EpistoNet: an ensemble of Epistocracy-optimized mixture of experts for detecting COVID-19 on chest X-ray images.

The Coronavirus has spread across the world and infected millions of people, causing devastating damage to the public health and global economies. To mitigate the impact of the coronavirus a reliable, fast, and accurate diagnostic system should be promptly implemented. In this study, we propose EpistoNet, a decision tree-based ensemble model using two mixtures of discriminative experts to classify COVID-19 lung infection from chest X-ray images. To optimize the architecture and hyper-parameters of the designed neural networks, we employed Epistocracy algorithm, a recently proposed hyper-heuristic evolutionary method. Using 2500 chest X-ray images consisting of 1250 COVID-19 and 1250 non-COVID-19 cases, we left out 500 images for testing and partitioned the remaining 2000 images into 5 different clusters using K-means clustering algorithm. We trained multiple deep convolutional neural networks on each cluster to help build a mixture of strong discriminative experts from the top-performing models supervised by a gating network. The final ensemble model obtained 95% accuracy on COVID-19 images and 93% accuracy on non-COVID-19. The experimental results show that EpistoNet can accurately, and reliably be used to detect COVID-19 infection in the chest X-ray images, and Epistocracy algorithm can be effectively used to optimize the hyper-parameters of the proposed models.

Surface Registration with Eigenvalues and Eigenvectors.

This paper presents a novel surface registration technique using the spectrum of the shapes, which can facilitate accurate localization and visualization of non-isometric deformations of the surfaces. In order to register two surfaces, we map both eigenvalues and eigenvectors of the Laplace-Beltrami of the shapes through optimizing an energy function. The function is defined by the integration of a smoothness term to align the eigenvalues and a distance term between the eigenvectors at feature points to align the eigenvectors. The feature points are generated using the static points of certain eigenvectors of the surfaces. By using both the eigenvalues and the eigenvectors on these feature points, the computational efficiency is improved considerably without losing the accuracy in comparison to the approaches that use the eigenvectors for all vertices. In our technique, the variation of the shape is expressed using a scale function defined at each vertex. Consequently, the total energy function to align the two given surfaces can be defined using the linear interpolation of the scale function derivatives. Through the optimization of the energy function, the scale function can be solved and the alignment is achieved. After the alignment, the eigenvectors can be employed to calculate the point-to-point correspondence of the surfaces. Therefore, the proposed method can accurately define the displacement of the vertices. We evaluate our method by conducting experiments on synthetic and real data using hippocampus, heart, and hand models. We also compare our method with non-rigid Iterative Closest Point (ICP) and a similar spectrum-based methods. These experiments demonstrate the advantages and accuracy of our method.

Bias and Stability of Single Variable Classifiers for Feature Ranking and Selection.

Feature rankings are often used for supervised dimension reduction especially when discriminating power of each feature is of interest, dimensionality of dataset is extremely high, or computational power is limited to perform more complicated methods. In practice, it is recommended to start dimension reduction via simple methods such as feature rankings before applying more complex approaches. Single Variable Classifier (SVC) ranking is a feature ranking based on the predictive performance of a classifier built using only a single feature. While benefiting from capabilities of classifiers, this ranking method is not as computationally intensive as wrappers. In this paper, we report the results of an extensive study on the bias and stability of such feature ranking method. We study whether the classifiers influence the SVC rankings or the discriminative power of features themselves has a dominant impact on the final rankings. We show the common intuition of using the same classifier for feature ranking and final classification does not always result in the best prediction performance. We then study if heterogeneous classifiers ensemble approaches provide more unbiased rankings and if they improve final classification performance. Furthermore, we calculate an empirical prediction performance loss for using the same classifier in SVC feature ranking and final classification from the optimal choices.

Confident Surgical Decision Making in Temporal Lobe Epilepsy by Heterogeneous Classifier Ensembles.

In medical domains with low tolerance for invalid predictions, classification confidence is highly important and traditional performance measures such as overall accuracy cannot provide adequate insight into classifications reliability. In this paper, a confident-prediction rate (CPR) which measures the upper limit of confident predictions has been proposed based on receiver operating characteristic (ROC) curves. It has been shown that heterogeneous ensemble of classifiers improves this measure. This ensemble approach has been applied to lateralization of focal epileptogenicity in temporal lobe epilepsy (TLE) and prediction of surgical outcomes. A goal of this study is to reduce extraoperative electrocorticography (eECoG) requirement which is the practice of using electrodes placed directly on the exposed surface of the brain. We have shown that such goal is achievable with application of data mining techniques. Furthermore, all TLE surgical operations do not result in complete relief from seizures and it is not always possible for human experts to identify such unsuccessful cases prior to surgery. This study demonstrates the capability of data mining techniques in prediction of undesirable outcome for a portion of such cases.

Automated segmentation and classification of high throughput yeast assay spots.

Several technologies for characterizing genes and proteins from humans and other organisms use yeast growth or color development as read outs. The yeast two-hybrid assay, for example, detects protein-protein interactions by measuring the growth of yeast on a specific solid medium, or the ability of the yeast to change color when grown on a medium containing a chromogenic substrate. Current systems for analyzing the results of these types of assays rely on subjective and inefficient scoring of growth or color by human experts. Here, an image analysis system is described for scoring yeast growth and color development in high throughput biological assays. The goal is to locate the spots and score them in color images of two types of plates named "X-Gal" and "growth assay" plates, with uniformly placed spots (cell areas) on each plate (both plates in one image). The scoring system relies on color for the X-Gal spots, and texture properties for the growth assay spots. A maximum likelihood projection-based segmentation is developed to automatically locate spots of yeast on each plate. Then color histogram and wavelet texture features are extracted for scoring using an optimal linear transformation. Finally, an artificial neural network is used to score the X-Gal and growth assay spots using the extracted features. The performance of the system is evaluated using spots of 60 images. After training the networks using training and validation sets, the system was assessed on the test set. The overall accuracies of 95.4% and 88.2% are achieved, respectively, for scoring the X-Gal and growth assay spots.

A proteome-wide protein interaction map for Campylobacter jejuni.

BACKGROUND: Data from large-scale protein interaction screens for humans and model eukaryotes have been invaluable for developing systems-level models of biological processes. Despite this value, only a limited amount of interaction data is available for prokaryotes. Here we report the systematic identification of protein interactions for the bacterium Campylobacter jejuni, a food-borne pathogen and a major cause of gastroenteritis worldwide. RESULTS: Using high-throughput yeast two-hybrid screens we detected and reproduced 11,687 interactions. The resulting interaction map includes 80% of the predicted C. jejuni NCTC11168 proteins and places a large number of poorly characterized proteins into networks that provide initial clues about their functions. We used the map to identify a number of conserved subnetworks by comparison to protein networks from Escherichia coli and Saccharomyces cerevisiae. We also demonstrate the value of the interactome data for mapping biological pathways by identifying the C. jejuni chemotaxis pathway. Finally, the interaction map also includes a large subnetwork of putative essential genes that may be used to identify potential new antimicrobial drug targets for C. jejuni and related organisms. CONCLUSION: The C. jejuni protein interaction map is one of the most comprehensive yet determined for a free-living organism and nearly doubles the binary interactions available for the prokaryotic kingdom. This high level of coverage facilitates pathway mapping and function prediction for a large number of C. jejuni proteins as well as orthologous proteins from other organisms. The broad coverage also facilitates cross-species comparisons for the identification of evolutionarily conserved subnetworks of protein interactions.

A database and tool, IM Browser, for exploring and integrating emerging gene and protein interaction data for Drosophila.

BACKGROUND: Biological processes are mediated by networks of interacting genes and proteins. Efforts to map and understand these networks are resulting in the proliferation of interaction data derived from both experimental and computational techniques for a number of organisms. The volume of this data combined with the variety of specific forms it can take has created a need for comprehensive databases that include all of the available data sets, and for exploration tools to facilitate data integration and analysis. One powerful paradigm for the navigation and analysis of interaction data is an interaction graph or map that represents proteins or genes as nodes linked by interactions. Several programs have been developed for graphical representation and analysis of interaction data, yet there remains a need for alternative programs that can provide casual users with rapid easy access to many existing and emerging data sets. DESCRIPTION: Here we describe a comprehensive database of Drosophila gene and protein interactions collected from a variety of sources, including low and high throughput screens, genetic interactions, and computational predictions. We also present a program for exploring multiple interaction data sets and for combining data from different sources. The program, referred to as the Interaction Map (IM) Browser, is a web-based application for searching and visualizing interaction data stored in a relational database system. Use of the application requires no downloads and minimal user configuration or training, thereby enabling rapid initial access to interaction data. IM Browser was designed to readily accommodate and integrate new types of interaction data as it becomes available. Moreover, all information associated with interaction measurements or predictions and the genes or proteins involved are accessible to the user. This allows combined searches and analyses based on either common or technique-specific attributes. The data can be visualized as an editable graph and all or part of the data can be downloaded for further analysis with other tools for specific applications. The database is available at http://proteome.wayne.edu/PIMdb.html CONCLUSION: The Drosophila Interactions Database described here places a variety of disparate data into one easily accessible location. The database has a simple structure that maintains all relevant information about how each interaction was determined. The IM Browser provides easy, complete access to this database and could readily be used to publish other sets of interaction data. By providing access to all of the available information from a variety of data types, the program will also facilitate advanced computational analyses.

Computational approaches for predicting protein-protein interactions: a survey.

Discovery of the protein interactions that take place within a cell can provide a starting point for understanding biological regulatory pathways. Global interaction patterns among proteins, for example, can suggest new drug targets and aid the design of new drugs by providing a clearer picture of the biological pathways in the neighborhoods of the drug targets. High-throughput experimental screens have been developed to detect protein-protein interactions, however, they show high rates of errors in terms of false positives and false negatives. Many computational approaches have been proposed to tackle the problem of protein-protein interaction prediction. They range from comparative genomics based methods to data integration based approaches. Challenging properties of protein-protein interaction data have to be addressed appropriately before a higher quality interaction map with better coverage can be achieved. This paper presents a survey of major works in computational prediction of protein-protein interactions, explaining their assumptions, main ideas, and limitations.

CASMIL: a comprehensive software/toolkit for image-guided neurosurgeries.

BACKGROUND: CASMIL aims to develop a cost-effective and efficient approach to monitor and predict deformation during surgery, allowing accurate, and real-time intra-operative information to be provided reliably to the surgeon. METHOD: CASMIL is a comprehensive Image-guided Neurosurgery System with extensive novel features. It is an integration of various modules including rigid and non-rigid body co-registration (image-image, image-atlas, and image-patient), automated 3D segmentation, brain shift predictor, knowledge based query tools, intelligent planning, and augmented reality. One of the vital and unique modules is the Intelligent Planning module, which displays the best surgical corridor on the computer screen based on tumor location, captured surgeon knowledge, and predicted brain shift using patient specific Finite Element Model. Also, it has multi-level parallel computing to provide near real-time interaction with iMRI (Intra-operative MRI). In addition, it has been securely web-enabled and optimized for remote web and PDA access. RESULTS: A version of this system is being used and tested using real patient data and is expected to be in use in the operating room at the Detroit Medical Center in the first half of 2006. CONCLUSION: CASMIL is currently under development and is targeted for minimally invasive surgeries. With minimal changes to the design, it can be easily extended and made available for other surgical procedures.

Content-based image database system for epilepsy.

We have designed and implemented a human brain multi-modality database system with content-based image management, navigation and retrieval support for epilepsy. The system consists of several modules including a database backbone, brain structure identification and localization, segmentation, registration, visual feature extraction, clustering/classification and query modules. Our newly developed anatomical landmark localization and brain structure identification method facilitates navigation through an image data and extracts useful information for segmentation, registration and query modules. The database stores T1-, T2-weighted and FLAIR MRI and ictal/interictal SPECT modalities with associated clinical data. We confine the visual feature extractors within anatomical structures to support semantically rich content-based procedures. The proposed system serves as a research tool to evaluate a vast number of hypotheses regarding the condition such as resection of the hippocampus with a relatively small volume and high average signal intensity on FLAIR. Once the database is populated, using data mining tools, partially invisible correlations between different modalities of data, modeled in database schema, can be discovered. The design and implementation aspects of the proposed system are the main focus of this paper.

Incremental genetic K-means algorithm and its application in gene expression data analysis.

BACKGROUND: In recent years, clustering algorithms have been effectively applied in molecular biology for gene expression data analysis. With the help of clustering algorithms such as K-means, hierarchical clustering, SOM, etc, genes are partitioned into groups based on the similarity between their expression profiles. In this way, functionally related genes are identified. As the amount of laboratory data in molecular biology grows exponentially each year due to advanced technologies such as Microarray, new efficient and effective methods for clustering must be developed to process this growing amount of biological data. RESULTS: In this paper, we propose a new clustering algorithm, Incremental Genetic K-means Algorithm (IGKA). IGKA is an extension to our previously proposed clustering algorithm, the Fast Genetic K-means Algorithm (FGKA). IGKA outperforms FGKA when the mutation probability is small. The main idea of IGKA is to calculate the objective value Total Within-Cluster Variation (TWCV) and to cluster centroids incrementally whenever the mutation probability is small. IGKA inherits the salient feature of FGKA of always converging to the global optimum. C program is freely available at http://database.cs.wayne.edu/proj/FGKA/index.htm. CONCLUSIONS: Our experiments indicate that, while the IGKA algorithm has a convergence pattern similar to FGKA, it has a better time performance when the mutation probability decreases to some point. Finally, we used IGKA to cluster a yeast dataset and found that it increased the enrichment of genes of similar function within the cluster.