Resource management challenges frequently manifest in systems and networks as tough online decision tasks, for which the proper solution is dependent on an understanding of the workload and environment and facilitates smooth use of mobile edge and cloud resources. Due to the geographical dispersion of resources, constrained resource capacity, unpredictable nature of tasks, and network hierarchy present in such contexts, it is difficult to efficiently schedule jobs in edge environments. Unfortunately, existing heuristic-based methods lack generality and fast adaptability and thus cannot optimally solve such problems. The advantage actor–critic (A2C) method, on the one hand, can quickly adapt to dynamic circumstances based on relatively few data, and deep reinforcement learning (DRL) agents can on the other hand rapidly learn from their experience of environmental interactions to make better judgments. Therefore, we present an A2C-DRL real-time task scheduling technique for stochastic edge–cloud environments that enables decentralized learning and simultaneous work scheduling across multiple servers. With the aim of producing efficient scheduling decisions, we develop reward values for various resources and model the update policy, server resource scheduling method, and policy learning method. The model is adaptive and includes various hyperparameters that can be adjusted in accordance with the application requirements. We evaluate the load balancing capability of the model by introducing a load balancing factor. Experiments on real datasets show that the proposed A2C-DRL method outperforms seven state-of-the-art algorithms in terms of the reward value, task rejection, and the load balancing factor.

Abstract Using convolutional neural networks (CNNs) in classifying hyperspectral images (HSIs) has achieved quite good results in recent years. It is widely used in agricultural remote sensing, geological exploration, environmental monitoring, and marine remote sensing. Unfortunately, the complexity of network structures used for hyperspectral image classification challenges the efficient delivery of HSI data extremely, and existing methods suffer from a large amount of redundancy in the network weight parameters during training, as they either require huge computational resources or make inefficient use of storage space when designing the network structure, and many of the parameters that waste computational resources contribute less to the rich spectral and spatial information transfer in HSI. So we introduce LCTCS, a better low-memory and less-parametric network approach. LCTCS aims to improve the efficiency of computational resource utilization with advanced classification performance and lower levels of computational resources. Unlike the conventional 2D and 3D convolution used previously, we use simple and efficient 3D grouped convolution as a vehicle to convey the semantic features of HSIs. More specifically, we design a novel two-channel sparse network to classify HSIs since grouped 3D convolution conveys the properties of hyperspectral data well in the time and space domains.We have compared LCTCS with eight widely used network methods on four publicly available hyperspectral datasets for learning HSI information. A series of experiments shows that the model architecture designed has $$65.89 \%$$ 65.89 % less storage space than the DBDA method, consumes $$67.36 \%$$ 67.36 % fewer computational resources than the SSRN method on the IP dataset, and accomplishes a highly accurate classification task with the number of parameters accounting for only $$1.99 \%$$ 1.99 % that of the DBMA method.

Human leukocyte antigen (HLA) complex molecules play an essential role in immune interactions by presenting peptides on the cell surface to T cells. With significant deep learning progress, a series of neural network-based models have been proposed and demonstrated with their excellent performances for peptide-HLA class I binding prediction. However, there is still a lack of effective binding prediction models for HLA class II protein binding with peptides due to its inherent challenges. We present a novel sequence-based pan-specific neural network structure, DeepSeaPanII, for peptide-HLA class II binding prediction in this work. Our model is an end-to-end neural network model without the need for pre-or post-processing on input samples compared with existing pan-specific models. Besides state-of-the-art performance in binding affinity prediction, DeepSeqPanII can also extract biological insight on the binding mechanism over the peptide by its attention mechanism-based binding core prediction capability. The leave-one-allele-out cross-validation and benchmark evaluation results show that our proposed network model achieved state-of-the-art performance in HLA-II peptide binding. The source code and trained models are freely available at https://github.com/pcpLiu/DeepSeqPanII.

Deep learning methods have recently made significant contributions to sound event detection. These methods either use a block-level approach to distinguish parts of audio containing the event, or analyze the small frames of the audio separately. In this paper, we introduce a new method, AudioMask, for rare sound event detection by combining these two approaches. AudioMask first applies Mask R-CNN, a state-of-the-art algorithm for detecting objects in images, to the log mel-spectrogram of the audio files. Mask R-CNN detects audio segments that might contain the target event by generating bounding boxes around them in time-frequency domain. Then we use a frame-based audio event classifier trained independently from Mask R-CNN, to analyze each individual frame in the candidate segments proposed by Mask R-CNN. A post-processing step combines the outputs of the Mask R-CNN and the frame-level classifier to identify the true events. By evaluating AudioMask over the data sets from 2017 Detection and Classification of Acoustic Scenes and Events (DCASE) Challenge Task 2, We show that our algorithm performs better than the baseline models by 13.3% in the average F-score and achieves better results compared to the other non-ensemble methods in the challenge.

We herein propose an evolutionary multi-agent system (EMAS for short) to build an ensemble of surrogates for prediction. In our EMAS, we employ six kinds of basic surrogates, including Gaussian process, Kriging model, polynomial response surface, radial basis function, radial basis function neural network, and support vector regression machine. We define each surrogate as one agent and co-evolve parameters of basic surrogates to obtain the evolutionary weighted average surrogate, where sample cross-validation errors evaluate an ensemble of surrogates. The preliminary results from predicting the benchmark function with high dimension showed the effectiveness of our EMAS for an ensemble of surrogates.

Hematoxylin-Eosin (H&E) stained breast cancer (BC) histopathological images are widely used in breast cancer diagnosis, for which segmenting the cells/nuclei from the images is a major operation. However, most of current methods are sensitive to the characteristics of the images. In this paper, we propose a self-adjusting method for adaptive cell/nuclei segmentation from HE BC images, which works on BC images of different types and quality. To deal with the diversity of nuclei sizes, our approach employs an ellipse detector to estimate nuclei size, which is essential for noise elimination and for addressing the cell overlap issue. Then we propose a novel seed detection method based on the topological skeleton model in the marker-controlled watershed transform framework. The experiment result shows that our method is effective and efficient, and outperforms the state-of-the-art approaches on various types of HE BC images.

Protein-peptide interactions are the promising targets for potential protein drugs due to their critical role in many signal pathways with its small interaction interface. As a significant step in virtual screening in peptide drug discovery, binding affinity prediction is still a unsolved problem compared with achievements in docking. Most of current binding affinity prediction models are limited to a specific domain and is thus not applicable to many receptor domains that have few or no affinity data. To address this issue, domain independent prediction models are strongly needed. Traditional energy-based affinity prediction models are domain-independent, but still cannot give satisfactory results while impeded by its high computational cost. In this paper, we proposed a geometry shape based affinity prediction model. We evaluated our model using cross-validation on a non-redundant dataset of 336 complexes and and an external datasest: 592 human SH3 domain compelxes. Our experiments showed that our model is fast and achieved higher accuracy than the energy based models. Our model could be a useful affinity prediction tool for peptide docking and virtual screening in peptide drug discovery.

BACKGROUND: Computational prediction of protein subcellular localization can greatly help to elucidate its functions. Despite the existence of dozens of protein localization prediction algorithms, the prediction accuracy and coverage are still low. Several ensemble algorithms have been proposed to improve the prediction performance, which usually include as many as 10 or more individual localization algorithms. However, their performance is still limited by the running complexity and redundancy among individual prediction algorithms. RESULTS: This paper proposed a novel method for rational design of minimalist ensemble algorithms for practical genome-wide protein subcellular localization prediction. The algorithm is based on combining a feature selection based filter and a logistic regression classifier. Using a novel concept of contribution scores, we analyzed issues of algorithm redundancy, consensus mistakes, and algorithm complementarity in designing ensemble algorithms. We applied the proposed minimalist logistic regression (LR) ensemble algorithm to two genome-wide datasets of Yeast and Human and compared its performance with current ensemble algorithms. Experimental results showed that the minimalist ensemble algorithm can achieve high prediction accuracy with only 1/3 to 1/2 of individual predictors of current ensemble algorithms, which greatly reduces computational complexity and running time. It was found that the high performance ensemble algorithms are usually composed of the predictors that together cover most of available features. Compared to the best individual predictor, our ensemble algorithm improved the prediction accuracy from AUC score of 0.558 to 0.707 for the Yeast dataset and from 0.628 to 0.646 for the Human dataset. Compared with popular weighted voting based ensemble algorithms, our classifier-based ensemble algorithms achieved much better performance without suffering from inclusion of too many individual predictors. CONCLUSIONS: We proposed a method for rational design of minimalist ensemble algorithms using feature selection and classifiers. The proposed minimalist ensemble algorithm based on logistic regression can achieve equal or better prediction performance while using only half or one-third of individual predictors compared to other ensemble algorithms. The results also suggested that meta-predictors that take advantage of a variety of features by combining individual predictors tend to achieve the best performance. The LR ensemble server and related benchmark datasets are available at http://mleg.cse.sc.edu/LRensemble/cgi-bin/predict.cgi.

The fatigue test for LY12 aluminum alloys specimens at various stresses and stress ratios were performed. The distributions of the crack-nucleating particle sizes and ratios were determined using scanning electron microscopy (SEM) measured from specimens fracture surfaces. The results show that the correlation between particles sizes and ratios versus fatigue lives. And with the increase of stress ratio, fatigue crack originated from deeper subsurface at the same maximal stress. With the increase of maximal stress, fatigue crack originated from shallower subsurface or surface at same stress ratio. It has been also observed that the incubated fatigue micro cracks source localizes at the triple junction adjacent particle boundaries. The calculated results show that microcracks could nucleate and propagate from a smaller particle under a higher stress ratio, which agree reasonably with experimental observation.

BACKGROUND: Accurate prediction of binding residues involved in the interactions between proteins and small ligands is one of the major challenges in structural bioinformatics. Heme is an essential and commonly used ligand that plays critical roles in electron transfer, catalysis, signal transduction and gene expression. Although much effort has been devoted to the development of various generic algorithms for ligand binding site prediction over the last decade, no algorithm has been specifically designed to complement experimental techniques for identification of heme binding residues. Consequently, an urgent need is to develop a computational method for recognizing these important residues. RESULTS: Here we introduced an efficient algorithm HemeBIND for predicting heme binding residues by integrating structural and sequence information. We systematically investigated the characteristics of binding interfaces based on a non-redundant dataset of heme-protein complexes. It was found that several sequence and structural attributes such as evolutionary conservation, solvent accessibility, depth and protrusion clearly illustrate the differences between heme binding and non-binding residues. These features can then be separately used or combined to build the structure-based classifiers using support vector machine (SVM). The results showed that the information contained in these features is largely complementary and their combination achieved the best performance. To further improve the performance, an attempt has been made to develop a post-processing procedure to reduce the number of false positives. In addition, we built a sequence-based classifier based on SVM and sequence profile as an alternative when only sequence information can be used. Finally, we employed a voting method to combine the outputs of structure-based and sequence-based classifiers, which demonstrated remarkably better performance than the individual classifier alone. CONCLUSIONS: HemeBIND is the first specialized algorithm used to predict binding residues in protein structures for heme ligands. Extensive experiments indicated that both the structure-based and sequence-based methods have effectively identified heme binding residues while the complementary relationship between them can result in a significant improvement in prediction performance. The value of our method is highlighted through the development of HemeBIND web server that is freely accessible at http://mleg.cse.sc.edu/hemeBIND/.

Recent studies showed that protein-protein interaction network based features can significantly improve the prediction of protein subcellular localization. However, it is unclear whether network prediction models or other types of protein-protein correlation networks would also improve localization prediction. We present NetLoc, a novel diffusion kernel-based logistic regression (KLR) algorithm for predicting protein subcellular localization using four types of protein networks including physical protein-protein interaction (PPPI) networks, genetic PPI networks (GPPI), mixed PPI networks (MPPI), and co-expression networks (COEXP). We applied NetLoc to yeast protein localization prediction. The results showed that protein networks can provide rich information for protein localization prediction, achieving prediction performance up to AUC score of 0.93. We also showed that networks with high connectivity and high percentage of interacting protein pairs targeting the same location lead to better prediction performance. We found that physical PPPI is better than GPPI which is better than COEXP in terms of localization prediction. The prediction performance (AUC) using the yeast PPPI network ranges between 0.71 and 0.93 for 7 locations. Compared to the previous network feature based prediction algorithm which achieved AUC scores of (0.49 and 0.52) on the yeast PPI network of the DIP database, NetLoc achieved significantly better overall performance with the AUC of 0.74.

AutoDock is a widely used automated protein docking program in structure-based drug-design. Different search algorithms, such as Simulated Annealing, traditional Genetic Algorithm and Lamarckian Genetic Algorithm are used in AutoDock. However, the docking performance is still limited by the local optima issue in simulated annealing or the premature convergence issue existing in traditional evolutionary algorithms (EA). Because of the stochastic nature of the search algorithms, usually users need to do multiple runs to get reasonable docking results, which is time-consuming. We have developed a new docking algorithm AutoDockX by applying a sustainable GA named ALPS to the protein docking problem. We tested the docking performance over three different proteins (pr, cox and hsp90) with more than 20 candidate ligands for each protein. Our experiments showed that the sustainable GA based AutodockX gives significantly better docking performance than all the existing search algorithms implemented in the latest version of AutoDock4. Our algorithm has the benefits of less running time and much higher robustness. Instead of running a genetic algorithm or LGA search many times (e.g. 10), a single run of AutoDockX allows us to get better results. AutodockX thus has unique advantages in large-scale drug-candidate virtual screening.

AutoDock is a widely used automated protein docking program in structure-based drug-design. Different search algorithms such as simulated annealing, traditional genetic algorithm (GA) and Lamarckian genetic algorithm (LGA) are implemented in AutoDock. However, the docking performance of these algorithms is still limited by the local optima issue of simulated annealing or the premature convergence issue typical in traditional evolutionary algorithms (EA). Due to the stochastic nature of these search algorithms, users usually need to run multiple times to get reasonable docking results, which is time-consuming. We have developed a new docking program AutoDockX by applying a sustainable GA, Age-Layered Population Structure (ALPS) to the protein docking problem. We tested the docking performance over three different proteins (pr, cox and hsp90) with more than 20 candidate ligands for each protein. Our experiments showed that the sustainable GA based AutodockX achieved significantly better docking performance in terms of running time and robustness than all the existing search algorithms implemented in the latest version of AutoDock. AutodockX thus has unique advantages in large-scale virtual screening.

BACKGROUND: Protein sorting is the process that newly synthesized proteins are transported to their target locations within or outside of the cell. This process is precisely regulated by protein sorting signals in different forms. A major category of sorting signals are amino acid sub-sequences usually located at the N-terminals or C-terminals of protein sequences. Genome-wide experimental identification of protein sorting signals is extremely time-consuming and costly. Effective computational algorithms for de novo discovery of protein sorting signals is needed to improve the understanding of protein sorting mechanisms. METHODS: We formulated the protein sorting motif discovery problem as a classification problem and proposed a Bayesian classifier based algorithm (BayesMotif) for de novo identification of a common type of protein sorting motifs in which a highly conserved anchor is present along with a less conserved motif regions. A false positive removal procedure is developed to iteratively remove sequences that are unlikely to contain true motifs so that the algorithm can identify motifs from impure input sequences. RESULTS: Experiments on both implanted motif datasets and real-world datasets showed that the enhanced BayesMotif algorithm can identify anchored sorting motifs from pure or impure protein sequence dataset. It also shows that the false positive removal procedure can help to identify true motifs even when there is only 20% of the input sequences containing true motif instances. CONCLUSION: We proposed BayesMotif, a novel Bayesian classification based algorithm for de novo discovery of a special category of anchored protein sorting motifs from impure datasets. Compared to conventional motif discovery algorithms such as MEME, our algorithm can find less-conserved motifs with short highly conserved anchors. Our algorithm also has the advantage of easy incorporation of additional meta-sequence features such as hydrophobicity or charge of the motifs which may help to overcome the limitations of PWM (position weight matrix) motif model.

A typical cell has a size of only 10 mum while it contains about a billion proteins. Transportation of these proteins from their synthesis sites to their target locations within or outside of the cell is precisely controlled by protein sorting signals. However, genome-wide understanding of protein sorting regulatory signals and mechanisms is still very limited. We formulate the protein sorting motif discovery problem as a classification problem and proposed a Bayesian classifier based motif discovery algorithm (BayesMotif) to find a common type of sorting motifs in which a highly conserved anchor is present along with a less conserved motif regions. Experiments showed that our algorithm has the advantage of finding long lowly conserved sorting signals compared to other protein motif discovery algorithms such as MEME. Our algorithm also has the advantage to easily include additional meta-sequence features that overcomes the limitation of PWM (position weight matrix).

BACKGROUND: Disease classification has been an important application of microarray technology. However, most microarray-based classifiers can only handle data generated within the same study, since microarray data generated by different laboratories or with different platforms can not be compared directly due to systematic variations. This issue has severely limited the practical use of microarray-based disease classification. RESULTS: In this study, we tested the feasibility of disease classification by integrating the large amount of heterogeneous microarray datasets from the public microarray repositories. Cross-platform data compatibility is created by deriving expression log-rank ratios within datasets. One may then compare vectors of log-rank ratios across datasets. In addition, we systematically map textual annotations of datasets to concepts in Unified Medical Language System (UMLS), permitting quantitative analysis of the phenotype "distance" between datasets and automated construction of disease classes. We design a new classification approach named ManiSVM, which integrates Manifold data transformation with SVM learning to exploit the data properties. Using the leave one dataset out cross validation, ManiSVM achieved the overall accuracy of 70.7% (68.6% precision and 76.9% recall) with many disease classes achieving the accuracy higher than 80%. CONCLUSION: Our results not only demonstrated the feasibility of the integrated disease classification approach, but also showed that the classification accuracy increases with the number of homogenous training datasets. Thus, the power of the integrative approach will increase with the continuous accumulation of microarray data in public repositories. Our study shows that automated disease diagnosis can be an important and promising application of the enormous amount of costly to generate, yet freely available, public microarray data.

The rapid accumulation of microarray data translates into an urgent need for tools to perform integrative microarray analysis. Integrative Array Analyzer is a comprehensive analysis and visualization software toolkit, which aims to facilitate the reuse of the large amount of cross-platform and cross-species microarray data. It is composed of the data preprocess module, the co-expression analysis module, the differential expression analysis module, the functional and transcriptional annotation module and the graph visualization module.

BACKGROUND: Missing value estimation is an important preprocessing step in microarray analysis. Although several methods have been developed to solve this problem, their performance is unsatisfactory for datasets with high rates of missing data, high measurement noise, or limited numbers of samples. In fact, more than 80% of the time-series datasets in Stanford Microarray Database contain less than eight samples. RESULTS: We present the integrative Missing Value Estimation method (iMISS) by incorporating information from multiple reference microarray datasets to improve missing value estimation. For each gene with missing data, we derive a consistent neighbor-gene list by taking reference data sets into consideration. To determine whether the given reference data sets are sufficiently informative for integration, we use a submatrix imputation approach. Our experiments showed that iMISS can significantly and consistently improve the accuracy of the state-of-the-art Local Least Square (LLS) imputation algorithm by up to 15% improvement in our benchmark tests. CONCLUSION: We demonstrated that the order-statistics-based integrative imputation algorithms can achieve significant improvements over the state-of-the-art missing value estimation approaches such as LLS and is especially good for imputing microarray datasets with a limited number of samples, high rates of missing data, or very noisy measurements. With the rapid accumulation of microarray datasets, the performance of our approach can be further improved by incorporating larger and more appropriate reference datasets.

BACKGROUND: Understanding gene regulatory networks has become one of the central research problems in bioinformatics. More than thirty algorithms have been proposed to identify DNA regulatory sites during the past thirty years. However, the prediction accuracy of these algorithms is still quite low. Ensemble algorithms have emerged as an effective strategy in bioinformatics for improving the prediction accuracy by exploiting the synergetic prediction capability of multiple algorithms. RESULTS: We proposed a novel clustering-based ensemble algorithm named EMD for de novo motif discovery by combining multiple predictions from multiple runs of one or more base component algorithms. The ensemble approach is applied to the motif discovery problem for the first time. The algorithm is tested on a benchmark dataset generated from E. coli RegulonDB. The EMD algorithm has achieved 22.4% improvement in terms of the nucleotide level prediction accuracy over the best stand-alone component algorithm. The advantage of the EMD algorithm is more significant for shorter input sequences, but most importantly, it always outperforms or at least stays at the same performance level of the stand-alone component algorithms even for longer sequences. CONCLUSION: We proposed an ensemble approach for the motif discovery problem by taking advantage of the availability of a large number of motif discovery programs. We have shown that the ensemble approach is an effective strategy for improving both sensitivity and specificity, thus the accuracy of the prediction. The advantage of the EMD algorithm is its flexibility in the sense that a new powerful algorithm can be easily added to the system.

Most existing research on robust design using evolutionary algorithms (EA) follows the paradigm of traditional robust design, in which parameters of a design solution are tuned to improve the robustness of the system. However, the topological structure of a system may set a limit on the possible robustness achievable through parameter tuning. This paper proposes a new robust design paradigm that exploits the open-ended topological synthesis capability of genetic programming to evolve more robust systems. As a case study, a methodology for automated synthesis of dynamic systems, based on genetic programming and bond graph modeling (GPBG), is applied to evolve robust low-pass and high-pass analog filters. Compared with a traditional robust design approach based on a state-of-the-art real-parameter genetic algorithm (GA), it is shown that open-ended topology search by genetic programming with a fitness criterion rewarding robustness can evolve more robust systems with respect to parameter perturbations than what was achieved through parameter tuning alone, for our test problems.

Many current Evolutionary Algorithms (EAs) suffer from a tendency to converge prematurely or stagnate without progress for complex problems. This may be due to the loss of or failure to discover certain valuable genetic material or the loss of the capability to discover new genetic material before convergence has limited the algorithm's ability to search widely. In this paper, the Hierarchical Fair Competition (HFC) model, including several variants, is proposed as a generic framework for sustainable evolutionary search by transforming the convergent nature of the current EA framework into a non-convergent search process. That is, the structure of HFC does not allow the convergence of the population to the vicinity of any set of optimal or locally optimal solutions. The sustainable search capability of HFC is achieved by ensuring a continuous supply and the incorporation of genetic material in a hierarchical manner, and by culturing and maintaining, but continually renewing, populations of individuals of intermediate fitness levels. HFC employs an assembly-line structure in which subpopulations are hierarchically organized into different fitness levels, reducing the selection pressure within each subpopulation while maintaining the global selection pressure to help ensure the exploitation of the good genetic material found. Three EAs based on the HFC principle are tested - two on the even-10-parity genetic programming benchmark problem and a real-world analog circuit synthesis problem, and another on the HIFF genetic algorithm (GA) benchmark problem. The significant gain in robustness, scalability and efficiency by HFC, with little additional computing effort, and its tolerance of small population sizes, demonstrates its effectiveness on these problems and shows promise of its potential for improving other existing EAs for difficult problems. A paradigm shift from that of most EAs is proposed: rather than trying to escape from local optima or delay convergence at a local optimum, HFC allows the emergence of new optima continually in a bottom-up manner, maintaining low local selection pressure at all fitness levels, while fostering exploitation of high-fitness individuals through promotion to higher levels.

We discuss the hierarchy that is involved in a typical MEMS design and how evolutionary approaches can be used to automate the hierarchical design and synthesis process for MEMS. At the system level, the approach combining bond graphs and genetic programming can lead to satisfactory design candidates of system level models that meet the predefined behavioral specifications for designers to tradeoff. At the physical layout synthesis level, the selection of geometric parameters for component devices is formulated as a constrained optimization problem and addressed using a constrained GA approach. A multiple-resonator microsystem design is used to illustrate the integrated design automation idea using evolutionary approaches.

The wireless access point configuration problem in wireless LAN deployment can be formulated as a nonlinear optimization problem with a variable number of parameters. In this paper, strongly-typed genetic programming is applied to solve an abstract version of this problem successfully. It is argued that this problem can be used as a potential benchmark problem for evaluating techniques and investigating issues in strongly typed genetic programming, topologically open-ended synthesis by genetic programming, and simultaneous topological and parametric search.

The HFC model for evolutionary computation is inspired by the stratified competition often seen in society and biology. Subpopulations are stratified by fitness. Individuals move from low-fitness subpopulations to higher-fitness subpopulations if and only if they exceed the fitness-based admission threshold of the receiving subpopulation, but not of a higher one. HFC's balanced exploration and exploitation, while avoiding premature convergence, is shown on a genetic programming example.