Front Plant Sci. 2024 Jun 17;15:1346182. doi: 10.3389/fpls.2024.1346182. eCollection 2024.

ABSTRACT

Accurate and real-time field wheat ear counting is of great significance for wheat yield prediction, genetic breeding and optimized planting management. In order to realize wheat ear detection and counting under the large-resolution Unmanned Aerial Vehicle (UAV) video, Space to depth (SPD) module was added to the deep learning model YOLOv7x. The Normalized Gaussian Wasserstein Distance (NWD) Loss function is designed to create a new detection model YOLOv7xSPD. The precision, recall, F1 score and AP of the model on the test set are 95.85%, 94.71%, 95.28%, and 94.99%, respectively. The AP value is 1.67% higher than that of YOLOv7x, and 10.41%, 39.32%, 2.96%, and 0.22% higher than that of Faster RCNN, SSD, YOLOv5s, and YOLOv7. YOLOv7xSPD is combined with the Kalman filter tracking and the Hungarian matching algorithm to establish a wheat ear counting model with the video flow, called YOLOv7xSPD Counter, which can realize real-time counting of wheat ears in the field. In the video with a resolution of 3840×2160, the detection frame rate of YOLOv7xSPD Counter is about 5.5FPS. The counting results are highly correlated with the ground truth number (R2 = 0.99), and can provide model basis for wheat yield prediction, genetic breeding and optimized planting management.

PMID:38952848 | PMC:PMC11215144 | DOI:10.3389/fpls.2024.1346182

Front Comput Neurosci. 2024 Jun 17;18:1415967. doi: 10.3389/fncom.2024.1415967. eCollection 2024.

ABSTRACT

Electroencephalogram (EEG) plays a pivotal role in the detection and analysis of epileptic seizures, which affects over 70 million people in the world. Nonetheless, the visual interpretation of EEG signals for epilepsy detection is laborious and time-consuming. To tackle this open challenge, we introduce a straightforward yet efficient hybrid deep learning approach, named ResBiLSTM, for detecting epileptic seizures using EEG signals. Firstly, a one-dimensional residual neural network (ResNet) is tailored to adeptly extract the local spatial features of EEG signals. Subsequently, the acquired features are input into a bidirectional long short-term memory (BiLSTM) layer to model temporal dependencies. These output features are further processed through two fully connected layers to achieve the final epileptic seizure detection. The performance of ResBiLSTM is assessed on the epileptic seizure datasets provided by the University of Bonn and Temple University Hospital (TUH). The ResBiLSTM model achieves epileptic seizure detection accuracy rates of 98.88-100% in binary and ternary classifications on the Bonn dataset. Experimental outcomes for seizure recognition across seven epilepsy seizure types on the TUH seizure corpus (TUSZ) dataset indicate that the ResBiLSTM model attains a classification accuracy of 95.03% and a weighted F1 score of 95.03% with 10-fold cross-validation. These findings illustrate that ResBiLSTM outperforms several recent deep learning state-of-the-art approaches.

PMID:38952709 | PMC:PMC11215953 | DOI:10.3389/fncom.2024.1415967

IEEE Int Conf Data Min Workshops. 2023 Dec;2023:329-337. doi: 10.1109/icdmw60847.2023.00048. Epub 2024 Feb 6.

ABSTRACT

Tailoring treatment for severely ill patients is crucial yet challenging to achieve optimal healthcare outcomes. Recent advances in reinforcement learning offer promising personalized treatment recommendations. However, they often rely solely on a patient's current physiological state, which may not accurately represent the true health status of the patient. This limitation hampers policy learning and evaluation, undermining the effectiveness of the treatment. In this study, we propose Deep Attention Q-Network for personalized treatment recommendation, leveraging the Transformer architecture within a deep reinforcement learning framework to efficiently integrate historical observations of patients. We evaluated our proposed method on two real-world datasets: sepsis and acute hypotension patients, demonstrating its superiority over state-of-the-art methods. The source code for our model is available at https://github.com/stevenmsm/RL-ICU-DAQN.

PMID:38952485 | PMC:PMC11216720 | DOI:10.1109/icdmw60847.2023.00048

Front Artif Intell. 2024 Jun 17;7:1321884. doi: 10.3389/frai.2024.1321884. eCollection 2024.

ABSTRACT

BACKGROUND: Carotid plaques are major risk factors for stroke. Carotid ultrasound can help to assess the risk and incidence rate of stroke. However, large-scale carotid artery screening is time-consuming and laborious, the diagnostic results inevitably involve the subjectivity of the diagnostician to a certain extent. Deep learning demonstrates the ability to solve the aforementioned challenges. Thus, we attempted to develop an automated algorithm to provide a more consistent and objective diagnostic method and to identify the presence and stability of carotid plaques using deep learning.

METHODS: A total of 3,860 ultrasound images from 1,339 participants who underwent carotid plaque assessment between January 2021 and March 2023 at the Shanghai Eighth People's Hospital were divided into a 4:1 ratio for training and internal testing. The external test included 1,564 ultrasound images from 674 participants who underwent carotid plaque assessment between January 2022 and May 2023 at Xinhua Hospital affiliated with Dalian University. Deep learning algorithms, based on the fusion of a bilinear convolutional neural network with a residual neural network (BCNN-ResNet), were used for modeling to detect carotid plaques and assess plaque stability. We chose AUC as the main evaluation index, along with accuracy, sensitivity, and specificity as auxiliary evaluation indices.

RESULTS: Modeling for detecting carotid plaques involved training and internal testing on 1,291 ultrasound images, with 617 images showing plaques and 674 without plaques. The external test comprised 470 ultrasound images, including 321 images with plaques and 149 without. Modeling for assessing plaque stability involved training and internal testing on 764 ultrasound images, consisting of 494 images with unstable plaques and 270 with stable plaques. The external test was composed of 279 ultrasound images, including 197 images with unstable plaques and 82 with stable plaques. For the task of identifying the presence of carotid plaques, our model achieved an AUC of 0.989 (95% CI: 0.840, 0.998) with a sensitivity of 93.2% and a specificity of 99.21% on the internal test. On the external test, the AUC was 0.951 (95% CI: 0.962, 0.939) with a sensitivity of 95.3% and a specificity of 82.24%. For the task of identifying the stability of carotid plaques, our model achieved an AUC of 0.896 (95% CI: 0.865, 0.922) on the internal test with a sensitivity of 81.63% and a specificity of 87.27%. On the external test, the AUC was 0.854 (95% CI: 0.889, 0.830) with a sensitivity of 68.52% and a specificity of 89.49%.

CONCLUSION: Deep learning using BCNN-ResNet algorithms based on routine ultrasound images could be useful for detecting carotid plaques and assessing plaque instability.

PMID:38952409 | PMC:PMC11215125 | DOI:10.3389/frai.2024.1321884

Front Artif Intell. 2024 Jun 17;7:1386753. doi: 10.3389/frai.2024.1386753. eCollection 2024.

ABSTRACT

INTRODUCTION: Computerized sentiment detection, based on artificial intelligence and computer vision, has become essential in recent years. Thanks to developments in deep neural networks, this technology can now account for environmental, social, and cultural factors, as well as facial expressions. We aim to create more empathetic systems for various purposes, from medicine to interpreting emotional interactions on social media.

METHODS: To develop this technology, we combined authentic images from various databases, including EMOTIC (ADE20K, MSCOCO), EMODB_SMALL, and FRAMESDB, to train our models. We developed two sophisticated algorithms based on deep learning techniques, DCNN and VGG19. By optimizing the hyperparameters of our models, we analyze context and body language to improve our understanding of human emotions in images. We merge the 26 discrete emotional categories with the three continuous emotional dimensions to identify emotions in context. The proposed pipeline is completed by fusing our models.

RESULTS: We adjusted the parameters to outperform previous methods in capturing various emotions in different contexts. Our study showed that the Sentiment_recognition_model and VGG19_contexte increased mAP by 42.81% and 44.12%, respectively, surpassing the results of previous studies.

DISCUSSION: This groundbreaking research could significantly improve contextual emotion recognition in images. The implications of these promising results are far-reaching, extending to diverse fields such as social robotics, affective computing, human-machine interaction, and human-robot communication.

PMID:38952408 | PMC:PMC11216079 | DOI:10.3389/frai.2024.1386753

Scand J Gastroenterol. 2024 Jul 1:1-8. doi: 10.1080/00365521.2024.2368241. Online ahead of print.

ABSTRACT

OBJECTIVES: Recently, artificial intelligence (AI) has been applied to clinical diagnosis. Although AI has already been developed for gastrointestinal (GI) tract endoscopy, few studies have applied AI to endoscopic ultrasound (EUS) images. In this study, we used a computer-assisted diagnosis (CAD) system with deep learning analysis of EUS images (EUS-CAD) and assessed its ability to differentiate GI stromal tumors (GISTs) from other mesenchymal tumors and their risk classification performance.

MATERIALS AND METHODS: A total of 101 pathologically confirmed cases of subepithelial lesions (SELs) arising from the muscularis propria layer, including 69 GISTs, 17 leiomyomas and 15 schwannomas, were examined. A total of 3283 EUS images were used for training and five-fold-cross-validation, and 827 images were independently tested for diagnosing GISTs. For the risk classification of 69 GISTs, including very-low-, low-, intermediate- and high-risk GISTs, 2,784 EUS images were used for training and three-fold-cross-validation.

RESULTS: For the differential diagnostic performance of GIST among all SELs, the accuracy, sensitivity, specificity and area under the receiver operating characteristic (ROC) curve were 80.4%, 82.9%, 75.3% and 0.865, respectively, whereas those for intermediate- and high-risk GISTs were 71.8%, 70.2%, 72.0% and 0.771, respectively.

CONCLUSIONS: The EUS-CAD system showed a good diagnostic yield in differentiating GISTs from other mesenchymal tumors and successfully demonstrated the GIST risk classification feasibility. This system can determine whether treatment is necessary based on EUS imaging alone without the need for additional invasive examinations.

PMID:38950889 | DOI:10.1080/00365521.2024.2368241

Nat Methods. 2024 Jul 1. doi: 10.1038/s41592-024-02315-5. Online ahead of print.

NO ABSTRACT

PMID:38951671 | DOI:10.1038/s41592-024-02315-5

Nat Methods. 2024 Jul 1. doi: 10.1038/s41592-024-02314-6. Online ahead of print.

ABSTRACT

Glycans constitute the most complicated post-translational modification, modulating protein activity in health and disease. However, structural annotation from tandem mass spectrometry (MS/MS) data is a bottleneck in glycomics, preventing high-throughput endeavors and relegating glycomics to a few experts. Trained on a newly curated set of 500,000 annotated MS/MS spectra, here we present CandyCrunch, a dilated residual neural network predicting glycan structure from raw liquid chromatography-MS/MS data in seconds (top-1 accuracy: 90.3%). We developed an open-access Python-based workflow of raw data conversion and prediction, followed by automated curation and fragment annotation, with predictions recapitulating and extending expert annotation. We demonstrate that this can be used for de novo annotation, diagnostic fragment identification and high-throughput glycomics. For maximum impact, this entire pipeline is tightly interlaced with our glycowork platform and can be easily tested at https://colab.research.google.com/github/BojarLab/CandyCrunch/blob/main/CandyCrunch.ipynb . We envision CandyCrunch to democratize structural glycomics and the elucidation of biological roles of glycans.

PMID:38951670 | DOI:10.1038/s41592-024-02314-6

Nat Med. 2024 Jul 1. doi: 10.1038/s41591-024-03126-z. Online ahead of print.

NO ABSTRACT

PMID:38951637 | DOI:10.1038/s41591-024-03126-z

Sci Rep. 2024 Jul 1;14(1):14771. doi: 10.1038/s41598-024-65639-4.

ABSTRACT

Software defect prediction aims to find a reliable method for predicting defects in a particular software project and assisting software engineers in allocating limited resources to release high-quality software products. While most earlier research has concentrated on employing traditional features, current methodologies are increasingly directed toward extracting semantic features from source code. Traditional features often fall short in identifying semantic differences within programs, differences that are essential for the development of reliable and effective prediction models. In contrast, semantic features cannot present statistical metrics about the source code, such as the code size and complexity. Thus, using only one kind of feature negatively affects prediction performance. To bridge the gap between the traditional and semantic features, we propose a novel defect prediction model that integrates traditional and semantic features using a hybrid deep learning approach to address this limitation. Specifically, our model employs a hybrid CNN-MLP classifier: the convolutional neural network (CNN) processes semantic features extracted from projects' abstract syntax trees (ASTs) using Word2vec. In contrast, the traditional features extracted from the dataset repository are processed by a multilayer perceptron (MLP). Outputs of CNN and MLP are then integrated and fed into a fully connected layer for defect prediction. Extensive experiments are conducted on various open-source projects to validate CNN-MLP's effectiveness. Experimental results indicate that CNN-MLP can significantly enhance defect prediction performance. Furthermore, CNN-MLP's improvements outperform existing methods in non-effort-aware and effort-aware cases.

PMID:38951608 | DOI:10.1038/s41598-024-65639-4

Sci Rep. 2024 Jul 1;14(1):15051. doi: 10.1038/s41598-024-65837-0.

ABSTRACT

Electrical conductivity (EC) is widely recognized as one of the most essential water quality metrics for predicting salinity and mineralization. In the current research, the EC of two Australian rivers (Albert River and Barratta Creek) was forecasted for up to 10 days using a novel deep learning algorithm (Convolutional Neural Network combined with Long Short-Term Memory Model, CNN-LSTM). The Boruta-XGBoost feature selection method was used to determine the significant inputs (time series lagged data) to the model. To compare the performance of Boruta-XGB-CNN-LSTM models, three machine learning approaches-multi-layer perceptron neural network (MLP), K-nearest neighbour (KNN), and extreme gradient boosting (XGBoost) were used. Different statistical metrics, such as correlation coefficient (R), root mean square error (RMSE), and mean absolute percentage error, were used to assess the models' performance. From 10 years of data in both rivers, 7 years (2012-2018) were used as a training set, and 3 years (2019-2021) were used for testing the models. Application of the Boruta-XGB-CNN-LSTM model in forecasting one day ahead of EC showed that in both stations, Boruta-XGB-CNN-LSTM can forecast the EC parameter better than other machine learning models for the test dataset (R = 0.9429, RMSE = 45.6896, MAPE = 5.9749 for Albert River, and R = 0.9215, RMSE = 43.8315, MAPE = 7.6029 for Barratta Creek). Considering the better performance of the Boruta-XGB-CNN-LSTM model in both rivers, this model was used to forecast 3-10 days ahead of EC. The results showed that the Boruta-XGB-CNN-LSTM model is very capable of forecasting the EC for the next 10 days. The results showed that by increasing the forecasting horizon from 3 to 10 days, the performance of the Boruta-XGB-CNN-LSTM model slightly decreased. The results of this study show that the Boruta-XGB-CNN-LSTM model can be used as a good soft computing method for accurately predicting how the EC will change in rivers.

PMID:38951605 | DOI:10.1038/s41598-024-65837-0

Sci Rep. 2024 Jul 1;14(1):14997. doi: 10.1038/s41598-024-65909-1.

ABSTRACT

Cracks in tunnel lining structures constitute a common and serious problem that jeopardizes the safety of traffic and the durability of the tunnel. The similarity between lining seams and cracks in terms of strength and morphological characteristics renders the detection of cracks in tunnel lining structures challenging. To address this issue, a new deep learning-based method for crack detection in tunnel lining structures is proposed. First, an improved attention mechanism is introduced for the morphological features of lining seams, which not only aggregates global spatial information but also features along two dimensions, height and width, to mine more long-distance feature information. Furthermore, a mixed strip convolution module leveraging four different directions of strip convolution is proposed. This module captures remote contextual information from various angles to avoid interference from background pixels. To evaluate the proposed approach, the two modules are integrated into a U-shaped network, and experiments are conducted on Tunnel200, a tunnel lining crack dataset, as well as the publicly available crack datasets Crack500 and DeepCrack. The results show that the approach outperforms existing methods and achieves superior performance on these datasets.

PMID:38951575 | DOI:10.1038/s41598-024-65909-1

Sci Rep. 2024 Jul 1;14(1):15041. doi: 10.1038/s41598-024-63930-y.

ABSTRACT

The Indian economy is greatly influenced by the Banana Industry, necessitating advancements in agricultural farming. Recent research emphasizes the imperative nature of addressing diseases that impact Banana Plants, with a particular focus on early detection to safeguard production. The urgency of early identification is underscored by the fact that diseases predominantly affect banana plant leaves. Automated systems that integrate machine learning and deep learning algorithms have proven to be effective in predicting diseases. This manuscript examines the prediction and detection of diseases in banana leaves, exploring various diseases, machine learning algorithms, and methodologies. The study makes a contribution by proposing two approaches for improved performance and suggesting future research directions. In summary, the objective is to advance understanding and stimulate progress in the prediction and detection of diseases in banana leaves. The need for enhanced disease identification processes is highlighted by the results of the survey. Existing models face a challenge due to their lack of rotation and scale invariance. While algorithms such as random forest and decision trees are less affected, initially convolutional neural networks (CNNs) is considered for disease prediction. Though the Convolutional Neural Network models demonstrated impressive accuracy in many research but it lacks in invariance to scale and rotation. Moreover, it is observed that due its inherent design it cannot be combined with feature extraction methods to identify the banana leaf diseases. Due to this reason two alternative models that combine ANN with scale-invariant Feature transform (SIFT) model or histogram of oriented gradients (HOG) combined with local binary patterns (LBP) model are suggested. The first model ANN with SIFT identify the disease by using the activation functions to process the features extracted by the SIFT by distinguishing the complex patterns. The second integrate the combined features of HOG and LBP to identify the disease thus by representing the local pattern and gradients in an image. This paves a way for the ANN to learn and identify the banana leaf disease. Moving forward, exploring datasets in video formats for disease detection in banana leaves through tailored machine learning algorithms presents a promising avenue for research.

PMID:38951552 | DOI:10.1038/s41598-024-63930-y

Int J Comput Assist Radiol Surg. 2024 Jul 1. doi: 10.1007/s11548-024-03202-2. Online ahead of print.

ABSTRACT

PURPOSE: Micro-electrode recordings (MERs) are a key intra-operative modality used during deep brain stimulation (DBS) electrode implantation, which allow for a trained neurophysiologist to infer the anatomy in which the electrode is placed. As DBS targets are small, such inference is necessary to confirm that the electrode is correctly positioned. Recently, machine learning techniques have been used to augment the neurophysiologist's capability. The goal of this paper is to investigate the generalisability of these methods with respect to different clinical centres and training paradigms.

METHODS: Five deep learning algorithms for binary classification of MER signals have been implemented. Three databases from two different clinical centres have also been collected with differing size, acquisition hardware, and annotation protocol. Each algorithm has initially been trained on the largest database, then either directly tested or fine-tuned on the smaller databases in order to estimate their generalisability. As a reference, they have also been trained from scratch on the smaller databases as well in order to estimate the effect of the differing database sizes and annotation systems.

RESULTS: Each network shows significantly reduced performance (on the order of a 6.5% to 16.0% reduction in balanced accuracy) when applied out-of-distribution. This reduction can be ameliorated through fine-tuning the network on the new database through transfer learning. Although, even for these small databases, it appears that retraining from scratch may still offer equivalent performance as fine-tuning with transfer learning. However, this is at the expense of significantly longer training times.

CONCLUSION: Generalisability is an important criterion for the success of machine learning algorithms in clinic. We have demonstrated that a variety of recent machine learning algorithms for MER classification are negatively affected by domain shift, but that this can be quickly ameliorated through simple transfer learning procedures that can be readily performed for new centres.

PMID:38951363 | DOI:10.1007/s11548-024-03202-2

Neurosurg Rev. 2024 Jun 29;47(1):300. doi: 10.1007/s10143-024-02537-3.

ABSTRACT

The diagnosis of Moyamoya disease (MMD) relies heavily on imaging, which could benefit from standardized machine learning tools. This study aims to evaluate the diagnostic efficacy of deep learning (DL) algorithms for MMD by analyzing sensitivity, specificity, and the area under the curve (AUC) compared to expert consensus. We conducted a systematic search of PubMed, Embase, and Web of Science for articles published from inception to February 2024. Eligible studies were required to report diagnostic accuracy metrics such as sensitivity, specificity, and AUC, excluding those not in English or using traditional machine learning methods. Seven studies were included, comprising a sample of 4,416 patients, of whom 1,358 had MMD. The pooled sensitivity for common and random effects models was 0.89 (95% CI: 0.85 to 0.92) and 0.92 (95% CI: 0.85 to 0.96), respectively. The pooled specificity was 0.89 (95% CI: 0.86 to 0.91) in the common effects model and 0.91 (95% CI: 0.75 to 0.97) in the random effects model. Two studies reported the AUC alongside their confidence intervals. A meta-analysis synthesizing these findings aggregated a mean AUC of 0.94 (95% CI: 0.92 to 0.96) for common effects and 0.89 (95% CI: 0.76 to 1.02) for random effects models. Deep learning models significantly enhance the diagnosis of MMD by efficiently extracting and identifying complex image patterns with high sensitivity and specificity. Trial registration: CRD42024524998 https://www.crd.york.ac.uk/prospero/displayrecord.php?RecordID=524998.

PMID:38951288 | DOI:10.1007/s10143-024-02537-3

Opt Lett. 2024 Jul 1;49(13):3741-3744. doi: 10.1364/OL.519973.

ABSTRACT

In general, delay operation is the most time-consuming stage in frequency-resolved optical gating (FROG) technology, which limits the use of FROG for high-speed measurement of ultrashort laser pulses. In this work, we propose and demonstrate the reconstruction of ultrashort optical pulses by employing the sequence-to-sequence (Seq2Seq) model with attention, theoretically. To our knowledge, this is the first deep learning framework capable of accurately reconstructing ultrashort pulses using very partial spectrograms. The root mean squared error (RMSE) of the pulse amplitude reconstruction and phase reconstruction on the overall test dataset are 9.5 × 10-4 and 0.20, respectively. Compared with the classic FROG recovery algorithm based on two-dimensional phase retrieval algorithms, the use of our model can shorten the spectral measurement time to 1/8 of the original time or even less. Meanwhile, the time required for pulse reconstruction using our model is roughly 0.2 s. To our knowledge, the pulse reconstruction speed of our model exceeds all current iteration-based FROG recovery algorithms. We believe that this study can greatly facilitate the use of FROG for high-speed measurements of ultrashort pulses.

PMID:38950256 | DOI:10.1364/OL.519973

J Chem Inf Model. 2024 Jul 1. doi: 10.1021/acs.jcim.4c00457. Online ahead of print.

ABSTRACT

Machine-learning (ML) and deep-learning (DL) approaches to predict the molecular properties of small molecules are increasingly deployed within the design-make-test-analyze (DMTA) drug design cycle to predict molecular properties of interest. Despite this uptake, there are only a few automated packages to aid their development and deployment that also support uncertainty estimation, model explainability, and other key aspects of model usage. This represents a key unmet need within the field, and the large number of molecular representations and algorithms (and associated parameters) means it is nontrivial to robustly optimize, evaluate, reproduce, and deploy models. Here, we present QSARtuna, a molecule property prediction modeling pipeline, written in Python and utilizing the Optuna, Scikit-learn, RDKit, and ChemProp packages, which enables the efficient and automated comparison between molecular representations and machine learning models. The platform was developed by considering the increasingly important aspect of model uncertainty quantification and explainability by design. We provide details for our framework and provide illustrative examples to demonstrate the capability of the software when applied to simple molecular property, reaction/reactivity prediction, and DNA encoded library enrichment classification. We hope that the release of QSARtuna will further spur innovation in automatic ML modeling and provide a platform for education of best practices in molecular property modeling. The code for the QSARtuna framework is made freely available via GitHub.

PMID:38950185 | DOI:10.1021/acs.jcim.4c00457

Bioinformatics. 2024 Jul 1:btae433. doi: 10.1093/bioinformatics/btae433. Online ahead of print.

ABSTRACT

MOTIVATION: The rise of single-cell RNA sequencing (scRNA-seq) technology presents new opportunities for constructing detailed cell type-specific gene regulatory networks (GRNs) to study cell heterogeneity. However, challenges caused by noises, technical errors, and dropout phenomena in scRNA-seq data pose significant obstacles to GRN inference, making the design of accurate GRN inference algorithms still essential. The recent growth of both single-cell and spatial transcriptomic sequencing data enables the development of supervised deep learning methods to infer GRNs on these diverse single-cell datasets.

RESULTS: In this study, we introduce a novel deep learning framework based on shared factor neighborhood and integrated neural network (SFINN) for inferring potential interactions and causalities between transcription factors and target genes from single-cell and spatial transcriptomic data. SFINN utilizes shared factor neighborhood to construct cellular neighborhood network based on gene expression data and additionally integrates cellular network generated from spatial location information. Subsequently, the cell adjacency matrix and gene pair expression are fed into an integrated neural network framework consisting of a graph convolutional neural network and a fully-connected neural network to determine whether the genes interact. Performance evaluation in the tasks of gene interaction and causality prediction against the existing GRN reconstruction algorithms demonstrates the usability and competitiveness of SFINN across different kinds of data. SFINN can be applied to infer gene regulatory networks from conventional single-cell sequencing data and spatial transcriptomic data.

AVAILABILITY: SFINN can be accessed at GitHub: https://github.com/JGuan-lab/SFINN.

SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

PMID:38950180 | DOI:10.1093/bioinformatics/btae433

Bioinformatics. 2024 Jul 1:btae434. doi: 10.1093/bioinformatics/btae434. Online ahead of print.

ABSTRACT

MOTIVATION: Gene set enrichment (GSE) analysis allows for an interpretation of gene expression through pre-defined gene set databases and is a critical step in understanding different phenotypes. With the rapid development of single-cell RNA sequencing (scRNA-seq) technology, GSE analysis can be performed on fine-grained gene expression data to gain a nuanced understanding of phenotypes of interest. However, with the cellular heterogeneity in single-cell gene profiles, current statistical GSE analysis methods sometimes fail to identify enriched gene sets. Meanwhile, deep learning has gained traction in applications like clustering and trajectory inference in single-cell studies due to its prowess in capturing complex data patterns. However, its use in GSE analysis remains limited, due to interpretability challenges.

RESULTS: In this paper, we present DeepGSEA, an explainable deep gene set enrichment analysis approach which leverages the expressiveness of interpretable, prototype-based neural networks to provide an in-depth analysis of GSE. DeepGSEA learns the ability to capture GSE information through our designed classification tasks, and significance tests can be performed on each gene set, enabling the identification of enriched sets. The underlying distribution of a gene set learned by DeepGSEA can be explicitly visualized using the encoded cell and cellular prototype embeddings. We demonstrate the performance of DeepGSEA over commonly used GSE analysis methods by examining their sensitivity and specificity with four simulation studies. In addition, we test our model on three real scRNA-seq datasets and illustrate the interpretability of DeepGSEA by showing how its results can be explained.

AVAILABILITY: https://github.com/Teddy-XiongGZ/DeepGSEA.

PMID:38950178 | DOI:10.1093/bioinformatics/btae434

ACS Nano. 2024 Jul 1. doi: 10.1021/acsnano.4c04851. Online ahead of print.

ABSTRACT

Prevailing over the bottleneck of von Neumann computing has been significant attention due to the inevitableness of proceeding through enormous data volumes in current digital technologies. Inspired by the human brain's operational principle, the artificial synapse of neuromorphic computing has been explored as an emerging solution. Especially, the optoelectronic synapse is of growing interest as vision is an essential source of information in which dealing with optical stimuli is vital. Herein, flexible optoelectronic synaptic devices composed of centimeter-scale tellurium dioxide (TeO2) films detecting and exhibiting synaptic characteristics to broadband wavelengths are presented. The TeO2-based flexible devices demonstrate a comprehensive set of emulating basic optoelectronic synaptic characteristics; i.e., excitatory postsynaptic current (EPSC), paired-pulse facilitation (PPF), conversion of short-term to long-term memory, and learning/forgetting. Furthermore, they feature linear and symmetric conductance synaptic weight updates at various wavelengths, which are applicable to broadband neuromorphic computations. Based on this large set of synaptic attributes, a variety of applications such as logistic functions or deep learning and image recognition as well as learning simulations are demonstrated. This work proposes a significant milestone of wafer-scale metal oxide semiconductor-based artificial synapses solely utilizing their optoelectronic features and mechanical flexibility, which is attractive toward scaled-up neuromorphic architectures.

PMID:38950148 | DOI:10.1021/acsnano.4c04851