Generative Neural Networks (GNN) have demonstrated remarkable power in creating novel graphic design images from text-to-image training. This work applies GNNs to design dynamic inductive power transfer systems to make charging electrical vehicles (EVs) more convenient and cheaper. Discovering optimal and safe coil implementations (for EV and road) is challenging because of the combinatorial explosion of possible configurations, along with multiple conflicting objective functions, such as maximizing the output power, while minimizing stray magnetic fields and the volume of windings and magnetic cores. To solve the problem, a differentiable simulator and evaluator define loss functions that train a generative neural network to only produce configurations that satisfy eight given design criteria. Before training, the rate of finding successful designs is 0.005%, but within 500 training epochs, the rate becomes 98% (about 30 seconds run time). Solution diversity and quality may be further improved by applying loss functions trained on pairwise Pareto-optimal generated examples.

Host-pathogen protein interactions (HPPIs) play vital roles in many biological processes and are directly involved in infectious diseases. With the outbreak of more frequent pandemics in the last couple of decades, such as the recent outburst of Covid-19 causing millions of deaths, it has become more critical to develop advanced methods to accurately predict pathogen interactions with their respective hosts. During the last decade, experimental methods to identify HPIs have been used to decipher host-pathogen systems with the caveat that those techniques are labor-intensive, expensive and time-consuming. Alternatively, accurate prediction of HPIs can be performed by the use of data-driven machine learning. To provide a more robust and accurate solution for the HPI prediction problem, we have developed a deepHPI tool based on deep learning. The web server delivers four host-pathogen model types: plant-pathogen, human-bacteria, human-virus and animal-pathogen, leveraging its operability to a wide range of analyses and cases of use. The deepHPI web tool is the first to use convolutional neural network models for HPI prediction. These models have been selected based on a comprehensive evaluation of protein features and neural network architectures. The best prediction models have been tested on independent validation datasets, which achieved an overall Matthews correlation coefficient value of 0.87 for animal-pathogen using the combined pseudo-amino acid composition and conjoint triad (PAAC_CT) features, 0.75 for human-bacteria using the combined pseudo-amino acid composition, conjoint triad and normalized Moreau-Broto feature (PAAC_CT_NMBroto), 0.96 for human-virus using PAAC_CT_NMBroto and 0.94 values for plant-pathogen interactions using the combined pseudo-amino acid composition, composition and transition feature (PAAC_CTDC_CTDT). Our server running deepHPI is deployed on a high-performance computing cluster that enables large and multiple user requests, and it provides more information about interactions discovered. It presents an enriched visualization of the resulting host-pathogen networks that is augmented with external links to various protein annotation resources. We believe that the deepHPI web server will be very useful to researchers, particularly those working on infectious diseases. Additionally, many novel and known host-pathogen systems can be further investigated to significantly advance our understanding of complex disease-causing agents. The developed models are established on a web server, which is freely accessible at http://bioinfo.usu.edu/deepHPI/.

Billions of dollars are traded automatically in the stock market every day, including algorithms that use neural networks, but there are still questions regarding how neural networks trade. The black box nature of a neural network gives pause to entrusting it with valuable trading funds. A more recent technique for the study of neural networks, feature map visualizations, yields insight into how a neural network generates an output. Utilizing a Convolutional Neural Network (CNN) with candlestick images as input and feature map visualizations gives a unique opportunity to determine what in the input images is causing the neural network to output a certain action. In this study, a CNN is utilized within a Double Deep Q-Network (DDQN) to outperform the S&P 500 Index returns, and also analyze how the system trades. The DDQN is trained and tested on the 30 largest stocks in the S&P 500. Following training the CNN is used to generate feature map visualizations to determine where the neural network is placing its attention on the candlestick images. Results show that the DDQN is able to yield higher returns than the S&P 500 Index between January 2, 2020 and June 30, 2020. Results also show that the CNN is able to switch its attention from all the candles in a candlestick image to the more recent candles in the image, based on an event such as the coronavirus stock market crash of 2020.

DBMS performance is dependent on many parameters, such as index selection, cache size, physical layout, and data partitioning. Some combinations of these parameters can lead to optimal performance for a given workload but selecting an optimal or near-optimal combination is challenging, especially for large databases with complex workloads. Among the hundreds of parameters, index selection is arguably the most critical parameter for performance. We propose a self-administered framework, called the Multiple Type and Attribute Index Selector (MANTIS), that automatically selects near-optimal indexes. The framework advances the state-of-the-art index selection by considering both multi-attribute and multiple types of indexes within a bounded storage size constraint, a combination not previously addressed. MANTIS combines supervised and reinforcement learning, a Deep Neural Network recommends the type of index for a given workload while a Deep Q-Learning network recommends the multi-attribute aspect. MANTIS is sensitive to storage cost constraints and incorporates noisy rewards in its reward function for better performance. Our experimental evaluation shows that MANTIS outperforms the current state-of-art methods by an average of 9.53% QphH@size.

Comparisons between machine learning and optimal transport-based approaches in classifying images are made in underwater orbital angular momentum (OAM) communications. A model is derived that justifies optimal transport for use in attenuated water environments. OAM pattern demultiplexing is performed using optimal transport and deep neural networks and compared to each other. Additionally, some of the complications introduced by signal attenuation are highlighted. The Radon cumulative distribution transform (R-CDT) is applied to OAM patterns to transform them to a linear subspace. The original OAM images and the R-CDT transformed patterns are used in several classification algorithms, and results are compared. The selected classification algorithms are the nearest subspace algorithm, a shallow convolutional neural network (CNN), and a deep neural network. It is shown that the R-CDT transformed images are more accurate than the original OAM images in pattern classification. Also, the nearest subspace algorithm performs better than the selected CNNs in OAM pattern classification in underwater environments.

Turbulence and attenuation are signal degrading factors that can severely hinder free-space and underwater OAM optical pattern demultiplexing. A variety of state-of-the-art convolutional neural network architectures are explored to identify which, if any, provide optimal performance under these non-ideal environmental conditions. Hyperparameter searches are performed on the architectures to ensure that near-ideal settings are used for training. Architectures are compared in various scenarios and the best performing, with their settings, are provided. We show that from the current state-of-the-art architectures, DenseNet outperforms all others when memory is not a constraint. When memory footprint is a factor, ShuffleNet is shown to performed the best.

Anaerobic granular sludge comprises of highly organized microorganisms with a sophisticated metabolic network. Such aggregates can withstand storage, temperature fluctuations and changes in the substrate supplied for anaerobic digestion. However, substrate change leads to long adaptation of granular consortia, creating lags in the reactor operations. To speed up adaptation and increase digestion efficiency, bioaugmentation with a robust consortium can be performed. The computational study described here aims to elucidate the mechanisms of bioaugmenting anaerobic granules, utilizing the current body of knowledge on metabolic and biochemical interactions between bacteria in such aggregates. Using a cDynoMiCs simulation environment, an agent-based model was developed to describe bioaugmentation for adaptation of cellobiose-degrading granular consortium to a lipid-rich feed. Lipolytic bacteria were successfully incorporated in silico to the stable granular consortia after 40 days of simulation. The ratio of cellobiose and the lipid-derivative, oleate, in the feed played key role to ensure augmentation. At 0.5 g/L of both cellobiose and oleate in the feed, a homogeneous stable augmented consortium was formed and converted the given amount of substrate to 10.9 mg/L of methane as a final product of anaerobic digestion. The demonstrated model can be used as a planning tool for anaerobic digestion facilities considering transition of the inoculum to a new type of feed.

Signal attenuation, resulting in low signal to noise ratio (SNR), in underwater optical communications (UWOC) is a problem that degrades classification performance. We develop and contrast several novel ways to create machine learning (ML) and optimal transport-based attenuation models and insert these models in a convolutional neural network (CNN) classification training pipeline. We show that including these ML-based attenuation models in the CNN classifier training, significantly improves classification performance.

Signal attenuation in underwater communications is a problem that degrades classification performance. Several novel CNN-based (SMART) models are developed to capture the physics of the attenuation process. One model is built and trained using automatic differentiation and another uses the radon cumulative distribution transform. These models are inserted in the classifier training pipeline. It is shown that including these attenuation models in classifier training significantly improves classification performance when the trained model is tested with environmentally attenuated images. The improved classification accuracy will be important in future OAM underwater optical communication applications.

BACKGROUND:Multicellular entities like mammalian tissues or microbial biofilms typically exhibit complex spatial arrangements that are adapted to their specific functions or environments. These structures result from intercellular signaling as well as from the interaction with the environment that allow cells of the same genotype to differentiate into well-organized communities of diversified cells. Despite its importance, our understanding how this cell-cell and metabolic coupling lead to functionally optimized structures is still limited.RESULTS:Here, we present a data-driven spatial framework to computationally investigate the development of yeast colonies as such a multicellular structure in dependence on metabolic capacity. For this purpose, we first developed and parameterized a dynamic cell state and growth model for yeast based on on experimental data from homogeneous liquid media conditions. The inferred model is subsequently used in a spatially coarse-grained model for colony development to investigate the effect of metabolic coupling by calibrating spatial parameters from experimental time-course data of colony growth using state-of-the-art statistical techniques for model uncertainty and parameter estimations. The model is finally validated by independent experimental data of an alternative yeast strain with distinct metabolic characteristics and illustrates the impact of metabolic coupling for structure formation.CONCLUSIONS:We introduce a novel model for yeast colony formation, present a statistical methodology for model calibration in a data-driven manner, and demonstrate how the established model can be used to generate predictions across scales by validation against independent measurements of genetically distinct yeast strains.

The complex morphologies observed in many biofilms play a critical role in the survival of these microbial communities. Recently, the formation of wrinkles has been the focus of many studies aimed at finding fundamental information on morphogenesis during development. While the underlying genetic mechanisms of wrinkling are not well-understood, recent discoveries have led to the counterintuitive idea that wrinkle formation is triggered by localized cell death. This work examines the hypothesis that the material properties of a biofilm both power and control wrinkle formation within biofilms in response to localized cell death. Using an agent-based model and a high-performance platform (Biocellion), we built a model that qualitatively reproduced wrinkle formation in biofilms due to cell death. Through the use of computational simulations, we determined important relationships between cellular level mechanical interactions and changes in colony morphology. These simulations were also used to identify significant cellular interactions that are required for wrinkle formation. These results are a first step towards more comprehensive models that, in combination with experimental observations, will improve our understanding of the morphological development of bacterial biofilms.

Efficient metabolism in bioengineered tissues requires a robust vascular system to provide healthy microenvironments to the cells and stroma. Such networks form spontaneously during embryogenesis from randomly distributed endothelial cells. There is a need to bioengineer endothelial cells so that network formation and operation is optimal for synthetic tissues. This work introduces a computational model that simulates de novo vascular development and assesses the effectiveness of the network in delivering nutrients and extracting waste from tissue. A genetic algorithm was employed to identify parameter values of the vaculogenesis model that lead to the most efficient and robust vascular structures. These parameter values control the behavior of cell-level mechanisms such as chemotaxis and adhesion. These studies demonstrate that genetic algorithms are effective at identifying model parameters that lead to near-optimal networks. This work suggests that computational modeling and optimization approaches may improve the effectiveness of engineered tissues by suggesting target cellular mechanisms for modification.

BACKGROUND:A unique combination of mechanical, physiochemical and biological forces influences granulation during processes of anaerobic digestion. Understanding this process requires a systems biology approach due to the need to consider not just single-cell metabolic processes, but also the multicellular organization and development of the granule.RESULTS:In this computational experiment, we address the role that physiochemical and biological processes play in granulation and provide a literature-validated working model of anaerobic granule de novo formation. The agent-based model developed in a cDynoMiCs simulation environment successfully demonstrated a de novo granulation in a glucose fed system, with the average specific methanogenic activity of 1.11 ml C H 4/g biomass and formation of a 0.5 mm mature granule in 33 days. The simulated granules exhibit experimental observations of radial stratification: a central dead core surrounded by methanogens then encased in acidogens. Practical application of the granulation model was assessed on the anaerobic digestion of low-strength wastewater by measuring the changes in methane yield as experimental configuration parameters were systematically searched.CONCLUSIONS:In the model, the emergence of multicellular organization of anaerobic granules from randomly mixed population of methanogens and acidogens was observed and validated. The model of anaerobic de novo granulation can be used to predict the morphology of the anaerobic granules in a alternative substrates of interest and to estimate methane potential of the resulting microbial consortia. The study demonstrates a successful integration of a systems biology approach to model multicellular systems with the engineering of an efficient anaerobic digestion system.

The dynamics of gene regulatory networks (GRNs) guide cellular differentiation. Determining the ways regulatory genes control expression of their targets is essential to understand and control cellular differentiation. The way a regulatory gene controls its target can be expressed as a gene regulatory function. Manual derivation of these regulatory functions is slow, error-prone and difficult to update as new information arises. Automating this process is a significant challenge and the subject of intensive effort. This work presents a novel approach to discovering biologically plausible gene regulatory interactions that control cellular differentiation. This method integrates known cell type expression data, genetic interactions, and knowledge of the effects of gene knockouts to determine likely GRN regulatory functions. We employ a genetic algorithm to search for candidate GRNs that use a set of transcription factors that control differentiation within a lineage. Nested canalyzing functions are used to constrain the search space to biologically plausible networks. The method identifies an ensemble of GRNs whose dynamics reproduce the gene expression pattern for each cell type within a particular lineage. The method's effectiveness was tested by inferring consensus GRNs for myeloid and pancreatic cell differentiation and comparing the predicted gene regulatory interactions to manually derived interactions. We identified many regulatory interactions reported in the literature and also found differences from published reports. These discrepancies suggest areas for biological studies of myeloid and pancreatic differentiation. We also performed a study that used defined synthetic networks to evaluate the accuracy of the automated search method and found that the search algorithm was able to discover the regulatory interactions in these defined networks with high accuracy. We suggest that the GRN functions derived from the methods described here can be used to fill gaps in knowledge about regulatory interactions and to offer hypotheses for experimental testing of GRNs that control differentiation and other biological processes.

The productivity of bioengineered cell factories is limited by inefficiencies in nutrient delivery and waste and product removal. Current solution approaches explore changes in the physical configurations of the bioreactors. This work investigates the possibilities of exploiting self-organizing vascular networks to support producer cells within the factory. A computational model simulates de novo vascular development of endothelial-like cells and the resultant network functioning to deliver nutrients and extract product and waste from the cell culture. Microbial factories with vascular networks are evaluated for their scalability, robustness, and productivity compared to the cell factories without a vascular network. Initial studies demonstrate that at least an order of magnitude increase in production is possible, the system can be scaled up, and the self-organization of an efficient vascular network is robust. The work suggests that bioengineered multicellularity may offer efficiency improvements difficult to achieve with physical engineering approaches.

BACKGROUND:The configuration of necrotic areas within the retinal pigmented epithelium is an important element in the progression of age-related macular degeneration (AMD). In the exudative (wet) and non-exudative (dry) forms of the disease, retinal pigment epithelial (RPE) cells respond to adjacent atrophied regions by secreting vascular endothelial growth factor (VEGF) that in turn recruits new blood vessels which lead to a further reduction in retinal function and vision. In vitro models exist for studying VEGF expression in wet AMD (Vargis et al., Biomaterials 35(13):3999-4004, 2014), but are limited in the patterns of necrotic and intact RPE epithelium they can produce and in their ability to finely resolve VEGF expression dynamics.RESULTS:In this work, an in silico hybrid agent-based model was developed and validated using the results of this cell culture model of VEGF expression in AMD. The computational model was used to extend the cell culture investigation to explore the dynamics of VEGF expression in different sized patches of RPE cells and the role of negative feedback in VEGF expression. Results of the simulation and the cell culture studies were in excellent qualitative agreement, and close quantitative agreement.CONCLUSIONS:The model indicated that the configuration of necrotic and RPE cell-containing regions have a major impact on VEGF expression dynamics and made precise predictions of VEGF expression dynamics by groups of RPE cells of various sizes and configurations. Coupled with biological studies, this model may give insights into key molecular mechanisms of AMD progression and open routes to more effective treatments.

Studies have shown that genetic regulatory networks (GRNs) consist of modules that are densely connected subnetworks that function quasi-autonomously. Modules may be recognized motifs that comprise of two or three genes with particular regulatory functions and connectivity or be purely structural and identified through connection density. It is unclear what evolutionary and developmental advantages modular structure and in particular motifs provide that have led to this enrichment. This study seeks to understand how modules within developmental GRNs influence the complexity of multicellular patterns that emerge from the dynamics of the regulatory networks. We apply an algorithmic complexity to measure the organization of the patterns. A computational study was performed by creating Boolean intracellular networks within a simulated epithelial field of embryonic cells, where each cell contains the same network and communicates with adjacent cells using contact-mediated signaling. Intracellular networks with random connectivity were compared to those with modular connectivity and with motifs. Results show that modularity effects network dynamics and pattern organization significantly. In particular: (1) modular connectivity alone increases complexity in network dynamics and patterns; (2) bistable switch motifs simplify both the pattern and network dynamics; (3) all other motifs with feedback loops increase multicellular pattern complexity while simplifying the network dynamics; (4) negative feedback loops affect the dynamics complexity more significantly than positive feedback loops.

Motifs are small gene interaction networks that frequently occur within larger genetic regulatory networks (GRNs). However, it is unclear what evolutionary and developmental advantages motifs provide that have led to this enrichment. This study seeks to understand how motifs within developmental GRNs influence the complexity of multicellular patterns that emerge from the dynamics of the regulatory networks. A computational study was performed by creating Boolean intracellular networks with varying inserted motifs within a simulated epithelial field of embryonic cells. Each cell contains the same network and communicates with adjacent cells using contact-mediated signaling. Comparison of random networks to those with motifs demonstrated that: (1) Bistable switches that encode mutual inhibition simplify both the pattern and network dynamics. (2) All other motifs with feedback loops increase information complexity of the multicellular patterns while simplifying the network dynamics. (3) Negative feedback loops affect the dynamics complexity more significantly than positive feedback loops. (4) Feed forward motifs without feedback have little effect on the complexity of patterns formed.

Multiscale models that link sub-cellular, cellular and multicellular components offer powerful insights in disease development. Such models need a realistic set of parameters to represent the physical and chemical mechanisms at the subcellular and cellular levels to produce high fidelity multicellular outcomes. However, determining correct values for some of the parameters is often difficult and expensive using high-throughput microfluidic approaches. This work presents an alternative approach that estimates cellular parameters from spatiotemporal data produced from bioengineered multicellular in vitro experiments. Specifically, we apply a search technique to an integrated cellular and multicellular model of retinal pigment epithelial (RPE) cells to estimate the binding rate and auto-regulation rate of vascular endothelial growth factor (VEGF). Understanding VEGF regulation is critical in treating age-related macular degeneration and many other diseases. The method successfully identifies realistic values for autoregulatory cellular parameters that reproduce the spatiotemporal in vitro experimental data.

MOTIVATIONBiological system behaviors are often the outcome of complex interactions among a large number of cells and their biotic and abiotic environment. Computational biologists attempt to understand, predict and manipulate biological system behavior through mathematical modeling and computer simulation. Discrete agent-based modeling (in combination with high-resolution grids to model the extracellular environment) is a popular approach for building biological system models. However, the computational complexity of this approach forces computational biologists to resort to coarser resolution approaches to simulate large biological systems. High-performance parallel computers have the potential to address the computing challenge, but writing efficient software for parallel computers is difficult and time-consuming.RESULTSWe have developed Biocellion, a high-performance software framework, to solve this computing challenge using parallel computers. To support a wide range of multicellular biological system models, Biocellion asks users to provide their model specifics by filling the function body of pre-defined model routines. Using Biocellion, modelers without parallel computing expertise can efficiently exploit parallel computers with less effort than writing sequential programs from scratch. We simulate cell sorting, microbial patterning and a bacterial system in soil aggregate as case studies.AVAILABILITY AND IMPLEMENTATIONBiocellion runs on x86 compatible systems with the 64 bit Linux operating system and is freely available for academic use. Visit http://biocellion.com for additional information.