Cross-manifold clustering is an extreme challenge learning problem. Since the low-density hypothesis is not satisfied in cross-manifold problems, many traditional clustering methods failed to discover the cross-manifold structures. In this article, we propose multiple flat projections clustering (MFPC) for cross-manifold clustering. In our MFPC, the given samples are projected into multiple localized flats to discover the global structures of implicit manifolds. Thus, the intersected clusters are distinguished in various projection flats. In MFPC, a series of nonconvex matrix optimization problems is solved by a proposed recursive algorithm. Furthermore, a nonlinear version of MFPC is extended via kernel tricks to deal with a more complex cross-manifold learning situation. The synthetic tests show that our MFPC works on the cross-manifold structures well. Moreover, experimental results on the benchmark datasets and object tracking videos show excellent performance of our MFPC compared with some state-of-the-art manifold clustering methods.

Based on collaborative representation, a novel supervised dimensionality reduction method called collaborative discriminative manifold embedding (CDME) is proposed for hyperspectral imagery. In the proposed CDME, we construct both an intraclass manifold graph and an interclass manifold graph based on two structured dictionaries. In the intraclass manifold graph, the neighborhood points are selected from the dictionary with the same class. Interclass manifold graph calculates the edge weight using all points that are sampled from the dictionary with the different classes. The goal of CDME is to learn a low-dimensional feature space by preserving the intraclass reconstructive structure and the interclass geometric structure simultaneously. Finally, the 1-NN classifier is employed to verify the performance of the CDME. Experimental results demonstrate that CDME outperforms other state-of-the-art dimensionality reduction methods.

MOTIVATION: With the booming of interactome studies, a lot of interactions can be measured in a high throughput way and large scale datasets are available. It is becoming apparent that many different types of interactions can be potential drug targets. Compared with inhibition of a single protein, inhibition of protein-protein interaction (PPI) is promising to improve the specificity with fewer adverse side-effects. Also it greatly broadens the drug target search space, which makes the drug target discovery difficult. Computational methods are highly desired to efficiently provide candidates for further experiments and hold the promise to greatly accelerate the discovery of novel drug targets. RESULTS: Here, we propose a machine learning method to predict PPI targets in a genomic-wide scale. Specifically, we develop a computational method, named as PrePPItar, to Predict PPIs as drug targets by uncovering the potential associations between drugs and PPIs. First, we survey the databases and manually construct a gold-standard positive dataset for drug and PPI interactions. This effort leads to a dataset with 227 associations among 63 PPIs and 113 FDA-approved drugs and allows us to build models to learn the association rules from the data. Second, we characterize drugs by profiling in chemical structure, drug ATC-code annotation, and side-effect space and represent PPI similarity by a symmetrical S-kernel based on protein amino acid sequence. Then the drugs and PPIs are correlated by Kronecker product kernel. Finally, a support vector machine (SVM), is trained to predict novel associations between drugs and PPIs. We validate our PrePPItar method on the well-established gold-standard dataset by cross-validation. We find that all chemical structure, drug ATC-code, and side-effect information are predictive for PPI target. Moreover, we can increase the PPI target prediction coverage by integrating multiple data sources. Follow-up database search and pathway analysis indicate that our new predictions are worthy of future experimental validation. CONCLUSION: In conclusion, PrePPItar can serve as a useful tool for PPI target discovery and provides a general heterogeneous data integrative framework. AVAILABILITY AND IMPLEMENTATION: PrePPItar is available at http://doc.aporc.org/wiki/PrePPItar. CONTACT: [email protected] or [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

The twin support vector machine (TWSVM) is one of the powerful classification methods. In this brief, a TWSVM-type clustering method, called twin support vector clustering (TWSVC), is proposed. Our TWSVC includes both linear and nonlinear versions. It determines k cluster center planes by solving a series of quadratic programming problems. To make TWSVC more efficient and stable, an initialization algorithm based on the nearest neighbor graph is also suggested. The experimental results on several benchmark data sets have shown a comparable performance of our TWSVC.

MOTIVATION: Discovering drug's Anatomical Therapeutic Chemical (ATC) classification rules at molecular level is of vital importance to understand a vast majority of drugs action. However, few studies attempt to annotate drug's potential ATC-codes by computational approaches. RESULTS: Here, we introduce drug-target network to computationally predict drug's ATC-codes and propose a novel method named NetPredATC. Starting from the assumption that drugs with similar chemical structures or target proteins share common ATC-codes, our method, NetPredATC, aims to assign drug's potential ATC-codes by integrating chemical structures and target proteins. Specifically, we first construct a gold-standard positive dataset from drugs' ATC-code annotation databases. Then we characterize ATC-code and drug by their similarity profiles and define kernel function to correlate them. Finally, we use a kernel method, support vector machine, to automatically predict drug's ATC-codes. Our method was validated on four drug datasets with various target proteins, including enzymes, ion channels, G-protein couple receptors and nuclear receptors. We found that both drug's chemical structure and target protein are predictive, and target protein information has better accuracy. Further integrating these two data sources revealed more experimentally validated ATC-codes for drugs. We extensively compared our NetPredATC with SuperPred, which is a chemical similarity-only based method. Experimental results showed that our NetPredATC outperforms SuperPred not only in predictive coverage but also in accuracy. In addition, database search and functional annotation analysis support that our novel predictions are worthy of future experimental validation. CONCLUSION: In conclusion, our new method, NetPredATC, can predict drug's ATC-codes more accurately by incorporating drug-target network and integrating data, which will promote drug mechanism understanding and drug repositioning and discovery. AVAILABILITY: NetPredATC is available at http://doc.aporc.org/wiki/NetPredATC. CONTACT: [email protected] or [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

In this letter, we propose an improved version of generalized eigenvalue proximal support vector machine (GEPSVM), called IGEPSVM for short. The main improvements are 1) the generalized eigenvalue decomposition is replaced by the standard eigenvalue decomposition, resulting in simpler optimization problems without the possible singularity. 2) An extra meaningful parameter is introduced, resulting in the stronger classification generalization ability. Experimental results on both the artificial datasets and several benchmark datasets show that our IGEPSVM is superior to GEPSVM in both computation time and classification accuracy.

Local learning has been successfully applied to transductive classification problems. In this paper, it is generalized to multi-class classification of transductive learning problems owing to its good classification ability. Meanwhile, there is essentially no ordinal meaning in class label of multi-class classification, and it belongs to discrete nominal variable. However, common binary series class label representation has the equal distance from one class to another, and it does not reflect the sparse and density relationship among classes distribution, so a learning and adjustable nominal class label representation method is presented. Experimental results on a set of benchmark multi-class datasets show the superiority of our algorithm.

MOTIVATION: Predicting protein interactions involving peptide recognition domains is essential for understanding the many important biological processes they mediate. It is important to consider the binding strength of these interactions to help us construct more biologically relevant protein interaction networks that consider cellular context and competition between potential binders. RESULTS: We developed a novel regression framework that considers both positive (quantitative) and negative (qualitative) interaction data available for mouse PDZ domains to quantitatively predict interactions between PDZ domains, a large peptide recognition domain family, and their peptide ligands using primary sequence information. First, we show that it is possible to learn from existing quantitative and negative interaction data to infer the relative binding strength of interactions involving previously unseen PDZ domains and/or peptides given their primary sequence. Performance was measured using cross-validated hold out testing and testing with previously unseen PDZ domain-peptide interactions. Second, we find that incorporating negative data improves quantitative interaction prediction. Third, we show that sequence similarity is an important prediction performance determinant, which suggests that experimentally collecting additional quantitative interaction data for underrepresented PDZ domain subfamilies will improve prediction. AVAILABILITY AND IMPLEMENTATION: The Matlab code for our SemiSVR predictor and all data used here are available at http://baderlab.org/Data/PDZAffinity.

BACKGROUND: With the development of genome-sequencing technologies, protein sequences are readily obtained by translating the measured mRNAs. Therefore predicting protein-protein interactions from the sequences is of great demand. The reason lies in the fact that identifying protein-protein interactions is becoming a bottleneck for eventually understanding the functions of proteins, especially for those organisms barely characterized. Although a few methods have been proposed, the converse problem, if the features used extract sufficient and unbiased information from protein sequences, is almost untouched. RESULTS: In this study, we interrogate this problem theoretically by an optimization scheme. Motivated by the theoretical investigation, we find novel encoding methods for both protein sequences and protein pairs. Our new methods exploit sufficiently the information of protein sequences and reduce artificial bias and computational cost. Thus, it significantly outperforms the available methods regarding sensitivity, specificity, precision, and recall with cross-validation evaluation and reaches ~80% and ~90% accuracy in Escherichia coli and Saccharomyces cerevisiae respectively. Our findings here hold important implication for other sequence-based prediction tasks because representation of biological sequence is always the first step in computational biology. CONCLUSIONS: By considering the converse problem, we propose new representation methods for both protein sequences and protein pairs. The results show that our method significantly improves the accuracy of protein-protein interaction predictions.

For classification problems, the generalized eigenvalue proximal support vector machine (GEPSVM) and twin support vector machine (TWSVM) are regarded as milestones in the development of the powerful SVMs, as they use the nonparallel hyperplane classifiers. In this brief, we propose an improved version, named twin bounded support vector machines (TBSVM), based on TWSVM. The significant advantage of our TBSVM over TWSVM is that the structural risk minimization principle is implemented by introducing the regularization term. This embodies the marrow of statistical learning theory, so this modification can improve the performance of classification. In addition, the successive overrelaxation technique is used to solve the optimization problems to speed up the training procedure. Experimental results show the effectiveness of our method in both computation time and classification accuracy, and therefore confirm the above conclusion further.