Motivation: The rapid development of sequencing technology has led to an explosive accumulation of genomic data. Clustering is often the first step to be performed in sequence analysis. However, existing methods scale poorly with respect to the unprecedented growth of input data size. As high-performance computing systems are becoming widely accessible, it is highly desired that a clustering method can easily scale to handle large-scale sequence datasets by leveraging the power of parallel computing. Results: In this paper, we introduce SLAD (Separation via Landmark-based Active Divisive clustering), a generic computational framework that can be used to parallelize various de novo operational taxonomic unit (OTU) picking methods and comes with theoretical guarantees on both accuracy and efficiency. The proposed framework was implemented on Apache Spark, which allows for easy and efficient utilization of parallel computing resources. Experiments performed on various datasets demonstrated that SLAD can significantly speed up a number of popular de novo OTU picking methods and meanwhile maintains the same level of accuracy. In particular, the experiment on the Earth Microbiome Project dataset (∼2.2B reads, 437 GB) demonstrated the excellent scalability of the proposed method. Availability and implementation: Open-source software for the proposed method is freely available at https://www.acsu.buffalo.edu/~yijunsun/lab/SLAD.html. Supplementary information: Supplementary data are available at Bioinformatics online.

We propose an alternative probabilistic dimensionality reduction framework that can naturally integrate the generative model and the locality information of data. Based on this framework, we present a new model, which is able to learn a set of embedding points in a low-dimensional space by retaining the inherent structure from high-dimensional data. The objective function of this new model can be equivalently interpreted as two coupled learning problems, i.e., structure learning and the learning of projection matrix. Inspired by this interesting interpretation, we propose another model, which finds a set of embedding points that can directly form an explicit graph structure. We proved that the model by learning explicit graphs generalizes the reversed graph embedding method, but leads to a natural interpretation from Bayesian perspective. This can greatly facilitate data visualization and scientific discovery in downstream analysis. Extensive experiments are performed that demonstrate that the proposed framework is able to retain the inherent structure of datasets and achieve competitive quantitative results in terms of various performance evaluation criteria.

Many scientific datasets are of high dimension, and the analysis usually requires retaining the most important structures of data. Principal curve is a widely used approach for this purpose. However, many existing methods work only for data with structures that are mathematically formulated by curves, which is quite restrictive for real applications. A few methods can overcome the above problem, but they either require complicated human-made rules for a specific task with lack of adaption flexibility to different tasks, or cannot obtain explicit structures of data. To address these issues, we develop a novel principal graph and structure learning framework that captures the local information of the underlying graph structure based on reversed graph embedding. As showcases, models that can learn a spanning tree or a weighted undirected `1 graph are proposed, and a new learning algorithm is developed that learns a set of principal points and a graph structure from data, simultaneously. The new algorithm is simple with guaranteed convergence. We then extend the proposed framework to deal with large-scale data. Experimental results on various synthetic and six real world datasets show that the proposed method compares favorably with baselines and can uncover the underlying structure correctly.

The rapid development of sequencing technology has led to an explosive accumulation of genomic sequence data. Clustering is often the first step to perform in sequence analysis, and hierarchical clustering is one of the most commonly used approaches for this purpose. However, it is currently computationally expensive to perform hierarchical clustering of extremely large sequence datasets due to its quadratic time and space complexities. In this paper we developed a new algorithm called ESPRIT-Forest for parallel hierarchical clustering of sequences. The algorithm achieves subquadratic time and space complexity and maintains a high clustering accuracy comparable to the standard method. The basic idea is to organize sequences into a pseudo-metric based partitioning tree for sub-linear time searching of nearest neighbors, and then use a new multiple-pair merging criterion to construct clusters in parallel using multiple threads. The new algorithm was tested on the human microbiome project (HMP) dataset, currently one of the largest published microbial 16S rRNA sequence dataset. Our experiment demonstrated that with the power of parallel computing it is now compu- tationally feasible to perform hierarchical clustering analysis of tens of millions of sequences. The software is available at http://www.acsu.buffalo.edu/∼yijunsun/lab/ESPRIT-Forest.html.

The rapid development of sequencing technology has led to an explosive accumulation of genomics data. Clustering is often the first step to perform in sequence analysis, and hierarchical clustering is one of the most commonly used approaches for this purpose. However, the standard hierarchical clustering method scales poorly due to its quadratic time and space complexities stemming mainly from the need of computing and storing a pairwise distance matrix. It is thus necessary to minimize the number of pairwise distances computed without degrading clustering performance. On the other hand, as high-performance computing systems are becoming widely accessible, it is highly desirable that a clustering method can be easily adapted to parallel computing environments for further speedup, which is not a trivial task for hierarchical clustering. We proposed a new hierarchical clustering method that achieves good clustering performance and high scalability on large sequence datasets. It consists of two stages. In the first stage, a new landmark-based active hierarchical divisive clustering method was proposed that partitions a large-scale sequence dataset into groups, and in the second stage, a fast hierarchical agglomerative clustering method is applied to each group. By assembling hierarchies from both stages, the hierarchy of the data can be easily recovered. Theoretical results showed that our method can recover the true hierarchy with a high probability under some mild conditions and has a linearithmic time complexity with respect to the number of input sequences. The proposed method also facilitates an efficient parallel implementation. Empirical results on various datasets showed that our method achieved clustering accuracy comparable to ESPRIT-Tree and ran faster than greedy heuristic methods.

We present a new dimensionality reduction setting for a large family of real-world problems. Unlike traditional methods, the new setting aims to explicitly represent and learn an intrinsic structure from data in a high-dimensional space, which can greatly facilitate data visualization and scientific discovery in downstream analysis. We propose a new dimensionality-reduction framework that involves the learning of a mapping function that projects data points in the original high-dimensional space to latent points in a low-dimensional space that are then used directly to construct a graph. Local geometric information of the projected data is naturally captured by the constructed graph. As a showcase, we develop a new method to obtain a discriminative and compact feature representation for clustering problems. In contrast to assumptions used in traditional clustering methods, we assume that centers of clusters should be close to each other if they are connected in a learned graph, and other cluster centers should be distant. Extensive experiments are performed that demonstrate that the proposed method is able to obtain discriminative feature representations yielding superior clustering performance, and correctly recover the intrinsic structures of various real-world datasets including curves, hierarchies and a cancer progression path.

Many scientific datasets are of high dimension, and the analysis usually requires visual manipulation by retaining the most important structures of data. Principal curve is a widely used approach for this purpose. However, many existing methods work only for data with structures that are not self-intersected, which is quite restrictive for real applications. To address this issue, we develop a new model, which captures the local information of the underlying graph structure based on reversed graph embedding. A generalization bound is derived that show that the model is consistent if the number of data points is sufficiently large. As a special case, a principal tree model is proposed and a new algorithm is developed that learns a tree structure automatically from data. The new algorithm is simple and parameter-free with guaranteed convergence. Experimental results on synthetic and breast cancer datasets show that the proposed method compares favorably with baselines and can discover a breast cancer progression path with multiple branches.

Multiple kernel learning (MKL) and classifier ensemble are two mainstream methods for solving learning problems in which some sets of features/views are more informative than others, or the features/views within a given set are inconsistent. In this paper, we first present a novel probabilistic interpretation of MKL such that maximum entropy discrimination with a noninformative prior over multiple views is equivalent to the formulation of MKL. Instead of using the noninformative prior, we introduce a novel data-dependent prior based on an ensemble of kernel predictors, which enhances the prediction performance of MKL by leveraging the merits of the classifier ensemble. With the proposed probabilistic framework of MKL, we propose a hierarchical Bayesian model to learn the proposed data-dependent prior and classification model simultaneously. The resultant problem is convex and other information (e.g., instances with either missing views or missing labels) can be seamlessly incorporated into the data-dependent priors. Furthermore, a variety of existing MKL models can be recovered under the proposed MKL framework and can be readily extended to incorporate these priors. Extensive experiments demonstrate the benefits of our proposed framework in supervised and semisupervised settings, as well as in tasks with partial correspondence among multiple views.

Feature selection with specific multivariate performance measures is the key to the success of many applications such as image retrieval and text classification. The existing feature selection methods are usually designed for classification error. In this paper, we propose a generalized sparse regularizer. Based on the proposed regularizer, we present a unified feature selection framework for general loss functions. In particular, we study the novel feature selection paradigm by optimizing multivariate performance measures. The resultant formulation is a challenging problem for high-dimensional data. Hence, a two-layer cutting plane algorithm is proposed to solve this problem, and the convergence is presented. In addition, we adapt the proposed method to optimize multivariate measures for multiple-instance learning problems. The analyses by comparing with the state-of-the-art feature selection methods show that the proposed method is superior to others. Extensive experiments on large-scale and high-dimensional real-world datasets show that the proposed method outperforms l₁-SVM and SVM-RFE when choosing a small subset of features, and achieves significantly improved performances over SVM(perf) in terms of F₁-score.

Automatic image annotation, which is usually formulated as a multi-label classification problem, is one of the major tools used to enhance the semantic understanding of web images. Many multimedia applications (e.g., tag-based image retrieval) can greatly benefit from image annotation. However, the insufficient performance of image annotation methods prevents these applications from being practical. On the other hand, specific measures are usually designed to evaluate how well one annotation method performs for a specific objective or application, but most image annotation methods do not consider optimization of these measures, so that they are inevitably trapped into suboptimal performance of these objective-specific measures. To address this issue, we first summarize a variety of objective-guided performance measures under a unified representation. Our analysis reveals that macro-averaging measures are very sensitive to infrequent keywords, and hamming measure is easily affected by skewed distributions. We then propose a unified multi-label learning framework, which directly optimizes a variety of objective-specific measures of multi-label learning tasks. Specifically, we first present a multilayer hierarchical structure of learning hypotheses for multi-label problems based on which a variety of loss functions with respect to objective-guided measures are defined. And then, we formulate these loss functions as relaxed surrogate functions and optimize them by structural SVMs. According to the analysis of various measures and the high time complexity of optimizing micro-averaging measures, in this paper, we focus on example-based measures that are tailor-made for image annotation tasks but are seldom explored in the literature. Experiments show consistency with the formal analysis on two widely used multi-label datasets, and demonstrate the superior performance of our proposed method over state-of-the-art baseline methods in terms of example-based measures on four image annotation datasets.

Conditional random fields (CRF) and structural support vector machines (structural SVM) are two state-of-the-art methods for structured prediction that captures the interdependencies among output variables. The success of these methods is attributed to the fact that their discriminative models are able to account for overlapping features on all input observations. These features are usually generated by applying a given set of templates on labeled data, but improper templates may lead to degraded performance. To alleviate this issue, in this paper we propose a novel multiple template learning paradigm to learn structured prediction and the importance of each template simultaneously, so that hundreds of arbitrary templates could be added into the learning model without caution. This paradigm can be formulated as a special multiple kernel learning problem with an exponential number of constraints. Then we introduce an efficient cutting-plane algorithm to solve this problem in the primal and present its convergence. We also evaluate the proposed learning paradigm on two widely studied structured prediction tasks, i.e., sequence labeling and dependency parsing. Extensive experimental results show that the proposed method outperforms CRFs and structural SVMs because of exploiting the importance of each template. Complexity analysis and empirical results also show that the proposed method is more efficient than Online multikernel learning on very sparse and high-dimensional data. We further extend this paradigm for structured prediction using generalized p-block norm regularization with p >; 1, and experiments show competitive performances when p ∈ [1,2).

In the absence of the labeled samples in a domain referred to as target domain, Domain Adaptation (DA) techniques come in handy. Generally, DA techniques assume there are available source domains that share similar predictive function with the target domain. Two core challenges of DA typically arise, variance that exists between source and target domains, and the inherent source hypothesis bias. In this paper, we first propose a Stability Transfer criterion for selecting relevant source domains with low variance. With this criterion, we introduce a TARget learning Assisted by Source Classifier Adaptation (TARASCA) method to address the two core challenges that have impeded the performances of DA techniques. To verify the robustness of TARASCA, extensive experimental studies are carried out with comparison to several state-of-the-art DA methods on the real-world Sentiment and Newsgroups datasets, where various settings for the class ratios of the source and target domains are considered.

Feature selection with specific multivariate performance measures is the key to the success of many applications, such as information retrieval and bioinformatics. The existing feature selection methods are usually designed for classification error. In this paper, we present a unified feature selection framework for general loss functions. In particular, we study the novel feature selection paradigm by optimizing multivariate performance measures. The resultant formulation is a challenging problem for high-dimensional data. Hence, a two-layer cutting plane algorithm is proposed to solve this problem, and the convergence is presented. Extensive experiments on large-scale and high-dimensional real world datasets show that the proposed method outperforms l1-SVM and SVM-RFE when choosing a small subset of features, and achieves significantly improved performances over SVMperfin terms of F1-score.