Search

Sequential recommendation aims to identify and recommend the next few items for a user that the user is most likely to purchase/review, given the user's purchase/rating trajectories. It becomes an effective tool to help users select favorite items from a variety of options. In this manuscript, we developed hybrid associations models (HAM) to generate sequential recommendations using three factors: 1) users long-term preferences, 2) sequential, high-order and low-order association patterns in the users most recent purchases/ratings, and 3) synergies among those items. HAM uses simplistic pooling to represent a set of items in the associations, and element-wise product to represent item synergies of arbitrary orders. We compared HAM models with the most recent, state-of-the-art methods on six public benchmark datasets in three different experimental settings. Our experimental results demonstrate that HAM models significantly outperform the state of the art in all the experimental settings, with an improvement as much as 46.6%. In addition, our run-time performance comparison in testing demonstrates that HAM models are much more efficient than the state-of-the-art methods, and are able to achieve significant speedup as much as 139.7 folds.

Background Cognitive assessments represent the most common clinical routine for the diagnosis of Alzheimer’s Disease (AD). Given a large number of cognitive assessment tools and time-limited office visits, it is important to determine a proper set of cognitive tests for different subjects. Most current studies create guidelines of cognitive test selection for a targeted population, but they are not customized for each individual subject. In this manuscript, we develop a machine learning paradigm enabling personalized cognitive assessments prioritization. Method We adapt a newly developed learning-to-rank approach $${\mathtt {PLTR}}$$ PLTR to implement our paradigm. This method learns the latent scoring function that pushes the most effective cognitive assessments onto the top of the prioritization list. We also extend $${\mathtt {PLTR}}$$ PLTR to better separate the most effective cognitive assessments and the less effective ones. Results Our empirical study on the ADNI data shows that the proposed paradigm outperforms the state-of-the-art baselines on identifying and prioritizing individual-specific cognitive biomarkers. We conduct experiments in cross validation and level-out validation settings. In the two settings, our paradigm significantly outperforms the best baselines with improvement as much as 22.1% and 19.7%, respectively, on prioritizing cognitive features. Conclusions The proposed paradigm achieves superior performance on prioritizing cognitive biomarkers. The cognitive biomarkers prioritized on top have great potentials to facilitate personalized diagnosis, disease subtyping, and ultimately precision medicine in AD.

Objectives This study aimed to create an objective predictive model for assessing the functional status of the cochlear nerve (CN) in individual cochlear implant (CI) users. Design Study participants included 23 children with cochlear nerve deficiency (CND), 29 children with normal-sized CNs (NSCNs), and 20 adults with various etiologies of hearing loss. Eight participants were bilateral CI users and were tested in both ears. As a result, a total of 80 ears were tested in this study. All participants used Cochlear Nucleus CIs in their test ears. For each participant, the CN refractory recovery function and input/output (I/O) function were measured using electrophysiological measures of the electrically evoked compound action potential (eCAP) at three electrode sites across the electrode array. Refractory recovery time constants were estimated using statistical modeling with an exponential decay function. Slopes of I/O functions were estimated using linear regression. The eCAP parameters used as input variables in the predictive model were absolute refractory recovery time estimated based on the refractory recovery function, eCAP threshold, slope of the eCAP I/O function, and negative-peak (i.e., N1) latency. The output variable of the predictive model was CN index, an indicator for the functional status of the CN. Predictive models were created by performing linear regression, support vector machine regression, and logistic regression with eCAP parameters from children with CND and the children with NSCNs. One-way analysis of variance with post hoc analysis with Tukey's honest significant difference criterion was used to compare study variables among study groups. Results All three machine learning algorithms created two distinct distributions of CN indices for children with CND and children with NSCNs. Variations in CN index when calculated using different machine learning techniques were observed for adult CI users. Regardless of these variations, CN indices calculated using all three techniques in adult CI users were significantly correlated with Consonant-Nucleus-Consonant word and AzBio sentence scores measured in quiet. The five oldest CI users had smaller CN indices than the five youngest CI users in this study. Conclusions The functional status of the CN for individual CI users was estimated by our newly developed analytical models. Model predictions of CN function for individual adult CI users were positively and significantly correlated with speech perception performance. The models presented in this study may be useful for understanding and/or predicting CI outcomes for individual patients.

Background Adverse drug events (ADEs) often occur as a result of drug-drug interactions (DDIs). The use of data mining for detecting effects of drug combinations on ADE has attracted growing attention and interest, however, most studies focused on analyzing pairwise DDIs. Recent efforts have been made to explore the directional relationships among high-dimensional drug combinations and have shown effectiveness on prediction of ADE risk. However, the existing approaches become inefficient from both computational and illustrative perspectives when considering more than three drugs. Methods We proposed an efficient approach to estimate the directional effects of high-order DDIs through frequent itemset mining, and further developed a novel visualization method to organize and present the high-order directional DDI effects involving more than three drugs in an interactive, concise and comprehensive manner. We demonstrated its performance by mining the directional DDIs associated with myopathy using a publicly available FAERS dataset. Results Directional effects of DDIs involving up to seven drugs were reported. Our analysis confirmed previously reported myopathy associated DDIs including interactions between fusidic acid with simvastatin and atorvastatin. Furthermore, we uncovered a number of novel DDIs leading to increased risk for myopathy, such as the co-administration of zoledronate with different types of drugs including antibiotics (ciprofloxacin, levofloxacin) and analgesics (acetaminophen, fentanyl, gabapentin, oxycodone). Finally, we visualized directional DDI findings via the proposed tool, which allows one to interactively select any drug combination as the baseline and zoom in/out to obtain both detailed and overall picture of interested drugs. Conclusions We developed a more efficient data mining strategy to identify high-order directional DDIs, and designed a scalable tool to visualize high-order DDI findings. The proposed method and tool have the potential to contribute to the drug interaction research and ultimately impact patient health care. Availability and implementation http://lishenlab.com/d3i/explorer.html

We propose an innovative machine learning paradigm enabling precision medicine for prioritizing cognitive assessments according to their relevance to Alzheimer's disease at the individual patient level. The paradigm tailors the cognitive biomarker discovery and cognitive assessment selection process to the brain morphometric characteristics of each individual patient. We implement this paradigm using a newly developed learning-to-rank method PLTR. Our empirical study on the ADNI data yields promising results to identify and prioritize individual-specific cognitive biomarkers as well as cognitive assessment tasks based on the individual's structural MRI data. The resulting top ranked cognitive biomarkers and assessment tasks have the potential to aid personalized diagnosis and disease subtyping.

Over the past decade low graduation and retention rates has plagued higher education institutions. To assist students in choosing a sequence of courses, choosing majors and successful academic pathways; many institutions provide several on-site academic advising services supported by data driven educational technologies. Accurate performance prediction can serve as the backbone for degree planning software, personalized advising systems and early warning systems that can identify students at-risk of dropping from their field of study. In this work, we present a deep learning based recommender system approach called Neural Collaborative Filtering (NCF) for predicting the grade a student will earn in a course that he/she plans to take in the next-term. Prior grade prediction methods are based on matrix factorization (MF) where students and courses are represented in a latent "knowledge" space. The deep learning inspired approach provides added flexibility in learning the latent spaces in comparison to MF approaches. The proposed approach also incorporates instructor information besides student and course information. Moreover, for proper analysis of the learned model parameters, we assume the embeddings obtained for students, courses and instructors should be non-negative. This non-negative NCF model referred by NCFnn model adds a rectified linear units (ReLU) on the embedding layer of NCF. The experimental results on datasets from George Mason University, a large, public university in the United States, demonstrate that the proposed NCF approaches significantly outperform competitive baselines across different test sets.

Polypharmacy increases the risk of drug-drug interactions (DDIs). Combining epidemiological studies with pharmacokinetic modeling, we detected and evaluated high-dimensional DDIs among 30 frequent drugs. Multidrug combinations that increased the risk of myopathy were identified in the US Food and Drug Administration Adverse Event Reporting System (FAERS) and electronic medical record (EMR) databases by a mixture drug-count response model. CYP450 inhibition was estimated among the 30 drugs in the presence of 1 to 4 inhibitors using in vitro / in vivo extrapolation. Twenty-eight three-way and 43 four-way DDIs had significant myopathy risk in both databases and predicted increases in the area under the concentration-time curve ratio (AUCR) >2-fold. The high-dimensional DDI of omeprazole, fluconazole, and clonidine was associated with a 6.41-fold (FAERS) and 18.46-fold (EMR) increased risk of myopathy local false discovery rate (

Drug-drug interactions (DDIs) and their associated adverse drug reactions (ADRs) represent a significant detriment to the public health. Existing research on DDIs is primarily focused on pairwise DDI detection and prediction. It is highly needed to develop effective computational tools for high-order DDI prediction. Here we show that deep learning can be effectively utilized to predict ADRs induced from high-order DDIs. In this manuscript, we present a deep learning model D3I for cardinality-invariant and order-invariant high-order DDI prediction. The D3I models achieve 0.740 F1 value and 0.847 AUC value on balanced high-order DDI prediction, and outperform other models on order-2 DDI prediction. These results demonstrate the strong potential of D3I and deep learning models in tackling the prediction problems of high-order DDIs and their induced ADRs. In addition, D3I is able to derive single drug representations, which conform to our current knowledge on single drugs, from their behaviors in drug combinations. D3I can also correctly predict ADRs for drug combinations in which no single drugs on their own induce ADRs, and improve ADR prediction on drug pairs by learning from all drug combinations. To the best of our knowledge, D3I is the first deep model for high-order DDI prediction.

Mining high-order drug-drug interaction (DDI) induced adverse drug effects from electronic health record (EHR) databases is an emerging area, and very few studies have explored the relationships between high-order drug combinations. We investigate a novel pharmacovigilance problem for mining directional DDI effects on myopathy using the FDA Adverse Event Reporting System (FAERS) database. Our work provides information on the risk of myopathy associated with adding new drugs on the already prescribed medication, and visualizes the identified directional DDI patterns as user-friendly graphical representation. We utilize the Apriori algorithm to extract frequent drug combinations from the FAERS database. We use odds ratio (OR) to estimate the risk of myopathy associated with directional DDI. We create a tree-structured graph to visualize the findings for easy interpretation. Our method confirmed myopathy association with previously reported HMG-CoA reductase inhibitors like rosuvastatin, fluvastatin, simvastatin and atorvastatin. New, previously unidentified but mechanistically plausible associations with myopathy were also observed, such as the DDI between pamidronate and levofloxacin. Additional top findings are gadolinium-based imaging agents, which however are often used in myopathy diagnosis. Other DDIs with no obvious mechanism are also reported, such as that of sulfamethoxazole with trimethoprim and potassium chloride. This study shows the feasibility to estimate high-order directional DDIs in a fast and accurate manner. The results of the analysis could become a useful tool in the specialists’ hands through an easy-to-understand graphic visualization.

High-order drug-drug interactions (DDIs) and associated adverse drug reactions (ADRs) are common, particularly for elderly people, and therefore represent a significant public health problem. In this paper, the problem of predicting whether a drug combination of arbitrary orders is likely to induce adverse drug reactions is considered. To solve this problem, novel kernels over drug combinations of arbitrary orders are developed within support vector machines (SVMs) for the prediction. Graph matching methods are used in the novel kernels to measure the similarities among drug combinations, in which drug co-medication patterns are leveraged to measure single drug similarities. The experimental results on a real-world dataset demonstrated that the new kernels achieve an area under the curve (AUC) value 0.912 for the prediction problem. The new methods with drug co-medication based single drug similarities can accurately predict whether a drug combination is likely to induce adverse drug reactions of interest.

The power supply system of muti-module spacecraft in domestic and foreign is investigated, and the power supply characteristics and typical power supply solutions of muti-module spacecraft is analyzed in this paper. Taking a deep space lunar probe as an example, a design scheme of combined power supply between multiple module is presented in this paper.The simulation model is established, and the different working modes of the combined power supply are simulated and verified. The comparison between the experimental data and the simulation results verifies the feasibility of the design. As the same time, this paper also provides an important theoretical basis for further improving the design capability of multi-module spacecraft power system and developing a more reliable spacecraft power supply system. It can be used to guide the design of multi-module spacecraft power system in the future.

Effective in silico compound prioritization is a critical step to identify promising drug candidates in the early stages of drug discovery. Current computational methods for compound prioritization usually focus on ranking the compounds based on one property, typically activity, with respect to a single target. However, compound selectivity is also a key property which should be deliberated simultaneously so as to minimize the likelihood of undesired side effects of future drugs. In this paper, we present a novel machine-learning based differential compound prioritization method dCPPP. This dCPPP method learns compound prioritization models that rank active compounds well, and meanwhile, preferably rank selective compounds higher via a bidirectional selectivity push strategy. The bidirectional push is enhanced by push powers that are determined by ranking difference of selective compounds over multiple bioassays. Our experiments demonstrate that the new method dCPPP achieves significant improvement on prioritizing selective compounds over baseline models.

Keyword-based online book retrieval can not fully understand the user's query intent. Query expansion is a typical solution, but the rate of recall and precision is still very low in existing methods. In response to these problems, this paper presents a semantic query expansion method based on domain ontology and local co-occurrence probability model. First, ontology reasoning and concepts related calculation are used to obtain the initial expansion terms. Furthermore, the local co-occurrence probability model is used to filter the candidate expansion terms and the filtering function is used for secondary selection. Experiment results show that this method can effectively improve retrieval efficiency.

The query of database must rely on professional query language, which is difficult to grasp for ordinary users. Natural language query interface (NLQI) can be applied to solve this problem with better human-computer interaction and intelligence. However, current natural language understanding technology is not mature. Though the query sentences’ structure is fixed, it is also difficult to accurately convert query sentences into understandable database query language for computer. Therefore a method of designing NLQI based on syntax and semantic analysis is proposed in this paper. First of all, Maximum forward and maximum reverse matching algorithms are used to analyse sentence by part of speech (POS) template, which is able to demonstrate the syntax relationship among the words. After syntax analysis, query conditions and targets are identified. Then, single conditions and targets are analyzed to relative standard forms by the relationship among entity, property, connect verb and unit. At last, these standard forms are transformed to SQL sentence. Regarding printing ERP system as experimental subject, experiment is done in this paper. The result shows that the method proposed in this paper can meet the needs of natural language query, and has a good transform effect for combination, omitting and other complex sentences.

Effective in silico compound prioritization is critical to identify promising candidates in the early stages of drug discovery. Current methods typically focus on compound ranking based on one single property, for example, activity, against a single target. However, compound selectivity is also a key property that should be deliberated simultaneously so as to reduce the likelihood of undesired side effects of future drugs. In this paper, we present a novel machine learning based differential compound prioritization method dCPPP. This dCPPP method learns compound prioritization models that rank active compounds well, and meanwhile, preferably rank selective compounds higher via a bi-directional push strategy. The bidirectional push is enhanced with push powers that are determined by ranking difference of selective compounds over multiple bioassays. Our experiments demonstrate that the dCPPP achieves an overall 19.221% improvement on prioritizing selective compounds over baseline models.

High-order Drug-Drug Interactions (DDI) are common particularly for elderly people. It is highly non-trivial to detect such interactions via in vivo/in vitro experiments. In this paper, we present SVM-based classification methods to predict whether a high-order directional drug-drug interaction (HoDDDI) instance is associated with adverse drug reactions (ADRs) and induced side effects. Specifically, we developed kernels for HoDDDI instances of arbitrary orders that are constructed from various single-drug information. The experiments over datasets extracted from electronic health records demonstrate that our classification methods can achieve the best F1 as 0.793.

Drug-Drug Interactions (DDIs) and associated Adverse Drug Reactions (ADRs) represent a significant public health problem in the United States. The research presented in this paper tackles the problems of representing, discovering, quantifying and visualizing patterns from high-order DDIs in a purely data-driven fashion. We formulate the problems based on a notion of directional DDI relations and correspondingly developed weighted hyper-graphlets for their representation. We also develop a convolutional scheme and its stochastic algorithm SD3ID2S to learn the directional DDI based drug-drug similarities. Our experimental results demonstrate that such approaches can well capture the patterns from high-order DDIs.

Drug-drug interactions (DDIs) are a common cause of adverse drug events (ADEs). The electronic medical record (EMR) database and the FDA's Adverse Event Reporting System (FAERS) database are the major data sources for mining and testing the ADE associated DDI signals. Most DDI data mining methods focus on pair-wise drug interactions, and methods to detect high-dimensional DDIs in medical databases are lacking. In this paper, we propose two novel mixture drug-count response models for detecting high dimensional drug combinations that induce myopathy. The “count” indicates the number of drugs in a combination. One model is called fixed probability mixture drug-count response model with a maximum risk threshold (FMDRM-MRT). The other model is called count-dependent probability mixture drug-count response model with a maximum risk threshold (CMDRM-MRT), in which the mixture probability is count-dependent. Compared to the previous mixture drug-count response model (MDRM) developed by our group, these two new models show a better likelihood in detecting high dimensional drug combinatory effects on myopathy. CMDRM-MRT identified and validated (54; 374; 637; 442; 131) 2-way to 6-way drug interactions, respectively, which induce myopathy in both EMR and FAERS databases. We further demonstrate FAERS data capture much higher maximum myopathy risk than EMR data do. The consistency of two mixture models’ parameters and local false discovery rate estimates are evaluated through statistical simulation studies.

In order to reuse and share thermal fatigue knowledge, the thermal fatigue knowledge ontology is exploited. The basic ontology relationship and characteristics are analyzed and the concept classification and axiom are devised. With the help of Protege3.1.1 software, the thermal fatigue knowledge ontology comes into being. Based on the crack axiom, the intelligent matching of thermal fatigue crack states of super-hard cutting tools is accomplished.

Ground control point (GCP) extraction is an essential step in automatic registration of remote sensing images. However the lack of quantitative and objective methods for analyzing the GCP quality becomes the bottleneck that prevents the broad development of automatic image registration. Although several measurements for evaluating the number, accuracy and distribution of GCPs have been proposed in recent years, some of them are redundant and the evaluation of dispersion is not efiective enough. In this paper, a method for an objective and quantitative evaluation of GCP quality is proposed. The proposed method consists of three parts: measurement calculation, cost function calculation and flnal validation. In the flrst part, two new measurements are proposed to evaluate the number, dispersion and isotropy of GCPs, and the root mean square of GCP residuals using leave- one-out method (RMSloo) is used to evaluate the accuracy. In the second part, seed cost functions are utilized to transform the measurements into a limited value range as well as to be desired on the ascending direction. Subsequently, all the seed cost functions are combined by a total cost function to provide an integrated evaluation. In the third part, the GCP scenario is validated by the accepted threshold depending on the value of the total cost function. To evaluate the performance of the proposed method, experiments using four typical emulated scenarios of GCP distribution and two sets of real GCPs in SAR images are considered. The results demonstrate that the proposed GCP evaluation method performs more efiectively than the existing methods, especially in the evaluation of dispersion quality.

CICS Web Support provided some CICS services, so that users could directly invoke and access the application in CICS through web browser. Users could use web-aware program to process web request and response. The paper used a case of CICS Web Support to process non-http request, to do application research of CICS Web Support IP direct connection to access mainframe. By determining the length of received data through the global pointer in CICS Web Support, the sample program realized the two times' invocation of Analyzer. At the same time, we finished the conversion of code page, and in user program, we received the requested data transmitted from Analyzer by applying BLI interface. In CICS side we used TSQ to record the program invocation in the non-HTTP processing. The conclusion is that it is feasible for IP direct connection to access mainframe using CICS Web Support function.

Operating system kernel-level tracers are popularly used in the post-development stage by black-box approaches. By inferring service request processing paths from kernel events, these approaches enabled system diagnosis and performance management that are application-logic aware. However, asynchronous communications and multi-threading behaviors make request path patterns dynamic on the kernel event level, this causes previous methods to focus on either software instrumentation techniques or better statistical inference models. In this paper, we propose a novel learning based approach called PInfer that infers request processing path patterns automatically with high precision. PInfer first learns dynamic event patterns of inter-thread and intra-thread service processing from the training data of sequential requests. On the testing data containing concurrent requests, PInfer infers individual request processing paths by effectively solving a graph matching problem and a generalized assignment problem based on the learned patterns. We have implemented our approach in a proprietary system performance diagnosis tool, and present performance results on 40 sets of kernel event traces. PInfer achieves on average 65% precision and 85% recall for profiling concurrent request processing paths.

Multi-view learning arouses vast amount of interest in the past decades with numerous real-world applications in web page analysis, bioinformatics, image processing and so on. Unlike the most previous works following the idea of co-training, in this paper we propose a new generative model for Multi-view Learning via Probabilistic Latent Semantic Analysis, called MVPLSA. In this model, we jointly model the co-occurrences of features and documents from different views. Specifically, in the model there are two latent variables y for the latent topic and z for the document cluster, and three visible variables d for the document, f for the feature, and v for the view label. The conditional probability p(z|d), which is independent of v, is used as the bridge to share knowledge among multiple views. Also, we have p(y|z, v) and p(f|y, v), which are dependent of v, to capture the specifical structures inside each view. Experiments are conducted on four real-world data sets to demonstrate the effectiveness and superiority of our model.

In this article, we review the recent development for in silico Structure-Activity-Relationship (SAR) models using machine-learning techniques. The review focuses on the following topics: machine-learning algorithms for computational SAR models, single-target-oriented SAR methodologies, Chemogenomics, and future trends. We try to provide the state-of-the-art SAR methods as well as the most up-to-date advancement, in order for the researchers to have a general overview at this area. Drug Dev Res 72: 138–146, 2011. © 2010 Wiley-Liss, Inc.

We present our solutions to the RecSys Challenge 2015. We propose a multi-perspective modeling scheme for click event prediction, which involves techniques from sophisticated feature engineering for both click sessions and clicked items, classification based on gradient boosting tree, semi-supervised learning that utilizes information from test data, multi-class classification for different categories of sessions and items, classifier-based feature fusion from multi-class classification and in the end classifier ensembles from multiple models. We demonstrate that our scheme is intuitive, flexible and powerful for the Challenge tasks. Our solution based on the scheme achieves a score of 49,517.2 in the Challenge.

Reducing attack surface is an effective preventive measure to strengthen security in large systems. However, it is challenging to apply this idea in an enterprise environment where systems are complex and evolving over time. In this paper, we empirically analyze and measure a real enterprise to identify unused services that expose attack surface. Interestingly, such unused services are known to exist and summarized by security best practices, yet such solutions require significant manual effort.We propose an automated approach to accurately detect the idling (most likely unused) services that are in either blocked or bookkeeping states. The idea is to identify repeating events with perfect time alignment, which is the indication of being idling. We implement this idea by developing a novel statistical algorithm based on autocorrelation with time information incorporated. From our measurement results, we find that 88.5% of the detected idling services can be constrained with a simple syscall-based policy, which confines the process behaviors within its bookkeeping states. In addition, working with two IT departments (one of which is a cross validation), we receive positive feedbacks which show that about 30.6% of such services can be safely disabled or uninstalled directly. In the future, the IT department plan to incorporate the results to build a "smaller" OS installation image. Finally, we believe our measurement results raise the awareness of the potential security risks of idling services.

Sparse coding plays a key role in high dimensional data analysis. One critical challenge of sparse coding is to design a dictionary that is both adaptive to the training data and generalizable to unseen data of same type. In this paper, we propose a novel dictionary learning method to build an adaptive dictionary regularized by an a-priori over-completed dictionary. This leads to a sparse structure of the learned dictionary over the a-priori dictionary, and a sparse structure of the data over the learned dictionary. We apply the hierarchical sparse dictionary learning approach on both synthetic data and real-world high-dimensional time series data. The experimental results demonstrate that the hierarchical sparse dictionary learning approach reduces overfitting and enhances the generalizability of the learned dictionary. Moreover, the learned dictionary is optimized to adapt to the given data and result in a more compact dictionary and a more robust sparse representation. The experimental results on real datasets demonstrate that the proposed approach can successfully characterize the heterogeneity of the given data, and leads to a better and more robust dictionary.

Streaming heterogeneous information is ubiquitous in the era of Big Data, which provides versatile perspectives for more comprehensive understanding of behaviors of an underlying system/process. Human analysis of these volumes is infeasible, leading to unprecedented demands for mathematical tools which effectively parse and distill such data. However, the complicated nature of streaming heterogeneous data prevents the conventional multivariate data analysis methods being applied immediately. In this paper, we propose a novel framework together with an online algorithm, denoted as \(\mathtt {LSTH}\), for latent space tracking from heterogeneous data. Our method leverages the advantages of dimension reduction, correlation analysis and sparse learning to better reveal the latent relations among heterogeneous information and adapt to slow variations in streaming data. We applied our method on both synthetic and real data, and it achieves results competitive with or superior to the state-of-the-art in detecting several different types of anomalies.

The effectiveness of existing top-N recommendation methods decreases as the sparsity of the datasets increases. To alleviate this problem, we present an item-based method for generating top-N recommendations that learns the item-item similarity matrix as the product of two low dimensional latent factor matrices. These matrices are learned using a structural equation modeling approach, wherein the value being estimated is not used for its own estimation. A comprehensive set of experiments on multiple datasets at three different sparsity levels indicate that the proposed methods can handle sparse datasets effectively and outperforms other state-of-the-art top-N recommendation methods. The experimental results also show that the relative performance gains compared to competing methods increase as the data gets sparser.

The increasing amount of side information associated with the items in E-commerce applications has provided a very rich source of information that, once properly exploited and incorporated, can significantly improve the performance of the conventional recommender systems. This paper focuses on developing effective algorithms that utilize item side information for top-N recommender systems. A set of sparse linear methods with side information (SSLIM) is proposed, which involve a regularized optimization process to learn a sparse aggregation coefficient matrix based on both user-item purchase profiles and item side information. This aggregation coefficient matrix is used within an item-based recommendation framework to generate recommendations for the users. Our experimental results demonstrate that SSLIM outperforms other methods in effectively utilizing side information and achieving performance improvement.

The identification of small potent compounds that selectively bind to the target under consideration with high affinities is a critical step toward successful drug discovery. However, there is still a lack of efficient and accurate computational methods to predict compound selectivity properties. In this paper, we propose a set of machine learning methods to do compound selectivity prediction. In particular, we propose a novel cascaded learning method and a multitask learning method. The cascaded method decomposes the selectivity prediction into two steps, one model for each step, so as to effectively filter out nonselective compounds. The multitask method incorporates both activity and selectivity models into one multitask model so as to better differentiate compound selectivity properties. We conducted a comprehensive set of experiments and compared the results with those of other conventional selectivity prediction methods, and our results demonstrated that the cascaded and multitask methods significantly improve the selectivity prediction performance.

Mining chemical compounds in silico has drawn increasing attention from both academia and pharmaceutical industry due to its effectiveness in aiding the drug discovery process. Since graphs are the natural representation for chemical compounds, most of the mining algorithms focus on mining chemical graphs. Chemical graph mining approaches have many applications in the drug discovery process that include structure-activity-relationship (SAR) model construction and bioactivity classification, similar compound search and retrieval from chemical compound database, target identification from phenotypic assays, etc. Solving such problems in silico through studying and mining chemical graphs can provide novel perspective to medicinal chemists, biologist and toxicologist. Moreover, since the large scale chemical graph mining is usually employed at the early stages of drug discovery, it has the potential to speed up the entire drug discovery process. In this chapter, we discuss various problems and algorithms related to mining chemical graphs and describe some of the state-of-the-art chemical graph mining methodologies and their applications.

Bio-relation extraction (bRE), an important goal in bio-text mining, involves subtasks identifying relationships between bio-entities in text at multiple levels, e.g., at the article, sentence or relation level. A key limitation of current bRE systems is that they are restricted by the availability of annotated corpora. In this work we introduce a semisupervised approach that can tackle multi-level bRE via string comparisons with mismatches in the string kernel framework. Our string kernel implements an abstraction step, which groups similar words to generate more abstract entities, which can be learnt with unlabeled data. Specifically, two unsupervised models are proposed to capture contextual (local or global) semantic similarities between words from a large unannotated corpus. This Abstraction-augmented String Kernel (ASK) allows for better generalization of patterns learned from annotated data and provides a unified framework for solving bRE with multiple degrees of detail. ASK shows effective improvements over classic string kernels on four datasets and achieves state-of-the-art bRE performance without the need for complex linguistic features.

Structure-activity relationship (SAR) models are used to inform and to guide the iterative optimization of chemical leads, and they play a fundamental role in modern drug discovery. In this paper, we present a new class of methods for building SAR models, referred to as multi-assay based, that utilize activity information from different targets. These methods first identify a set of targets that are related to the target under consideration, and then they employ various machine learning techniques that utilize activity information from these targets in order to build the desired SAR model. We developed different methods for identifying the set of related targets, which take into account the primary sequence of the targets or the structure of their ligands, and we also developed different machine learning techniques that were derived by using principles of semi-supervised learning, multi-task learning, and classifier ensembles. The comprehensive evaluation of these methods shows that they lead to considerable improvements over the standard SAR models that are based only on the ligands of the target under consideration. On a set of 117 protein targets, obtained from PubChem, these multi-assay-based methods achieve a receiver-operating characteristic score that is, on the average, 7.0 -7.2% higher than that achieved by the standard SAR models. Moreover, on a set of targets belonging to six protein families, the multi-assay-based methods outperform chemogenomics-based approaches by 4.33%.

This paper focuses on developing classification algorithms for problems in which there is a need to predict the class based on multiple observations (examples) of the same phenomenon (class). These problems give rise to a new classification problem, referred to as set classification, that requires the prediction of a set of instances given the prior knowledge that all the instances of the set belong to the same unknown class. This problem falls under the general class of problems whose instances have class label dependencies. Four methods for solving the set classification problem are developed and studied. The first is based on a straightforward extension of the traditional classification paradigm whereas the other three are designed to explicitly take into account the known dependencies among the instances of the unlabeled set during learning or classification. A comprehensive experimental evaluation of the various methods and their underlying parameters shows that some of them lead to significant gains in performance.

Many enterprises nowadays are faced with a problem: Their mainframe legacy systems have great differences with the interface of the newly-developed system. Then a realistic problem is how to seamlessly integrate existing legacy systems into new systems, while not change the interface of legacy systems. By discussing the Improve transition method to transit the mainframe 3270 applications to the SOA environment, the mainframe application usage in the SOA environment is proposed. The article's focal point is the modeling of the business process, which does some expansion on the foundation of the Web service. Using CICS service flow feature, by Service Flow Modeler tool, an entire or partial business process is published as web services through service flow modeling method.