DRSPRING: Graph convolutional network (GCN)-Based drug synergy prediction utilizing drug-induced gene expression profile

Great efforts have been made over the years to identify novel drug pairs with synergistic effects. Although numerous computational approaches have been proposed to analyze diverse types of biological big data, the pharmacogenomic profiles, presumably the most direct proxy of drug effects, have been rarely used due to the data sparsity problem. In this study, we developed a composite deep-learning-based model that predicts the drug synergy effect utilizing pharmacogenomic profiles as well as molecular properties. Graph convolutional network (GCN) was used to represent and integrate the chemical structure, genetic interactions, drug-target information, and gene expression profiles of cell lines. Insufficient amount of pharmacogenomic data, i.e., drug-induced expression profiles from the LINCS project, was resolved by augmenting the data with the predicted profiles. Our method learned and predicted the Loewe synergy score in the DrugComb database and achieved a better or comparable performance compared to other published methods in a benchmark test. We also investigated contribution of various input features, which highlighted the value of basal gene expression and pharmacogenomic profiles of each cell line. Importantly, DRSPRING (DRug Synergy PRediction by INtegrated GCN) can be applied to any drug pairs and any cell lines, greatly expanding its applicability compared to previous methods.

ECG-surv: A deep learning-based model to predict time to 1-year mortality from 12-lead electrocardiogram

BACKGROUND

Electrocardiogram (ECG) abnormalities have demonstrated potential as prognostic indicators of patient survival. However, the traditional statistical approach is constrained by structured data input, limiting its ability to fully leverage the predictive value of ECG data in prognostic modeling.

METHODS

This study aims to introduce and evaluate a deep-learning model to simultaneously handle censored data and unstructured ECG data for survival analysis. We herein introduce a novel deep neural network called ECG-surv, which includes a feature extraction neural network and a time-to-event analysis neural network. The proposed model is specifically designed to predict the time to 1-year mortality by extracting and analyzing unique features from 12-lead ECG data. ECG-surv was evaluated using both an independent test set and an external set, which were collected using different ECG devices.

RESULTS

The performance of ECG-surv surpassed that of the Cox proportional model, which included demographics and ECG waveform parameters, in predicting 1-year all-cause mortality, with a significantly higher concordance index (C-index) in ECG-surv than in the Cox model using both the independent test set (0.860 [95% CI: 0.859- 0.861] vs. 0.796 [95% CI: 0.791- 0.800]) and the external test set (0.813 [95% CI: 0.807- 0.814] vs. 0.764 [95% CI: 0.755- 0.770]). ECG-surv also demonstrated exceptional predictive ability for cardiovascular death (C-index of 0.891 [95% CI: 0.890- 0.893]), outperforming the Framingham risk Cox model (C-index of 0.734 [95% CI: 0.715-0.752]).

CONCLUSION

ECG-surv effectively utilized unstructured ECG data in a survival analysis. It outperformed traditional statistical approaches in predicting 1-year all-cause mortality and cardiovascular death, which makes it a valuable tool for predicting patient survival.

Carbon sequestration costs and spatial spillover effects in China's collective forests

BACKGROUND

Global climate change is one of the major challenges facing the world today, and forests play a crucial role as significant carbon sinks and providers of ecosystem services in mitigating climate change and protecting the environment. China, as one of the largest developing countries globally, owns 60% of its forest resources collectively. Evaluating the carbon sequestration cost of collective forests not only helps assess the contribution of China's forest resources to global climate change mitigation but also provides important evidence for formulating relevant policies and measures.

RESULTS

Over the past 30 years, the carbon sequestration cost of collective forests in China has shown an overall upward trend. Except for coastal provinces, southern collective forest areas, as well as some southwestern and northeastern regions, have the advantage of lower carbon sequestration costs. Furthermore, LSTM network predictions indicate that the carbon sequestration cost of collective forests in China will continue to rise. By 2030, the average carbon sequestration cost of collective forests is projected to reach 125 CNY per ton(= 16.06 Euros/t). Additionally, there is spatial correlation in the carbon sequestration cost of collective forests. Timber production, labor costs, and labor prices have negative spatial spillover effects on carbon sequestration costs, while land opportunity costs, forest accumulation, and rural resident consumption have positive spatial spillover effects.

CONCLUSION

The results of this study indicate regional disparities in the spatial distribution of carbon sequestration costs of collective forests, with an undeniable upward trend in future cost growth. It is essential to focus on areas with lower carbon sequestration costs and formulate targeted carbon sink economic policies and management measures to maximize the carbon sequestration potential of collective forests and promote the sustainable development of forestry.

Multi-scale full spike pattern for semantic segmentation

Spiking neural networks (SNNs), as the brain-inspired neural networks, encode information in spatio-temporal dynamics. They have the potential to serve as low-power alternatives to artificial neural networks (ANNs) due to their sparse and event-driven nature. However, existing SNN-based models for pixel-level semantic segmentation tasks suffer from poor performance and high memory overhead, failing to fully exploit the computational effectiveness and efficiency of SNNs. To address these challenges, we propose the multi-scale and full spike segmentation network (MFS-Seg), which is based on the deep direct trained SNN and represents the first attempt to train a deep SNN with surrogate gradients for semantic segmentation. Specifically, we design an efficient fully-spike residual block (EFS-Res) to alleviate representation issues caused by spiking noise on different channels. EFS-Res utilizes depthwise separable convolution to improve the distributions of spiking feature maps. The visualization shows that our model can effectively extract the edge features of segmented objects. Furthermore, it can significantly reduce the memory overhead and energy consumption of the network. In addition, we theoretically analyze and prove that EFS-Res can avoid the degradation problem based on block dynamical isometry theory. Experimental results on the Camvid dataset, the DDD17 dataset, and the DSEC-Semantic dataset show that our model achieves comparable performance to the mainstream UNet network with up to 31× fewer parameters, while significantly reducing power consumption by over 13×. Overall, our MFS-Seg model demonstrates promising results in terms of performance, memory efficiency, and energy consumption, showcasing the potential of deep SNNs for semantic segmentation tasks. Our code is available in https://github.com/BICLab/MFS-Seg.

Quantitative detection of multicomponent SF6 decomposition products based on Fourier transform infrared spectroscopy combined with SCARS-DNN

Accurate and efficient quantitative analysis of the decomposition products of the insulating medium SF6 in gas-insulated switchgear (GIS) is important for an effective assessment of its internal insulation status. In this work, a quantitative calibration model of Fourier Transform Infrared Spectroscopy (FTIR) combined with SCARS-DNN (Stability Competitive Adaptive Reweighted Sampling-Deep Neural Network) is proposed for the rapid non-destructive detection of SF6 decomposition products. First, the interference of the background gas SF6 on the absorption spectra of the decomposition products is eliminated according to the Lambert-Beer law, while baseline correction and Savitzky-Golay (S-G) smoothing are used to remove baseline drift and noise. Subsequently, a Monte Carlo cross-validation method is used to detect and eliminate the anomalous samples. Then feature selection is performed using uninformative variable elimination (UVE) and stability competitive adaptive reweighted sampling (SCARS), and finally quantitative calibration models of FULL-DNN (full spectral band), UVE-DNN, and SCARS-DNN are developed. For the quantitative detection of SF6 decomposition products, the SCARS-DNN model had the best prediction performance with a maximum reduction of 96.18% in the root mean square error (RMSE) and 96.11% in the mean absolute percentage error (MAPE). Results reveal that the relative errors are basically kept below 1.36% when predicting the three decomposition products, even in the presence of a high level of SF6 interference. Therefore, the SCARS-DNN model is suitable for high-precision quantitative detection of SF6 decomposition gas.

Investigation of multidirectional toxicity induced by high-dose molybdenum exposure with Allium test

In this study, the multifaceted toxicity induced by high doses of the essential trace element molybdenum in Allium cepa L. was investigated. Germination, root elongation, weight gain, mitotic index (MI), micronucleus (MN), chromosomal abnormalities (CAs), Comet assay, malondialdehyde (MDA), proline, superoxide dismutase (SOD), catalase (CAT) and anatomical parameters were used as biomarkers of toxicity. In addition, detailed correlation and PCA analyzes were performed for all parameters discussed. On the other hand, this study focused on the development of a two hidden layer deep neural network (DNN) using Matlab. Four experimental groups were designed: control group bulbs were germinated in tap water and application group bulbs were germinated with 1000, 2000 and 4000 mg/L doses of molybdenum for 72 h. After germination, root tips were collected and prepared for analysis. As a result, molybdenum exposure caused a dose-dependent decrease (p < 0.05) in the investigated physiological parameter values, and an increase (p < 0.05) in the cytogenetic (except MI) and biochemical parameter values. Molybdenum exposure induced different types of CAs and various anatomical damages in root meristem cells. Comet assay results showed that the severity of DNA damage increased depending on the increasing molybdenum dose. Detailed correlation and PCA analysis results determined significant positive and negative interactions between the investigated parameters and confirmed the relationships of these parameters with molybdenum doses. It has been found that the DNN model is in close agreement with the actual data showing the accuracy of the predictions. MAE, MAPE, RMSE and R2 were used to evaluate the effectiveness of the DNN model. Collective analysis of these metrics showed that the DNN model performed well. As a result, it has been determined once again that high doses of molybdenum cause multiple toxicity in A. cepa and the Allium test is a reliable universal test for determining this toxicity. Therefore, periodic measurement of molybdenum levels in agricultural soils should be the first priority in preventing molybdenum toxicity.

Deep neural network-estimated age using optical coherence tomography predicts mortality

The concept of biological age has emerged as a measurement that reflects physiological and functional decline with ageing. Here we aimed to develop a deep neural network (DNN) model that predicts biological age from optical coherence tomography (OCT). A total of 84,753 high-quality OCT images from 53,159 individuals in the UK Biobank were included, among which 12,631 3D-OCT images from 8,541 participants without any reported medical conditions at baseline were used to develop an age prediction model. For the remaining 44,618 participants, OCT age gap, the difference between the OCT-predicted age and chronological age, was calculated for each participant. Cox regression models assessed the association between OCT age gap and mortality. The DNN model predicted age with a mean absolute error of 3.27 years and showed a strong correlation of 0.85 with chronological age. After a median follow-up of 11.0 years (IQR 10.9-11.1 years), 2,429 deaths (5.44%) were recorded. For each 5-year increase in OCT age gap, there was an 8% increased mortality risk (hazard ratio [HR] = 1.08, CI:1.02-1.13, P = 0.004). Compared with an OCT age gap within ± 4 years, OCT age gap less than minus 4 years was associated with a 16% decreased mortality risk (HR = 0.84, CI: 0.75-0.94, P = 0.002) and OCT age gap more than 4 years showed an 18% increased risk of death incidence (HR = 1.18, CI: 1.02-1.37, P = 0.026). OCT imaging could serve as an ageing biomarker to predict biological age with high accuracy and the OCT age gap, defined as the difference between the OCT-predicted age and chronological age, can be used as a marker of the risk of mortality.

Invariant feature based label correction for DNN when Learning with Noisy Labels

Learning with Noisy Labels (LNL) methods have been widely studied in recent years, which aims to improve the performance of Deep Neural Networks (DNNs) when the training dataset contains incorrectly annotated labels. Popular existing LNL methods rely on semantic features extracted by the DNN to detect and mitigate label noise. However, these extracted features are often spurious and contain unstable correlations with the label across different environments (domains), which can occasionally lead to incorrect prediction and compromise the efficacy of LNL methods. To mitigate this insufficiency, we propose Invariant Feature based Label Correction (IFLC), which reduces spurious features and accurately utilizes the learned invariant features that contain stable correlation to correct label noise. To the best of our knowledge, this is the first attempt to mitigate the issue of spurious features for LNL methods. IFLC consists of two critical processes: The Label Disturbing (LD) process and the Representation Decorrelation (RD) process. The LD process aims to encourage DNN to attain stable performance across different environments, thus reducing the captured spurious features. The RD process strengthens independence between each dimension of the representation vector, thus enabling accurate utilization of the learned invariant features for label correction. We then utilize robust linear regression for the feature representation to conduct label correction. We evaluated the effectiveness of our proposed method and compared it with state-of-the-art (sota) LNL methods on four benchmark datasets, CIFAR-10, CIFAR-100, Animal-10N, and Clothing1M. The experimental results show that our proposed method achieved comparable or even better performance than the existing sota methods. The source codes are available at https://github.com/yangbo1973/IFLC.

Multi-modal deep learning networks for RGB-D pavement waste detection and recognition

To create a clean living environment, governments around the world have hired a large number of workers to clean up waste on pavements, which is inefficient for waste management. To better alleviate this problem, relevant scholars have proposed several deep learning methods based on RGB images to achieve waste detection and recognition. Considering the limitations of color images, we propose an efficient multi-modal learning solution for pavement waste detection and recognition. Specifically, we construct a high-quality outdoor pavement waste dataset called OPWaste, which is more in line with real needs. Compared to other waste datasets, OPWaste dataset not only has the advantages of rich background and high diversity, but also provides color and depth images. Meanwhile, we explore six different multi-modal fusion methods and propose a novel multi-modal multi-scale network (MM-Net) for RGB-D waste detection and recognition. MM-Net introduces a novel multi-scale refinement module (MRM) and multi-scale interaction module (MIM). MRM can effectively refine critical features using attention mechanisms. MIM can gradually realize information interaction between hierarchical features. In addition, we select several representative methods and perform comparative experiments. Experimental results show that MM-Net based on the image addition fusion method outperforms other deep learning models and reaches 97.3% and 84.4% on mAP0.5 and AR metrics. In fact, multi-modal learning plays an important role in intelligent waste recycling. As a promising auxiliary tool, our solution can be applied to intelligent cleaning robots for automatic outdoor waste management.

Short-term wind power forecasting through stacked and bi directional LSTM techniques

Background

Computational intelligence (CI) based prediction models increase the efficient and effective utilization of resources for wind prediction. However, the traditional recurrent neural networks (RNN) are difficult to train on data having long-term temporal dependencies, thus susceptible to an inherent problem of vanishing gradient. This work proposed a method based on an advanced version of RNN known as long short-term memory (LSTM) architecture, which updates recurrent weights to overcome the vanishing gradient problem. This, in turn, improves training performance.

Methods

The RNN model is developed based on stack LSTM and bidirectional LSTM. The parameters like mean absolute error (MAE), standard deviation error (SDE), and root mean squared error (RMSE) are utilized as performance measures for comparison with recent state-of-the-art techniques.

Results

Results showed that the proposed technique outperformed the existing techniques in terms of RMSE and MAE against all the used wind farm datasets. Whereas, a reduction in SDE is observed for larger wind farm datasets. The proposed RNN approach performed better than the existing models despite fewer parameters. In addition, the approach requires minimum processing power to achieve compatible results.

Levenberg-Marquardt deep neural watermarking for 3D mesh using nearest centroid salient point learning

Watermarking is one of the crucial techniques in the domain of information security, preventing the exploitation of 3D Mesh models in the era of Internet. In 3D Mesh watermark embedding, moderately perturbing the vertices is commonly required to retain them in certain pre-arranged relationship with their neighboring vertices. This paper proposes a novel watermarking authentication method, called Nearest Centroid Discrete Gaussian and Levenberg-Marquardt (NCDG-LV), for distortion detection and recovery using salient point detection. In this method, the salient points are selected using the Nearest Centroid and Discrete Gaussian Geometric (NC-DGG) salient point detection model. Map segmentation is applied to the 3D Mesh model to segment into distinct sub regions according to the selected salient points. Finally, the watermark is embedded by employing the Multi-function Barycenter into each spatially selected and segmented region. In the extraction process, the embedded 3D Mesh image is extracted from each re-segmented region by means of Levenberg-Marquardt Deep Neural Network Watermark Extraction. In the authentication stage, watermark bits are extracted by analyzing the geometry via Levenberg-Marquardt back-propagation. Based on a performance evaluation, the proposed method exhibits high imperceptibility and tolerance against attacks, such as smoothing, cropping, translation, and rotation. The experimental results further demonstrate that the proposed method is superior in terms of salient point detection time, distortion rate, true positive rate, peak signal to noise ratio, bit error rate, and root mean square error compared to the state-of-the-art methods.

scFSNN: a feature selection method based on neural network for single-cell RNA-seq data

While single-cell RNA sequencing (scRNA-seq) allows researchers to analyze gene expression in individual cells, its unique characteristics like over-dispersion, zero-inflation, high gene-gene correlation, and large data volume with many features pose challenges for most existing feature selection methods. In this paper, we present a feature selection method based on neural network (scFSNN) to solve classification problem for the scRNA-seq data. scFSNN is an embedded method that can automatically select features (genes) during model training, control the false discovery rate of selected features and adaptively determine the number of features to be eliminated. Extensive simulation and real data studies demonstrate its excellent feature selection ability and predictive performance.

A comparative study of a traditional localization algorithm and a deep learning model for radioactive particle tracking application

Radioactive particle tracking is a nuclear technique that tracks a sealed radioactive particle inside a volume through a mathematical location algorithm, which is widely applied in many fields such as chemical and civil engineering in hydrodynamics flows. It is possible to reconstruct the trajectory of the radioactive particle using a traditional mathematical algorithm or artificial intelligence methods. In this paper, the traditional algorithm is based on solving a minimization problem between the simulated events and a calibration dataset, and it was written using C++ language. The artificial intelligence method is represented by a deep neural network, in which hyperparameters were defined using a Python optimization library called Optuna. This paper aims to compare the potentiality of both methods to evaluate the accuracy of the radioactive particle tracking technique. This study proposes a simplified model of a concrete mixer, six NaI(Tl) detectors, and a137Cs sealed radioactive particle. The simulated measurement geometry and the dataset (3615 patterns) were developed using the MCNPX code, which is a mathematical code based on the Monte Carlo Method. The results show a mean absolute percentage error (MAPE) of 20.81%, 10.33%, and 16.84% for x, y and z coordinates, respectively, for the traditional algorithm. For the deep neural network, MAPE is 6.87%, 2.70%, and 22.79% respectively for x, y and z coordinates. In addition, an investigation is carried out to analyze whether the size of the calibration dataset influences the performance of both methods.

Recent developments in machine learning modeling methods for hypertension treatment

Hypertension is the leading cause of cardiovascular complications. This review focuses on the advancements in medical artificial intelligence (AI) models aimed at individualized treatment for hypertension, with particular emphasis on the approach to time-series big data on blood pressure and the development of interpretable medical AI models. The digitalization of daily blood pressure records and the downsizing of measurement devices enable the accumulation and utilization of time-series data. As mainstream blood pressure data shift from snapshots to time series, the clinical significance of blood pressure variability will be clarified. The time-series blood pressure prediction model demonstrated the capability to forecast blood pressure variabilities with a reasonable degree of accuracy for up to four weeks in advance. In recent years, various explainable AI techniques have been proposed for different purposes of model interpretation. It is essential to select the appropriate technique based on the clinical aspects; for example, actionable path-planning techniques can present individualized intervention plans to efficiently improve outcomes such as hypertension. Despite considerable progress in this field, challenges remain, such as the need for the prospective validation of AI-driven interventions and the development of comprehensive systems that integrate multiple AI methods. Future research should focus on addressing these challenges and refining the AI models to ensure their practical applicability in real-world clinical settings. Furthermore, the implementation of interdisciplinary collaborations among AI experts, clinicians, and healthcare providers are crucial to further optimizing and validate AI-driven solutions for hypertension management.

Understanding the performance of RHO type zeolite membrane for CH4/N2 separation based on molecular dynamics and deep neural network methods

This study shows a molecular dynamics (MD) simulation study on the performance of the RHO zeolite membrane for separating nitrogen from methane/nitrogen gas mixtures. The contamination of natural gas, predominantly composed of methane, with nitrogen diminishes its value. Zeolite membranes offer promising prospects for gas separation due to their stability, rigid pore structure, and molecular sieving properties. The study investigates the impact of pressure difference (up to 30 MPa), feed composition, and membrane thickness on the separation rate at a system temperature of 298 K. Results demonstrate that the RHO zeolite membrane exhibits high permeability and selectivity for N2 separation, surpassing the upper limit defined by Robson with a maximum permeability of 2.14 × 105 GPU (Gas Permeation Units). Exceptional selectivity of N2 over CH4 molecules is observed. Additionally, altering the feed composition and membrane thickness positively influences the membrane's separation performance, thereby enhancing its efficiency. The findings contribute to the advancement of separation technologies, providing valuable insights into the potential application of zeolite membranes for efficient N2 separation from CH4/N2 gas mixtures in natural gas processing. Furthermore, the study explores the use of Deep Neural Network (DNN) models to predict the membrane's performance under diverse operating conditions. The DNN models, trained using simulation data from MD simulations, exhibit high accuracy with a coefficient of determination (R2) exceeding 0.9, ensuring reliable predictions. The integration of DNN models facilitates the optimization of zeolite membrane-based gas separation systems, improving their design and operation.

Interpreting drug synergy in breast cancer with deep learning using target-protein inhibition profiles

BACKGROUND

Breast cancer is the most common malignancy among women worldwide. Despite advances in treating breast cancer over the past decades, drug resistance and adverse effects remain challenging. Recent therapeutic progress has shifted toward using drug combinations for better treatment efficiency. However, with a growing number of potential small-molecule cancer inhibitors, in silico strategies to predict pharmacological synergy before experimental trials are required to compensate for time and cost restrictions. Many deep learning models have been previously proposed to predict the synergistic effects of drug combinations with high performance. However, these models heavily relied on a large number of drug chemical structural fingerprints as their main features, which made model interpretation a challenge.

RESULTS

This study developed a deep neural network model that predicts synergy between small-molecule pairs based on their inhibitory activities against 13 selected key proteins. The synergy prediction model achieved a Pearson correlation coefficient between model predictions and experimental data of 0.63 across five breast cancer cell lines. BT-549 and MCF-7 achieved the highest correlation of 0.67 when considering individual cell lines. Despite achieving a moderate correlation compared to previous deep learning models, our model offers a distinctive advantage in terms of interpretability. Using the inhibitory activities against key protein targets as the main features allowed a straightforward interpretation of the model since the individual features had direct biological meaning. By tracing the synergistic interactions of compounds through their target proteins, we gained insights into the patterns our model recognized as indicative of synergistic effects.

CONCLUSIONS

The framework employed in the present study lays the groundwork for future advancements, especially in model interpretation. By combining deep learning techniques and target-specific models, this study shed light on potential patterns of target-protein inhibition profiles that could be exploited in breast cancer treatment.

Highly accurate and effective deep neural networks in pathological diagnosis of prostate cancer

PURPOSE

To established an AI system to make the pathological diagnosis of prostate cancer.

METHODS

Prostate histopathological whole mount (WM) sections from patients underwent robot-assisted laparoscopic prostatectomy were prepared. All the prostate WM pathological sections were converted to digital image data and marked with different colors on the basis of the ISUP Gleason grade group. The image was then fed into a segmentation algorithm. We chose modified U-Net as our fundamental network architecture.

RESULTS

172 patients were involved in this study. 896 pieces of prostate WM pathological sections from 160 patients, in which 826 pieces of WM sections from 148 patients were assigned to the training set randomly. After image segmentation there were totally 2,138,895 patches, of which 1,646,535 patches were valid for training. The other WM section was arranged for testing. Based on the whole image testing, AI and pathologists presented the same answers among 21 of 22 pieces of sections. To evaluate the diagnostic results at the pixel level, we anticipated correct cancer or non-cancer diagnose from this AI system. The area under the ROC curve as 96.8%. The value of pixel accuracy of three methods (binary analysis, clinically oriented analysis and analysis for different ISUP Gleason grade) were 96.93%, 95.43% and 93.88%, respectively. The value of frequency weighted IoU were 94.32%, 92.13% and 90.21%, respectively.

CONCLUSIONS

This AI system is able to assist pathologists to make a final diagnosis, indicating the great potential and a wide-range of applications of AI in the medical field.

Geospatial prediction of total soil carbon in European agricultural land based on deep learning

Accurate geospatial prediction of soil parameters provides a basis for large-scale digital soil mapping, making efficient use of the expensive and time-consuming process of field soil sampling. To date, few studies have used deep learning for geospatial prediction of soil parameters, but there is evidence that it may provide higher accuracy compared to machine learning methods. To address this research gap, this study proposed a deep neural network (DNN) for geospatial prediction of total soil carbon (TC) in European agricultural land and compared it with the eight most commonly used machine learning methods based on studies indexed in the Web of Science Core Collection. A total of 6209 preprocessed soil samples from the Geochemical mapping of agricultural and grazing land soil (GEMAS) dataset in heterogeneous agricultural areas covering 4,899,602 km2 in Europe were used. Prediction was performed based on 96 environmental covariates from climate and remote sensing sources, with extensive comprehensive hyperparameter tuning for all evaluated methods. DNN outperformed all evaluated machine learning methods (R2 = 0.663, RMSE = 9.595, MAE = 5.565), followed by Quantile Random Forest (QRF) (R2 = 0.635, RMSE = 25.993, MAE = 22.081). The ability of DNN to accurately predict small TC values and thus produce relatively low absolute residuals was a major reason for the higher prediction accuracy compared to machine learning methods. Climate parameters were the main factors in the achieved prediction accuracy, with 23 of the 25 environmental covariates with the highest variable importance being climate or land surface temperature parameters. These results demonstrate the superiority of DNN over machine learning methods for TC prediction, while highlighting the need for more recent soil sampling to assess the impact of climate change on TC content in European agricultural land.

RUBICON: a framework for designing efficient deep learning-based genomic basecallers

Nanopore sequencing generates noisy electrical signals that need to be converted into a standard string of DNA nucleotide bases using a computational step called basecalling. The performance of basecalling has critical implications for all later steps in genome analysis. Therefore, there is a need to reduce the computation and memory cost of basecalling while maintaining accuracy. We present RUBICON, a framework to develop efficient hardware-optimized basecallers. We demonstrate the effectiveness of RUBICON by developing RUBICALL, the first hardware-optimized mixed-precision basecaller that performs efficient basecalling, outperforming the state-of-the-art basecallers. We believe RUBICON offers a promising path to develop future hardware-optimized basecallers.

DeepMPTB: a vaginal microbiome-based deep neural network as artificial intelligence strategy for efficient preterm birth prediction

In recent decades, preterm birth (PTB) has become a significant research focus in the healthcare field, as it is a leading cause of neonatal mortality worldwide. Using five independent study cohorts including 1290 vaginal samples from 561 pregnant women who delivered at term (n = 1029) or prematurely (n = 261), we analysed vaginal metagenomics data for precise microbiome structure characterization. Then, a deep neural network (DNN) was trained to predict term birth (TB) and PTB with an accuracy of 84.10% and an area under the receiver operating characteristic curve (AUROC) of 0.875 ± 0.11. During a benchmarking process, we demonstrated that our DL model outperformed seven currently used machine learning algorithms. Finally, our results indicate that overall diversity of the vaginal microbiota should be taken in account to predict PTB and not specific species. This artificial-intelligence based strategy should be highly helpful for clinicians in predicting preterm birth risk, allowing personalized assistance to address various health issues. DeepMPTB is open source and free for academic use. It is licensed under a GNU Affero General Public License 3.0 and is available at https://deepmptb.streamlit.app/ . Source code is available at https://github.com/oschakoory/DeepMPTB and can be easily installed using Docker ( https://www.docker.com/ ).

Explainable deep-neural-network supported scheme for tuberculosis detection from chest radiographs

Chest radiographs are examined in typical clinical settings by competent physicians for tuberculosis diagnosis. However, this procedure is time consuming and subjective. Due to the growing usage of machine learning techniques in applied sciences, researchers have begun applying comparable concepts to medical diagnostics, such as tuberculosis screening. In the period of extremely deep neural nets which comprised of hundreds of convolution layers for feature extraction, we create a shallow-CNN for screening of TB condition from Chest X-rays so that the model is able to offer appropriate interpretation for right diagnosis. The suggested model consists of four convolution-maxpooling layers with various hyperparameters that were optimized for optimal performance using a Bayesian optimization technique. The model was reported with a peak classification accuracy, F1-score, sensitivity and specificity of 0.95. In addition, the receiver operating characteristic (ROC) curve for the proposed shallow-CNN showed a peak area under the curve value of 0.976. Moreover, we have employed class activation maps (CAM) and Local Interpretable Model-agnostic Explanations (LIME), explainer systems for assessing the transparency and explainability of the model in comparison to a state-of-the-art pre-trained neural net such as the DenseNet.

Differential effects of intrinsic properties of natural scenes and interference mechanisms on recognition processes in long-term visual memory

Humans display remarkable long-term visual memory (LTVM) processes. Even though images may be intrinsically memorable, the fidelity of their visual representations, and consequently the likelihood of successfully retrieving them, hinges on their similarity when concurrently held in LTVM. In this debate, it is still unclear whether intrinsic features of images (perceptual and semantic) may be mediated by mechanisms of interference generated at encoding, or during retrieval, and how these factors impinge on recognition processes. In the current study, participants (32) studied a stream of 120 natural scenes from 8 semantic categories, which varied in frequencies (4, 8, 16 or 32 exemplars per category) to generate different levels of category interference, in preparation for a recognition test. Then they were asked to indicate which of two images, presented side by side (i.e. two-alternative forced-choice), they remembered. The two images belonged to the same semantic category but varied in their perceptual similarity (similar or dissimilar). Participants also expressed their confidence (sure/not sure) about their recognition response, enabling us to tap into their metacognitive efficacy (meta-d'). Additionally, we extracted the activation of perceptual and semantic features in images (i.e. their informational richness) through deep neural network modelling and examined their impact on recognition processes. Corroborating previous literature, we found that category interference and perceptual similarity negatively impact recognition processes, as well as response times and metacognitive efficacy. Moreover, images semantically rich were less likely remembered, an effect that trumped a positive memorability boost coming from perceptual information. Critically, we did not observe any significant interaction between intrinsic features of images and interference generated either at encoding or during retrieval. All in all, our study calls for a more integrative understanding of the representational dynamics during encoding and recognition enabling us to form, maintain and access visual information.

Ultrahigh-porosity Ranunculus-like MgO adsorbent coupled with predictive deep belief networks: A transformative method for phosphorus treatment

Phosphorus is a nonrenewable material with a finite supply on Earth; however, due to the rapid growth of the manufacturing industry, phosphorus contamination has become a global concern. Therefore, this study highlights the remarkable potential of ranunculus-like MgO (MO4-MO6) as superior adsorbents for phosphate removal and recovery. Furthermore, MO6 stands out with an impressive adsorption capacity of 596.88 mg/g and a high efficacy across a wide pH range (2-10) under varying coexisting ion concentrations. MO6 outperforms the top current adsorbents for phosphate removal. The process follows Pseudo-second-order and Langmuir models, indicating chemical interactions between the phosphate species and homogeneous MO6 monolayer. MO6 maintains 80 % removal and 96 % recovery after five cycles and adheres to the WHO and EUWFD regulations for residual elements in water. FT-IR and XPS analyses further reveal the underlying mechanisms, including ion exchange, electrostatic, and acid-base interactions. Ten machine learning (ML) models were applied to simultaneously predict multi-criteria (sorption capacity, removal efficiency, final pH, and Mg leakage) affected by 15 diverse environmental conditions. Traditional ML models and deep neural networks have poor accuracy, particularly for removal efficiency. However, a breakthrough was achieved by the developed deep belief network (DBN) with unparalleled performance (MAE = 1.3289, RMSE = 5.2552, R2 = 0.9926) across all output features, surpassing all current studies using thousands of data points for only one output factor. These captivating MO6 and DBN models also have immense potential for effectively applying in the real water test with error < 5 %, opening immense horizons for transformative methods, particularly in phosphate removal and recovery.

A computational investigation of DMSO/water separation through functionalized GO multilayer nanosheet membrane using molecular dynamics simulation and deep neural network model for membrane performance prediction

In this molecular dynamics (MD) simulation study, the separation of dimethyl sulfoxide (DMSO) from water was investigated using multilayer functionalized graphene oxide (GO) membranes. The GO nanosheets were modified with chemical groups (-F, -H) to alter their properties. The study analyzed the influence of pressure and functional groups on the separation rate. Additionally, a deep neural network (DNN) model was developed to predict membrane behavior under different conditions in water treatment processes. Results revealed that the fluorine-functionalized membrane exhibited higher permeation compared to the hydrogen-functionalized one, with potential of mean force (PMF) analysis indicating higher energy barriers for water molecules passing through the hydrogen-functionalized membrane. The study used density profile, water density map analysis, and radial distribution function (RDF) analysis to understand water and DMSO molecule interactions. The diffusion coefficient of water molecules was also calculated, showing higher diffusion in the fluorine-functionalized system. Overall, the findings suggest that functionalized GO membranes are effective for DMSO-water separation, with the fluorine-functionalized membrane showing superior performance. The DNN model accurately predicts membrane behavior, contributing to the optimization of membrane separation systems.

Deep Learning Algorithm Based on Molecular Fingerprint for Prediction of Drug-Induced Liver Injury

Drug-induced liver injury (DILI) is one the rare adverse drug reaction (ADR) and multifactorial endpoints. Current preclinical animal models struggle to anticipate it, and in silico methods have emerged as a way with significant potential for doing so. In this study, a high-quality dataset of 1573 compounds was assembled. The 48 classification models, which depended on six different molecular fingerprints, were built via deep neural network (DNN) and seven machine learning algorithms. Comparing the results of the DNN and machine learning models, the optional performing model was found as the one developed based on the DNN with ECFP_6 as input, which achieved the area under the receiver operating characteristic curve (AUC) of 0.713, balanced accuracy (BA) of 0.680, and F1 of 0.753. In addition, we used the SHapley Additive exPlanations (SHAP) algorithm to interpret the models, identified the crucial structural fragments related to DILI risk, and selected the top ten substructures with the highest contribution rankings to serve as warning indicators for subsequent drug hepatotoxicity screening studies. The study demonstrates that the DNN models developed based on molecular fingerprints can be a trustworthy and efficient tool for determining the risk of DILI during the pre-development of novel medications.