability) and is associated medically with over 40 human diseases. Although a number of computational models have been developed for predicting aggregation propensity and identifying aggregation-prone regions in proteins, little systematic research has been done to determine physicochemical properties relevant to aggregation and their relative importance to this important process. Such studies may result in not only accurately predicting peptide aggregation propensities and identifying aggregation prone regions in proteins, but also aid in discovering additional underlying mechanisms governing this process. RESULTS: We use two feature selection algorithms to identify 16 features, out of a total of 560 physicochemical properties, presumably important to protein aggregation. Two predictors (ProA-SVM and ProA-RF) using selected features are built for predicting peptide aggregation propensity and identifying aggregation prone regions in proteins. Both methods are compared favourably to other state-of-the-art algorithms in cross validation. The identified important properties are fairly consistent with previous studies and bring some new insights into protein and peptide aggregation. One interesting new finding is that aggregation prone peptide sequences have similar properties to signal peptide and signal anchor sequences. CONCLUSIONS: Both predictors are implemented in a freely available web application (http://www.abl.ku.edu/ProA/). We suggest that the quaternary structure of protein aggregates, especially soluble oligomers, may allow the formation of new molecular recognition signals that guide aggregate targeting to specific cellular sites.

MOTIVATION: Complex diseases induce perturbations to interaction and regulation networks in living systems, resulting in dynamic equilibrium states that differ for different diseases and also normal states. Thus identifying gene expression patterns corresponding to different equilibrium states is of great benefit to the diagnosis and treatment of complex diseases. However, it remains a major challenge to deal with the high dimensionality and small size of available complex disease gene expression datasets currently used for discovering gene expression patterns. RESULTS: Here we present a phase-only correlation (POC) based classification method for recognizing the type of complex diseases. First, a virtual sample template is constructed for each subclass by averaging all samples of each subclass in a training dataset. Then the label of a test sample is determined by measuring the similarity between the test sample and each template. This novel method can detect the similarity of overall patterns emerged from the differentially expressed genes or proteins while ignoring small mismatches. CONCLUSIONS: The experimental results obtained on seven publicly available complex disease datasets including microarray and protein array data demonstrate that the proposed POC-based disease classification method is effective and robust for diagnosing complex diseases with regard to the number of initially selected features, and its recognition accuracy is better than or comparable to other state-of-the-art machine learning methods. In addition, the proposed method does not require parameter tuning and data scaling, which can effectively reduce the occurrence of over-fitting and bias.

BACKGROUND: Previous studies on tumor classification based on gene expression profiles suggest that gene selection plays a key role in improving the classification performance. Moreover, finding important tumor-related genes with the highest accuracy is a very important task because these genes might serve as tumor biomarkers, which is of great benefit to not only tumor molecular diagnosis but also drug development. RESULTS: This paper proposes a novel gene selection method with rich biomedical meaning based on Heuristic Breadth-first Search Algorithm (HBSA) to find as many optimal gene subsets as possible. Due to the curse of dimensionality, this type of method could suffer from over-fitting and selection bias problems. To address these potential problems, a HBSA-based ensemble classifier is constructed using majority voting strategy from individual classifiers constructed by the selected gene subsets, and a novel HBSA-based gene ranking method is designed to find important tumor-related genes by measuring the significance of genes using their occurrence frequencies in the selected gene subsets. The experimental results on nine tumor datasets including three pairs of cross-platform datasets indicate that the proposed method can not only obtain better generalization performance but also find many important tumor-related genes. CONCLUSIONS: It is found that the frequencies of the selected genes follow a power-law distribution, indicating that only a few top-ranked genes can be used as potential diagnosis biomarkers. Moreover, the top-ranked genes leading to very high prediction accuracy are closely related to specific tumor subtype and even hub genes. Compared with other related methods, the proposed method can achieve higher prediction accuracy with fewer genes. Moreover, they are further justified by analyzing the top-ranked genes in the context of individual gene function, biological pathway, and protein-protein interaction network.

Identification of phosphorylation sites by computational methods is becoming increasingly important as it may reduce labor-intensive and costly experiments and improve our understanding in the common properties and underlying mechanisms of protein phosphorylation. A multitask learning framework for learning phosphorylation sites from four kinase families simultaneously, instead of studying each one of them separately, is presented in the study. Our strategy successfully selects 18 common features shared by four kinase families of phosphorylation sites. The reliability of selected features is demonstrated by the consistent performance in two multi-task learning methods. Those features can be used to build efficient multitask classifiers with good performance, suggesting they are important to protein phosphorylation.

MOTIVATION: Protein residue-residue contact prediction can be useful in predicting protein 3D structures. Current algorithms for such a purpose leave room for improvement. RESULTS: We develop ProC_S3, a set of Random Forest algorithm-based models, for predicting residue-residue contact maps. The models are constructed based on a collection of 1490 non-redundant, high-resolution protein structures using >1280 sequence-based features. A new amino acid residue contact propensity matrix and a new set of seven amino acid groups based on contact preference are developed and used in ProC_S3. ProC_S3 delivers a 3-fold cross-validated accuracy of 26.9% with coverage of 4.7% for top L/5 predictions (L is the number of residues in a protein) of long-range contacts (sequence separation ≥24). Further benchmark tests deliver an accuracy of 29.7% and coverage of 5.6% for an independent set of 329 proteins. In the recently completed Ninth Community Wide Experiment on the Critical Assessment of Techniques for Protein Structure Prediction (CASP9), ProC_S3 is ranked as No. 1, No. 3, and No. 2 accuracies in the top L/5, L/10 and best 5 predictions of long-range contacts, respectively, among 18 automatic prediction servers. AVAILABILITY: http://www.abl.ku.edu/proc/proc_s3.html. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: The ability to design thermostable proteins is theoretically important and practically useful. Robust and accurate algorithms, however, remain elusive. One critical problem is the lack of reliable methods to estimate the relative thermostability of possible mutants. RESULTS: We report a novel scoring function for discriminating hyperthermophilic and mesophilic proteins with application to predicting the relative thermostability of protein mutants. The scoring function was developed based on an elaborate analysis of a set of features calculated or predicted from 540 pairs of hyperthermophilic and mesophilic protein ortholog sequences. It was constructed by a linear combination of ten important features identified by a feature ranking procedure based on the random forest classification algorithm. The weights of these features in the scoring function were fitted by a hill-climbing algorithm. This scoring function has shown an excellent ability to discriminate hyperthermophilic from mesophilic sequences. The prediction accuracies reached 98.9% and 97.3% in discriminating orthologous pairs in training and the holdout testing datasets, respectively. Moreover, the scoring function can distinguish non-homologous sequences with an accuracy of 88.4%. Additional blind tests using two datasets of experimentally investigated mutations demonstrated that the scoring function can be used to predict the relative thermostability of proteins and their mutants at very high accuracies (92.9% and 94.4%). We also developed an amino acid substitution preference matrix between mesophilic and hyperthermophilic proteins, which may be useful in designing more thermostable proteins. CONCLUSIONS: We have presented a novel scoring function which can distinguish not only HP/MP ortholog pairs, but also non-homologous pairs at high accuracies. Most importantly, it can be used to accurately predict the relative stability of proteins and their mutants, as demonstrated in two blind tests. In addition, the residue substitution preference matrix assembled in this study may reflect the thermal adaptation induced substitution biases. A web server implementing the scoring function and the dataset used in this study are freely available at http://www.abl.ku.edu/thermorank/.

BACKGROUND: The toxic effects of many simple organic compounds stem from their biotransformation to chemically reactive metabolites which bind covalently to cellular proteins. To understand the mechanisms of cytotoxic responses it may be important to know which proteins become adducted and whether some may be common targets of multiple toxins. The literature of this field is widely scattered but expanding rapidly, suggesting the need for a comprehensive, searchable database of reactive metabolite target proteins. DESCRIPTION: The Reactive Metabolite Target Protein Database (TPDB) is a comprehensive, curated, searchable, documented compilation of publicly available information on the protein targets of reactive metabolites of 18 well-studied chemicals and drugs of known toxicity. TPDB software enables i) string searches for author names and proteins names/synonyms, ii) more complex searches by selecting chemical compound, animal species, target tissue and protein names/synonyms from pull-down menus, and iii) commonality searches over multiple chemicals. Tabulated search results provide information, references and links to other databases. CONCLUSION: The TPDB is a unique on-line compilation of information on the covalent modification of cellular proteins by reactive metabolites of chemicals and drugs. Its comprehensiveness and searchability should facilitate the elucidation of mechanisms of reactive metabolite toxicity. The database is freely available at http://tpdb.medchem.ku.edu/tpdb.html.

SUMMARY: Tracker is a web-based email alert system for monitoring protein database searches using HMMER and Blast-P, nucleotide searches using Blast-N and literature searches of the PubMed database. Users submit searches via a web-based interface. Searches are saved and run against updated databases to alert users about new information. If there are new results from the saved searches, users will be notified by email and will then be able to access results and link to additional information on the NCBI website. Tracker supports Boolean AND/OR operations on HMMER and BLASTP result sets to allow users to broaden or narrow protein searches. AVAILABILITY: The server is located at http://jay.bioinformatics.ku.edu/tracker/index.html. A distribution package including detailed installation procedure is freely available from http://jay.bioinformatics.ku.edu/download/tracker/.

BACKGROUND: The development of high-throughput technologies such as yeast two-hybrid systems and mass spectrometry technologies has made it possible to generate large protein-protein interaction (PPI) datasets. Mining these datasets for underlying biological knowledge has, however, remained a challenge. RESULTS: A total of 3108 sequence signatures were found, each of which was shared by a set of guest proteins interacting with one of 944 host proteins in Saccharomyces cerevisiae genome. Approximately 94% of these sequence signatures matched entries in InterPro member databases. We identified 84 distinct sequence signatures from the remaining 172 unknown signatures. The signature sharing information was then applied in predicting sub-cellular localization of yeast proteins and the novel signatures were used in identifying possible interacting sites. CONCLUSION: We reported a method of PPI data mining that facilitated the discovery of novel sequence signatures using a large PPI dataset from S. cerevisiae genome as input. The fact that 94% of discovered signatures were known validated the ability of the approach to identify large numbers of signatures from PPI data. The significance of these discovered signatures was demonstrated by their application in predicting sub-cellular localizations and identifying potential interaction binding sites of yeast proteins.