Deep Artificial Neural Networks and Neuromorphic Chips for Big Data Analysis: Pharmaceutical and Bioinformatics Applications

Machine learning models for prediction of lymph node metastasis in patients with gastric cancer: a Chinese single-centre study with external validation in an Asian American population

OBJECTIVE

To develop and validate machine learning (ML)-based models to predict lymph node metastasis (LNM) in patients with gastric cancer (GC).

DESIGN

Retrospective cohort study.

SETTING

Second Affiliated Hospital of Soochow University.

PARTICIPANTS

A total of 500 inpatients from the Second Affiliated Hospital of Soochow University, collected retrospectively between 1 April 2018 and 31 March 2023, were used as the training set, while 824 Asian patients from the Surveillance, Epidemiology and End Results database comprised the external validation set.

MAIN OUTCOME MEASURES

Prediction models were developed using multiple ML algorithms, including logistic regression, support vector machine, k-nearest neighbours, naive Bayes, decision tree (DT), gradient boosting DT, random forest and artificial neural network (ANN). The predictive value of these models was validated and evaluated through receiver operating characteristic curves, precision-recall (PR) curves, calibration curves, decision curve analysis and accuracy metrics.

RESULTS

Among the ML algorithms, the ANN outperformed others, achieving the highest accuracy (0.722; 95% CI: 0.692 to 0.751), precision (0.732; 95% CI: 0.694 to 0.776), F1 score (0.733; 95% CI: 0.695 to 0.773), specificity (0.728; 95% CI: 0.684 to 0.770) and area under the PR curve (0.781; 95% CI: 0.740 to 0.821) in the external validation results. Moreover, it demonstrated superior calibration and clinical utility. Shapley Additive Explanations analysis identified the depth of invasion, tumour size and Lauren classification as the most influential predictors of LNM in patients with GC. Furthermore, a user-friendly web application was developed to provide individual prediction results.

CONCLUSIONS

This study introduces an accurate, reliable and clinically applicable approach for predicting the risk of LNM in patients with GC. The model demonstrates its potential to enhance the personalised management of GC in diverse populations, supported by external validation and an accessible web application for practical use.

Spiking Neuron-Astrocyte Networks for Image Recognition

From biological and artificial network perspectives, researchers have started acknowledging astrocytes as computational units mediating neural processes. Here, we propose a novel biologically inspired neuron-astrocyte network model for image recognition, one of the first attempts at implementing astrocytes in spiking neuron networks (SNNs) using a standard data set. The architecture for image recognition has three primary units: the preprocessing unit for converting the image pixels into spiking patterns, the neuron-astrocyte network forming bipartite (neural connections) and tripartite synapses (neural and astrocytic connections), and the classifier unit. In the astrocyte-mediated SNNs, an astrocyte integrates neural signals following the simplified Postnov model. It then modulates the integrate-and-fire (IF) neurons via gliotransmission, thereby strengthening the synaptic connections of the neurons within the astrocytic territory. We develop an architecture derived from a baseline SNN model for unsupervised digit classification. The spiking neuron-astrocyte networks (SNANs) display better network performance with an optimal variance-bias trade-off than SNN alone. We demonstrate that astrocytes promote faster learning, support memory formation and recognition, and provide a simplified network architecture. Our proposed SNAN can serve as a benchmark for future researchers on astrocyte implementation in artificial networks, particularly in neuromorphic systems, for its simplified design.

Free-space optical spiking neural network

Neuromorphic engineering has emerged as a promising avenue for developing brain-inspired computational systems. However, conventional electronic AI-based processors often encounter challenges related to processing speed and thermal dissipation. As an alternative, optical implementations of such processors have been proposed, capitalizing on the intrinsic information-processing capabilities of light. Among the various Optical Neural Networks (ONNs) explored within the realm of optical neuromorphic engineering, Spiking Neural Networks (SNNs) have exhibited notable success in emulating the computational principles of the human brain. The event-based spiking nature of optical SNNs offers capabilities in low-power operation, speed, temporal processing, analog computing, and hardware efficiency that are difficult or impossible to match with other ONN types. In this work, we introduce the pioneering Free-space Optical Deep Spiking Convolutional Neural Network (OSCNN), a novel approach inspired by the computational model of the human eye. Our OSCNN leverages free-space optics to enhance power efficiency and processing speed while maintaining high accuracy in pattern detection. Specifically, our model employs Gabor filters in the initial layer for effective feature extraction, and utilizes optical components such as Intensity-to-Delay conversion and a synchronizer, designed using readily available optical components. The OSCNN was rigorously tested on benchmark datasets, including MNIST, ETH80, and Caltech, demonstrating competitive classification accuracy. Our comparative analysis reveals that the OSCNN consumes only 1.6 W of power with a processing speed of 2.44 ms, significantly outperforming conventional electronic CNNs on GPUs, which typically consume 150-300 W with processing speeds of 1-5 ms, and competing favorably with other free-space ONNs. Our contributions include addressing several key challenges in optical neural network implementation. To ensure nanometer-scale precision in component alignment, we propose advanced micro-positioning systems and active feedback control mechanisms. To enhance signal integrity, we employ high-quality optical components, error correction algorithms, adaptive optics, and noise-resistant coding schemes. The integration of optical and electronic components is optimized through the design of high-speed opto-electronic converters, custom integrated circuits, and advanced packaging techniques. Moreover, we utilize highly efficient, compact semiconductor laser diodes and develop novel cooling strategies to minimize power consumption and footprint.

Potential and challenges of computing with molecular materials

We are at an inflection point in computing where traditional technologies are incapable of keeping up with the demands of exploding data collection and artificial intelligence. This challenge demands a leap to a new platform as transformative as the digital silicon revolution. Over the past 30 years molecular materials for computing have generated great excitement but continually fallen short of performance and reliability requirements. However, recent reports indicate that those historical limitations may have been resolved. Here we assess the current state of computing with molecular-based materials, especially using transition metal complexes of redox active ligands, in the context of neuromorphic computing. We describe two complementary research paths necessary to determine whether molecular materials can be the basis of a new computing technology: continued exploration of the molecular electronic properties that enable computation and, equally important, the process development for on-chip integration of molecular materials.

Evaluating and Reducing Subgroup Disparity in AI Models: An Analysis of Pediatric COVID-19 Test Outcomes

Artificial Intelligence (AI) fairness in healthcare settings has attracted significant attention due to the concerns to propagate existing health disparities. Despite ongoing research, the frequency and extent of subgroup fairness have not been sufficiently studied. In this study, we extracted a nationally representative pediatric dataset (ages 0-17, n=9,935) from the US National Health Interview Survey (NHIS) concerning COVID-19 test outcomes. For subgroup disparity assessment, we trained 50 models using five machine learning algorithms. We assessed the models' area under the curve (AUC) on 12 small (<15% of the total n) subgroups defined using social economic factors versus the on the overall population. Our results show that subgroup disparities were prevalent (50.7%) in the models. Subgroup AUCs were generally lower, with a mean difference of 0.01, ranging from -0.29 to +0.41. Notably, the disparities were not always statistically significant, with four out of 12 subgroups having statistically significant disparities across models. Additionally, we explored the efficacy of synthetic data in mitigating identified disparities. The introduction of synthetic data enhanced subgroup disparity in 57.7% of the models. The mean AUC disparities for models with synthetic data decreased on average by 0.03 via resampling and 0.04 via generative adverbial network methods.

Deep learning algorithms applied to computational chemistry

Recently, there has been a significant increase in the use of deep learning techniques in the molecular sciences, which have shown high performance on datasets and the ability to generalize across data. However, no model has achieved perfect performance in solving all problems, and the pros and cons of each approach remain unclear to those new to the field. Therefore, this paper aims to review deep learning algorithms that have been applied to solve molecular challenges in computational chemistry. We proposed a comprehensive categorization that encompasses two primary approaches; conventional deep learning and geometric deep learning models. This classification takes into account the distinct techniques employed by the algorithms within each approach. We present an up-to-date analysis of these algorithms, emphasizing their key features and open issues. This includes details of input descriptors, datasets used, open-source code availability, task solutions, and actual research applications, focusing on general applications rather than specific ones such as drug discovery. Furthermore, our report discusses trends and future directions in molecular algorithm design, including the input descriptors used for each deep learning model, GPU usage, training and forward processing time, model parameters, the most commonly used datasets, libraries, and optimization schemes. This information aids in identifying the most suitable algorithms for a given task. It also serves as a reference for the datasets and input data frequently used for each algorithm technique. In addition, it provides insights into the benefits and open issues of each technique, and supports the development of novel computational chemistry systems.

Self-learning activation functions to increase accuracy of privacy-preserving Convolutional Neural Networks with homomorphic encryption

The widespread adoption of cloud computing necessitates privacy-preserving techniques that allow information to be processed without disclosure. This paper proposes a method to increase the accuracy and performance of privacy-preserving Convolutional Neural Networks with Homomorphic Encryption (CNN-HE) by Self-Learning Activation Functions (SLAF). SLAFs are polynomials with trainable coefficients updated during training, together with synaptic weights, for each polynomial independently to learn task-specific and CNN-specific features. We theoretically prove its feasibility to approximate any continuous activation function to the desired error as a function of the SLAF degree. Two CNN-HE models are proposed: CNN-HE-SLAF and CNN-HE-SLAF-R. In the first model, all activation functions are replaced by SLAFs, and CNN is trained to find weights and coefficients. In the second one, CNN is trained with the original activation, then weights are fixed, activation is substituted by SLAF, and CNN is shortly re-trained to adapt SLAF coefficients. We show that such self-learning can achieve the same accuracy 99.38% as a non-polynomial ReLU over non-homomorphic CNNs and lead to an increase in accuracy (99.21%) and higher performance (6.26 times faster) than the state-of-the-art CNN-HE CryptoNets on the MNIST optical character recognition benchmark dataset.

Improving Psychiatry Services with Artificial Intelligence: Opportunities and Challenges

Mental disorders are a critical global public health problem due to their increasing prevalence, rising costs, and significant economic burden. Despite efforts to increase the mental health workforce in Türkiye, there is a significant shortage of psychiatrists, limiting the quality and accessibility of mental health services. This review examines the potential of artificial intelligence (AI), especially large language models, to transform psychiatric care in the world and in Türkiye. AI technologies, including machine learning and deep learning, offer innovative solutions for the diagnosis, personalization of treatment, and monitoring of mental disorders using a variety of data sources, such as speech patterns, neuroimaging, and behavioral measures. Although AI has shown promising capabilities in improving diagnostic accuracy and access to mental health services, challenges such as algorithmic biases, data privacy concerns, ethical implications, and the confabulation phenomenon of large language models prevent the full implementation of AI in practice. The review highlights the need for interdisciplinary collaboration to develop culturally and linguistically adapted AI tools, particularly in the Turkish context, and suggests strategies such as fine-tuning, retrieval-augmented generation, and reinforcement learning from human feedback to increase AI reliability. Advances suggest that AI can improve mental health care by increasing diagnostic accuracy and accessibility while preserving the essential human elements of medical care. Current limitations need to be addressed through rigorous research and ethical frameworks for effective and equitable integration of AI into mental health care. Keywords: Artificial İntelligence, Health, Large Language Model, Machine Learning, Psychiatry.

Deep learning in bioinformatics

Deep learning is a powerful machine learning technique that can learn from large amounts of data using multiple layers of artificial neural networks. This paper reviews some applications of deep learning in bioinformatics, a field that deals with analyzing and interpreting biological data. We first introduce the basic concepts of deep learning and then survey the recent advances and challenges of applying deep learning to various bioinformatics problems, such as genome sequencing, gene expression analysis, protein structure prediction, drug discovery, and disease diagnosis. We also discuss future directions and opportunities for deep learning in bioinformatics. We aim to provide an overview of deep learning so that bioinformaticians applying deep learning models can consider all critical technical and ethical aspects. Thus, our target audience is biomedical informatics researchers who use deep learning models for inference. This review will inspire more bioinformatics researchers to adopt deep-learning methods for their research questions while considering fairness, potential biases, explainability, and accountability.

Age-Associated UBE2O Reduction Promotes Neuronal Death in Alzheimer's Disease

BACKGROUND

Alzheimer's disease (AD) is the most common neurodegenerative disease leading to dementia in the elderly. Ubiquitin proteasome system (UPS) is critical for protein homeostasis, while the functional decline of UPS with age contributes to the pathogenesis of AD. Ubiquitin-conjugating enzyme E2O (UBE2O), an E2-E3 hybrid enzyme, is a major component of UPS. However, its role in AD pathogenesis has not been fully defined.

OBJECTIVE

We aimed to identify the age-associated expression of UBE2O and its role AD pathogenesis.

METHODS

Western blot analysis were used to assess expression of UBE2O in organs/tissues and cell lines. Immunofluorescence staining was performed to examine the cellular distribution of UBE2O. Neuronal death was determined by the activity of lactate dehydrogenase.

RESULTS

UBE2O is highly expressed in the cortex and hippocampus. It is predominantly expressed in neurons but not in glial cells. The peak expression of UBE2O is at postnatal day 17 and 14 in the cortex and hippocampus, respectively. Moreover its expression is gradually reduced with age. Importantly, UBE2O is significantly reduced in both cortex and hippocampus of AD mice. Consistently, overexpression of amyloid-β protein precursor (AβPP) with a pathogenic mutation (AβPPswe) for AD reduces the expression of UBE2O and promotes neuronal death, while increased expression of UBE2O rescues AβPPswe-induced neuronal death.

CONCLUSION

Our study indicates that age-associated reduction of UBE2O may facilitates neuronal death in AD, while increasing UBE2O expression or activity may be a potential approach for AD treatment by inhibiting neuronal death.

Contribution of Deep Learning in the Investigation of Possible Dual LOX-3 Inhibitors/DPPH Scavengers: The Case of Recently Synthesized Compounds

Even though non-steroidal anti-inflammatory drugs are the most effective treatment for inflammatory conditions, they have been linked to negative side effects. A promising approach to mitigating potential risks, is the development of new compounds able to combine anti-inflammatory with antioxidant activity to enhance activity and reduce toxicity. The implication of reactive oxygen species in inflammatory conditions has been extensively studied, based on the pro-inflammatory properties of generated free radicals. Drugs with dual activity (i.e., inhibiting inflammation related enzymes, e.g., LOX-3 and scavenging free radicals, e.g., DPPH) could find various therapeutic applications, such as in cardiovascular or neurodegenerating disorders. The challenge we embarked on using deep learning was the creation of appropriate classification and regression models to discriminate pharmacological activity and selectivity as well as to discover future compounds with dual activity prior to synthesis. An accurate filter algorithm was established, based on knowledge from compounds already evaluated in vitro, that can separate compounds with low, moderate or high activity. In this study, we constructed a customized highly effective one dimensional convolutional neural network (CONV1D), with accuracy scores up to 95.2%, that was able to identify dual active compounds, being LOX-3 inhibitors and DPPH scavengers, as an indication of simultaneous anti-inflammatory and antioxidant activity. Additionally, we created a highly accurate regression model that predicted the exact value of effectiveness of a set of recently synthesized compounds with anti-inflammatory activity, scoring a root mean square error value of 0.8. Eventually, we succeeded in observing the manner in which those newly synthesized compounds differentiate from each other, regarding a specific pharmacological target, using deep learning algorithms.

A High-Accuracy and Energy-Efficient CORDIC Based Izhikevich Neuron With Error Suppression and Compensation

Bio-inspired neuron models are the key building blocks of brain-like neural networks for brain-science exploration and neuromorphic engineering applications. The efficient hardware design of bio-inspired neuron models is one of the challenges to implement brain-like neural networks, as the balancing of model accuracy, energy consumption and hardware cost is very challenging. This paper proposes a high-accuracy and energy-efficient Fast-Convergence COordinate Rotation DIgital Computer (FC-CORDIC) based Izhikevich neuron design. For ensuring the model accuracy, an error propagation model of the Izhikevich neuron is presented for systematic error analysis and effective error reduction. Parameter-Tuning Error Compensation (PTEC) method and Bitwidth-Extension Error Suppression (BEES) method are proposed to reduce the error of Izhikevich neuron design effectively. In addition, by utilizing the FC-CORDIC instead of conventional CORDIC for square calculation in the Izhikevich model, the redundant CORDIC iterations are removed and therefore, both the accumulated errors and required computation are effectively reduced, which significantly improve the accuracy and energy efficiency. An optimized fixed-point design of FC-CORDIC is also proposed to save hardware overhead while ensuring the accuracy. FPGA implementation results exhibit that the proposed Izhikevich neuron design can achieve high accuracy and energy efficiency with an acceptable hardware overhead, among the state-of-the-art designs.

3D Neural Network Composed of Neurospheroid and Bionanohybrid on Microelectrode Array to Realize the Spatial Input Signal Recognition in Neurospheroid

There have been several studies for demonstration of 2D neural network using living cells or organic/inorganic molecules, but to date, there is no report of development of a 3D neural network in vitro. Based on developed bionanohybrid composed of protein, DNA, molybdenum disulfide nanoparticles, and peptides for controlling electrophysiological states of living cells, here, the in vitro 3D neural network composed of the bionanohybrid, 3D neurospheroid and the microelectrode array (MEA) is developed. After production of the 3D neurospheroid derived from human neural stem cells, the bionanohybrid developed on the MEA successfully semi-penetrates the neurites of the 3D neurospheroid and forms the 3D neural network. The developed 3D neural network successfully exhibited the electrophysiological output signals of the 3D neurospheroid by transmitting the input signal applied by the bionanohybrid. Moreover, by using the selectively immobilized bionanohybrid on the MEA, the spatial input signal recognition in the neurospheroid of 3D neural network is realized for the first time. This newly developed in vitro 3D neural network provides a promising strategy to be applied in brain-on-a-chip, brain disease-related drug efficacy evaluation, bioelectronics, and bioelectronic medicine.

A Saliency based Feature Fusion Model for EEG Emotion Estimation

Among the different modalities to assess emotion, electroencephalogram (EEG), representing the electrical brain activity, achieved motivating results over the last decade. Emotion estimation from EEG could help in the diagnosis or rehabilitation of certain diseases. In this paper, we propose a dual model considering two different representations of EEG feature maps: 1) a sequential based representation of EEG band power, 2) an image-based representation of the feature vectors. We also propose an innovative method to combine the information based on a saliency analysis of the image-based model to promote joint learning of both model parts. The model has been evaluated on four publicly available datasets: SEED-IV, SEED, DEAP and MPED. The achieved results outperform results from state-of-the-art approaches for three of the proposed datasets with a lower standard deviation that reflects higher stability. For sake of reproducibility, the codes and models proposed in this paper are available at https://github.com/VDelv/Emotion-EEG.

Modulating the Filamentary-Based Resistive Switching Properties of HfO2 Memristive Devices by Adding Al2O3 Layers

The resistive switching properties of HfO2 based 1T-1R memristive devices are electrically modified by adding ultra-thin layers of Al2O3 into the memristive device. Three different types of memristive stacks are fabricated in the 130 nm CMOS technology of IHP. The switching properties of the memristive devices are discussed with respect to forming voltages, low resistance state and high resistance state characteristics and their variabilities. The experimental I–V characteristics of set and reset operations are evaluated by using the quantum point contact model. The properties of the conduction filament in the on and off states of the memristive devices are discussed with respect to the model parameters obtained from the QPC fit.

Detection and analysis of COVID-19 in medical images using deep learning techniques

The main purpose of this work is to investigate and compare several deep learning enhanced techniques applied to X-ray and CT-scan medical images for the detection of COVID-19. In this paper, we used four powerful pre-trained CNN models, VGG16, DenseNet121, ResNet50,and ResNet152, for the COVID-19 CT-scan binary classification task. The proposed Fast.AI ResNet framework was designed to find out the best architecture, pre-processing, and training parameters for the models largely automatically. The accuracy and F1-score were both above 96% in the diagnosis of COVID-19 using CT-scan images. In addition, we applied transfer learning techniques to overcome the insufficient data and to improve the training time. The binary and multi-class classification of X-ray images tasks were performed by utilizing enhanced VGG16 deep transfer learning architecture. High accuracy of 99% was achieved by enhanced VGG16 in the detection of X-ray images from COVID-19 and pneumonia. The accuracy and validity of the algorithms were assessed on X-ray and CT-scan well-known public datasets. The proposed methods have better results for COVID-19 diagnosis than other related in literature. In our opinion, our work can help virologists and radiologists to make a better and faster diagnosis in the struggle against the outbreak of COVID-19.

The rise of deep learning and transformations in bioactivity prediction power of molecular modeling tools

The search and design for the better use of bioactive compounds are used in many experiments to best mimic compounds' functions in the human body. However, finding a cost-effective and timesaving approach is a top priority in different disciplines. Nowadays, artificial intelligence (AI) and particularly deep learning (DL) methods are widely applied to improve the precision and accuracy of models used in the drug discovery process. DL approaches have been used to provide more opportunities for a faster, efficient, cost-effective, and reliable computer-aided drug discovery. Moreover, the increasing biomedical data volume in areas, like genome sequences, medical images, protein structures, etc., has made data mining algorithms very important in finding novel compounds that could be drugs, uncovering or repurposing drugs and improving the area of genetic markers-based personalized medicine. Furthermore, deep neural networks (DNNs) have been demonstrated to outperform other techniques such as random forests and SVMs for QSAR studies and ligand-based virtual screening. Despite this, in QSAR studies, the quality of different data sources and potential experimental errors has greatly affected the accuracy of QSAR predictions. Therefore, further researches are still needed to improve the accuracy, selectivity, and sensitivity of the DL approach in building the best models of drug discovery.

The Feasibility of Using Machine Learning to Classify Calls to South African Emergency Dispatch Centres According to Prehospital Diagnosis, by Utilising Caller Descriptions of the Incident

This paper presents the application of machine learning for classifying time-critical conditions namely sepsis, myocardial infarction and cardiac arrest, based off transcriptions of emergency calls from emergency services dispatch centers in South Africa. In this study we present results from the application of four multi-class classification algorithms: Support Vector Machine (SVM), Logistic Regression, Random Forest and K-Nearest Neighbor (kNN). The application of machine learning for classifying time-critical diseases may allow for earlier identification, adequate telephonic triage, and quicker response times of the appropriate cadre of emergency care personnel. The data set consisted of an original data set of 93 examples which was further expanded through the use of data augmentation. Two feature extraction techniques were investigated namely; TF-IDF and handcrafted features. The results were further improved using hyper-parameter tuning and feature selection. In our work, within the limitations of a limited data set, classification results yielded an accuracy of up to 100% when training with 10-fold cross validation, and 95% accuracy when predicted on unseen data. The results are encouraging and show that automated diagnosis based on emergency dispatch centre transcriptions is feasible. When implemented in real time, this can have multiple utilities, e.g. enabling the call-takers to take the right action with the right priority.

Understanding Demographic Risk Factors for Adverse Outcomes in COVID-19 Patients: Explanation of a Deep Learning Model

This study was to understand the impacts of three key demographic variables, age, gender, and race, on the adverse outcome of all-cause hospitalization or all-cause mortality in patients with COVID-19, using a deep neural network (DNN) analysis. We created a cohort of Veterans who were tested positive for COVID-19, extracted data on age, gender, and race, and clinical characteristics from their electronic health records, and trained a DNN model for predicting the adverse outcome. Then, we analyzed the association of the demographic variables with the risks of the adverse outcome using the impact scores and interaction scores for explaining DNN models. The results showed that, on average, older age and African American race were associated with higher risks while female gender was associated with lower risks. However, individual-level impact scores of age showed that age was a more impactful risk factor in younger patients and in older patients with fewer comorbidities. The individual-level impact scores of gender and race variables had a wide span covering both positive and negative values. The interaction scores between the demographic variables showed that the interaction effects were minimal compared to the impact scores associated with them. In conclusion, the DNN model is able to capture the non-linear relationship between the risk factors and the adverse outcome, and the impact scores and interaction scores can help explain the complicated non-linear effects between the demographic variables and the risk of the outcome.

Shedding Light on the Black Box: Explaining Deep Neural Network Prediction of Clinical Outcomes

Deep neural network models are emerging as an important method in healthcare delivery, following the recent success in other domains such as image recognition. Due to the multiple non-linear inner transformations, deep neural networks are viewed by many as black boxes. For practical use, deep learning models require explanations that are intuitive to clinicians. In this study, we developed a deep neural network model to predict outcomes following major cardiovascular procedures, using temporal image representation of past medical history as input. We created a novel explanation for the prediction of the model by defining impact scores that associate clinical observations with the outcome. For comparison, a logistic regression model was fitted to the same dataset. We compared the impact scores and log odds ratios by calculating three types of correlations, which provided a partial validation of the impact scores. The deep neural network model achieved an area under the receiver operating characteristics curve (AUC) of 0.787, compared to 0.746 for the logistic regression model. Moderate correlations were found between the impact scores and the log odds ratios. Impact scores generated by the explanation algorithm has the potential to shed light on the "black box" deep neural network model and could facilitate its adoption by clinicians.

AI-based investigation of molecular biomarkers of longevity

In this paper, I build deep neural networks of various structures and hyperparameters in order to predict human chronological age based on open-access biochemical indicators and their specifications from the NHANES database. In total, 1152 neural networks are trained and tested. The algorithms are trained and tested on incomplete data: missing values in data records are extrapolated by mean or median values for each parameter. I select the best neural networks in terms of validation accuracy (coefficient of determination and mean absolute error). It turns out that the most accurate results are delivered by multilayer networks (6 layers) with recurrent layers. Neural network types are selected by trial and error. The algorithms reached an accuracy of 78% in terms of coefficient of determination and 6.5 in terms of mean absolute error. I also list empirically determined features of neural networks that increase accuracy for the task of chronological age prediction. Obtained results can be considered as an approximation of human biological age. Parameters in training datasets are selected the most broadly: all potentially relevant parameters (926) from the NHANES database are used. Although the networks are trained on the incomplete data, they demonstrated the ability to make reasonable predictions (with R² > 0.7) based on no more than 100 biochemical indicators. Hence, for practical reasons the full data on each of 926 indicators are not required, although the analysis of the impact of each indicator is useful for theoretical developments.

Semantic Segmentation of the Choroid in Swept Source Optical Coherence Tomography Images for Volumetrics

The choroid is a complex vascular tissue that is covered with the retinal pigment epithelium. Ultra high speed swept source optical coherence tomography (SS-OCT) provides us with high-resolution cube scan images of the choroid. Robust segmentation techniques are required to reconstruct choroidal volume using SS-OCT images. For automated segmentation, the delineation of the choroidal-scleral (C-S) boundary is key to accurate segmentation. Low contrast of the boundary, scleral canals formed by the vessel and the nerve, and the posterior stromal layer, may cause segmentation errors. Semantic segmentation is one of the applications of deep learning used to classify the parts of images related to the meanings of the subjects. We applied semantic segmentation to choroidal segmentation and measured the volume of the choroid. The measurement results were validated through comparison with those of other segmentation methods. As a result, semantic segmentation was able to segment the C-S boundary and choroidal volume adequately.

Deep learning in omics: a survey and guideline

Omics, such as genomics, transcriptome and proteomics, has been affected by the era of big data. A huge amount of high dimensional and complex structured data has made it no longer applicable for conventional machine learning algorithms. Fortunately, deep learning technology can contribute toward resolving these challenges. There is evidence that deep learning can handle omics data well and resolve omics problems. This survey aims to provide an entry-level guideline for researchers, to understand and use deep learning in order to solve omics problems. We first introduce several deep learning models and then discuss several research areas which have combined omics and deep learning in recent years. In addition, we summarize the general steps involved in using deep learning which have not yet been systematically discussed in the existent literature on this topic. Finally, we compare the features and performance of current mainstream open source deep learning frameworks and present the opportunities and challenges involved in deep learning. This survey will be a good starting point and guideline for omics researchers to understand deep learning.

Society of Toxicologic Pathology Digital Pathology and Image Analysis Special Interest Group Article*: Opinion on the Application of Artificial Intelligence and Machine Learning to Digital Toxicologic Pathology

Toxicologic pathology is transitioning from analog to digital methods. This transition seems inevitable due to a host of ongoing social and medical technological forces. Of these, artificial intelligence (AI) and in particular machine learning (ML) are globally disruptive, rapidly growing sectors of technology whose impact on the long-established field of histopathology is quickly being realized. The development of increasing numbers of algorithms, peering ever deeper into the histopathological space, has demonstrated to the scientific community that AI pathology platforms are now poised to truly impact the future of precision and personalized medicine. However, as with all great technological advances, there are implementation and adoption challenges. This review aims to define common and relevant AI and ML terminology, describe data generation and interpretation, outline current and potential future business cases, discuss validation and regulatory hurdles, and most importantly, propose how overcoming the challenges of this burgeoning technology may shape toxicologic pathology for years to come, enabling pathologists to contribute even more effectively to answering scientific questions and solving global health issues. [Box: see text]

Analysis of parameters affecting blood oxygen saturation and modeling of fuzzy logic system for inspired oxygen prediction

BACKGROUND AND OBJECTIVE

Fraction of Inspired Oxygen is one of the arbitrary set ventilator parameters which has critical influence on the concentration of blood oxygen. Normally mechanical ventilators providing respiratory assistance are tuned manually to supply required inspired oxygen to keep the oxygen saturation at the desired level. Maintaining oxygen saturation in the desired limit is so vital since excess supply of inspired oxygen leads to hypercapnia and respiratory acidosis which lead to increased risk in cell damage and death. On the other side a sudden drop in oxygen saturation will lead to severe cardiac arrest and seizure. Hence intelligent real time control of blood oxygen level saturation is highly significant for patients in intensive care units.

METHODS

This paper gives statistical pair wise analysis for finding out deeply correlated physiological parameters from clinical data for fixing fuzzy variables. An advisory fuzzy controller using Mamdani model is developed with R programming to predict FiO₂ which is to be delivered from the ventilator to maintain SaO₂ with in required levels.

RESULTS

Fuzzy variables for the fuzzy model is fixed using 75% of the clinical data collected. Remaining 25% of the data is used for checking the system. Compared the predictive output of the system with physicians' decisions and found to be accurate with less than five percentage error.

CONCLUSIONS

Based on the comparison the system is proved to be effective and can be used as assist mode for physicians for effective decision making.

Virtual Screening Meets Deep Learning

BACKGROUND

Automated compound testing is currently the de facto standard method for drug screening, but it has not brought the great increase in the number of new drugs that was expected. Computer- aided compounds search, known as Virtual Screening, has shown the benefits to this field as a complement or even alternative to the robotic drug discovery. There are different methods and approaches to address this problem and most of them are often included in one of the main screening strategies. Machine learning, however, has established itself as a virtual screening methodology in its own right and it may grow in popularity with the new trends on artificial intelligence.

OBJECTIVE

This paper will attempt to provide a comprehensive and structured review that collects the most important proposals made so far in this area of research. Particular attention is given to some recent developments carried out in the machine learning field: the deep learning approach, which is pointed out as a future key player in the virtual screening landscape.

Optimization of a Deep-Learning Method Based on the Classification of Images Generated by Parameterized Deep Snap a Novel Molecular-Image-Input Technique for Quantitative Structure-Activity Relationship (QSAR) Analysis

Numerous chemical compounds are distributed around the world and may affect the homeostasis of the endocrine system by disrupting the normal functions of hormone receptors. Although the risks associated with these compounds have been evaluated by acute toxicity testing in mammalian models, the chronic toxicity of many chemicals remains due to high cost of the compounds and the testing, etc. However, computational approaches may be promising alternatives and reduce these evaluations. Recently, deep learning (DL) has been shown to be promising prediction models with high accuracy for recognition of images, speech, signals, and videos since it greatly benefits from large datasets. Recently, a novel DL-based technique called DeepSnap was developed to conduct QSAR analysis using three-dimensional images of chemical structures. It can be used to predict the potential toxicity of many different chemicals to various receptors without extraction of descriptors. DeepSnap has been shown to have a very high capacity in tests using Tox21 quantitative qHTP datasets. Numerous parameters must be adjusted to use the DeepSnap method but they have not been optimized. In this study, the effects of these parameters on the performance of the DL prediction model were evaluated in terms of the loss in validation as an indicator for evaluating the performance of the DL using the toxicity information in the Tox21 qHTP database. The relations of the parameters of DeepSnap such as (1) number of molecules per SDF split into (2) zoom factor percentage, (3) atom size for van der waals percentage, (4) bond radius, (5) minimum bond distance, and (6) bond tolerance, with the validation loss following quadratic function curves, which suggests that optimal thresholds exist to attain the best performance with these prediction models. Using the parameter values set with the best performance, the prediction model of chemical compounds for CAR agonist was built using 64 images, at 105° angle, with AUC of 0.791. Thus, based on these parameters, the proposed DeepSnap-DL approach will be highly reliable and beneficial to establish models to assess the risk associated with various chemicals.

In Silico Prediction of Chemical-Induced Hepatocellular Hypertrophy Using Molecular Descriptors

In silico prediction for toxicity of chemicals is required to reduce cost, time, and animal testing. However, predicting hepatocellular hypertrophy, which often affects the derivation of the No-Observed-Adverse-Effect Level in repeated dose toxicity studies, is difficult because pathological findings are diverse, mechanisms are largely unknown, and a wide variety of chemical structures exists. Therefore, a method for predicting the hepatocellular hypertrophy of diverse chemicals without complete understanding of their mechanisms is necessary. In this study, we developed predictive classification models of hepatocellular hypertrophy using machine learning-specifically, deep learning, random forest, and support vector machine. We extracted hepatocellular hypertrophy data on rats from 2 toxicological databases, our original database developed from risk assessment reports such as pesticides, and the Hazard Evaluation Support System Integrated Platform. Then, we constructed prediction models based on molecular descriptors and evaluated their performance using independent test chemicals datasets, which differed from the training chemicals datasets. Further, we defined the applicability domain (AD), which generally limits the application for chemicals, as structurally similar to the training chemicals dataset. The best model was found to be the support vector machine model using the Hazard Evaluation Support System Integrated Platform dataset, which was trained with 251 chemicals and predicted 214 test chemicals inside the applicability domain. It afforded a prediction accuracy of 0.76, sensitivity of 0.90, and area under the curve of 0.81. These in silico predictive classification models could be reliable tools for hepatocellular hypertrophy assessments and can facilitate the development of in silico models for toxicity prediction.

Using deep neural networks along with dimensionality reduction techniques to assist the diagnosis of neurodegenerative disorders

The analysis of neuroimaging data is frequently used to assist the diagnosis of neurodegenerative disorders such as Alzheimer's disease (AD) or Parkinson's disease (PD) and has become a routine procedure in the clinical practice. During the past decade, the pattern recognition community has proposed a number of machine learning-based systems that automatically analyse neuroimaging data in order to improve the diagnosis. However, the high dimensionality of the data is still a challenge and there is room for improvement. The development of novel classification frameworks as TensorFlow, recently released as open source by Google Inc., represents an opportunity to continue evolving these systems. In this work, we demonstrate several computer-aided diagnosis (CAD) systems based on Deep Neural Networks that improve the diagnosis for AD and PD and outperform those based on classical classifiers. In order to address the small sample size problem we evaluate two dimensionality reduction algorithms based on Principal Component Analysis and Non-Negative Matrix Factorization (NNMF), respectively. The performance of developed CAD systems is assessed using 4 datasets with neuroimaging data of different modalities.

Machine Learning Based Toxicity Prediction: From Chemical Structural Description to Transcriptome Analysis

Toxicity prediction is very important to public health. Among its many applications, toxicity prediction is essential to reduce the cost and labor of a drug's preclinical and clinical trials, because a lot of drug evaluations (cellular, animal, and clinical) can be spared due to the predicted toxicity. In the era of Big Data and artificial intelligence, toxicity prediction can benefit from machine learning, which has been widely used in many fields such as natural language processing, speech recognition, image recognition, computational chemistry, and bioinformatics, with excellent performance. In this article, we review machine learning methods that have been applied to toxicity prediction, including deep learning, random forests, k-nearest neighbors, and support vector machines. We also discuss the input parameter to the machine learning algorithm, especially its shift from chemical structural description only to that combined with human transcriptome data analysis, which can greatly enhance prediction accuracy.

Reinforced Adversarial Neural Computer for de Novo Molecular Design

In silico modeling is a crucial milestone in modern drug design and development. Although computer-aided approaches in this field are well-studied, the application of deep learning methods in this research area is at the beginning. In this work, we present an original deep neural network (DNN) architecture named RANC (Reinforced Adversarial Neural Computer) for the de novo design of novel small-molecule organic structures based on the generative adversarial network (GAN) paradigm and reinforcement learning (RL). As a generator RANC uses a differentiable neural computer (DNC), a category of neural networks, with increased generation capabilities due to the addition of an explicit memory bank, which can mitigate common problems found in adversarial settings. The comparative results have shown that RANC trained on the SMILES string representation of the molecules outperforms its first DNN-based counterpart ORGANIC by several metrics relevant to drug discovery: the number of unique structures, passing medicinal chemistry filters (MCFs), Muegge criteria, and high QED scores. RANC is able to generate structures that match the distributions of the key chemical features/descriptors (e.g., MW, logP, TPSA) and lengths of the SMILES strings in the training data set. Therefore, RANC can be reasonably regarded as a promising starting point to develop novel molecules with activity against different biological targets or pathways. In addition, this approach allows scientists to save time and covers a broad chemical space populated with novel and diverse compounds.

An Efficient ABC_DE_Based Hybrid Algorithm for Protein-Ligand Docking

Protein–ligand docking is a process of searching for the optimal binding conformation between the receptor and the ligand. Automated docking plays an important role in drug design, and an efficient search algorithm is needed to tackle the docking problem. To tackle the protein–ligand docking problem more efficiently, An ABC_DE_based hybrid algorithm (ADHDOCK), integrating artificial bee colony (ABC) algorithm and differential evolution (DE) algorithm, is proposed in the article. ADHDOCK applies an adaptive population partition (APP) mechanism to reasonably allocate the computational resources of the population in each iteration process, which helps the novel method make better use of the advantages of ABC and DE. The experiment tested fifty protein–ligand docking problems to compare the performance of ADHDOCK, ABC, DE, Lamarckian genetic algorithm (LGA), running history information guided genetic algorithm (HIGA), and swarm optimization for highly flexible protein–ligand docking (SODOCK). The results clearly exhibit the capability of ADHDOCK toward finding the lowest energy and the smallest root-mean-square deviation (RMSD) on most of the protein–ligand docking problems with respect to the other five algorithms.

Deep Learning for Drug Design: an Artificial Intelligence Paradigm for Drug Discovery in the Big Data Era

Over the last decade, deep learning (DL) methods have been extremely successful and widely used to develop artificial intelligence (AI) in almost every domain, especially after it achieved its proud record on computational Go. Compared to traditional machine learning (ML) algorithms, DL methods still have a long way to go to achieve recognition in small molecular drug discovery and development. And there is still lots of work to do for the popularization and application of DL for research purpose, e.g., for small molecule drug research and development. In this review, we mainly discussed several most powerful and mainstream architectures, including the convolutional neural network (CNN), recurrent neural network (RNN), and deep auto-encoder networks (DAENs), for supervised learning and nonsupervised learning; summarized most of the representative applications in small molecule drug design; and briefly introduced how DL methods were used in those applications. The discussion for the pros and cons of DL methods as well as the main challenges we need to tackle were also emphasized.

A FPGA-Based, Granularity-Variable Neuromorphic Processor and Its Application in a MIMO Real-Time Control System

Artificial Neural Networks (ANNs), including Deep Neural Networks (DNNs), have become the state-of-the-art methods in machine learning and achieved amazing success in speech recognition, visual object recognition, and many other domains. There are several hardware platforms for developing accelerated implementation of ANN models. Since Field Programmable Gate Array (FPGA) architectures are flexible and can provide high performance per watt of power consumption, they have drawn a number of applications from scientists. In this paper, we propose a FPGA-based, granularity-variable neuromorphic processor (FBGVNP). The traits of FBGVNP can be summarized as granularity variability, scalability, integrated computing, and addressing ability: first, the number of neurons is variable rather than constant in one core; second, the multi-core network scale can be extended in various forms; third, the neuron addressing and computing processes are executed simultaneously. These make the processor more flexible and better suited for different applications. Moreover, a neural network-based controller is mapped to FBGVNP and applied in a multi-input, multi-output, (MIMO) real-time, temperature-sensing and control system. Experiments validate the effectiveness of the neuromorphic processor. The FBGVNP provides a new scheme for building ANNs, which is flexible, highly energy-efficient, and can be applied in many areas.

From in silico astrocyte cell models to neuron-astrocyte network models: A review

The idea that astrocytes may be active partners in synaptic information processing has recently emerged from abundant experimental reports. Because of their spatial proximity to neurons and their bidirectional communication with them, astrocytes are now considered as an important third element of the synapse. Astrocytes integrate and process synaptic information and by doing so generate cytosolic calcium signals that are believed to reflect neuronal transmitter release. Moreover, they regulate neuronal information transmission by releasing gliotransmitters into the synaptic cleft affecting both pre- and postsynaptic receptors. Concurrent with the first experimental reports of the astrocytic impact on neural network dynamics, computational models describing astrocytic functions have been developed. In this review, we give an overview over the published computational models of astrocytic functions, from single-cell dynamics to the tripartite synapse level and network models of astrocytes and neurons.

Deep Learning for Health Informatics

With a massive influx of multimodality data, the role of data analytics in health informatics has grown rapidly in the last decade. This has also prompted increasing interests in the generation of analytical, data driven models based on machine learning in health informatics. Deep learning, a technique with its foundation in artificial neural networks, is emerging in recent years as a powerful tool for machine learning, promising to reshape the future of artificial intelligence. Rapid improvements in computational power, fast data storage, and parallelization have also contributed to the rapid uptake of the technology in addition to its predictive power and ability to generate automatically optimized high-level features and semantic interpretation from the input data. This article presents a comprehensive up-to-date review of research employing deep learning in health informatics, providing a critical analysis of the relative merit, and potential pitfalls of the technique as well as its future outlook. The paper mainly focuses on key applications of deep learning in the fields of translational bioinformatics, medical imaging, pervasive sensing, medical informatics, and public health.

The BioGRID interaction database: 2017 update

The Biological General Repository for Interaction Datasets (BioGRID: https://thebiogrid.org) is an open access database dedicated to the annotation and archival of protein, genetic and chemical interactions for all major model organism species and humans. As of September 2016 (build 3.4.140), the BioGRID contains 1 072 173 genetic and protein interactions, and 38 559 post-translational modifications, as manually annotated from 48 114 publications. This dataset represents interaction records for 66 model organisms and represents a 30% increase compared to the previous 2015 BioGRID update. BioGRID curates the biomedical literature for major model organism species, including humans, with a recent emphasis on central biological processes and specific human diseases. To facilitate network-based approaches to drug discovery, BioGRID now incorporates 27 501 chemical-protein interactions for human drug targets, as drawn from the DrugBank database. A new dynamic interaction network viewer allows the easy navigation and filtering of all genetic and protein interaction data, as well as for bioactive compounds and their established targets. BioGRID data are directly downloadable without restriction in a variety of standardized formats and are freely distributed through partner model organism databases and meta-databases.

Data Analysis in Chemistry and Bio-Medical Sciences

n/a.