Design high-entropy carbide ceramics from machine learning

High-entropy ceramics (HECs) have shown great application potential under demanding conditions, such as high stresses and temperatures. However, the immense phase space poses great challenges for the rational design of new high-performance HECs. In this work, we develop machine-learning (ML) models to discover high-entropy ceramic carbides (HECCs). Built upon attributes of HECCs and their constituent precursors, our ML models demonstrate a high prediction accuracy (0.982). Using the well-trained ML models, we evaluate the single-phase probability of 90 HECCs that are not experimentally reported so far. Several of these predictions are validated by our experiments. We further establish the phase diagrams for non-equiatomic HECCs spanning the whole composition space by which the single-phase regime can be easily identified. Our ML models can predict both equiatomic and non-equiatomic HECs based solely on the chemical descriptors of constituent transition-metal-carbide precursors, which paves the way for the high-throughput design of HECCs with superior properties.

A Combination of Chest Radiography and Estimated Plasma Volume May Predict In-Hospital Mortality in Acute Heart Failure

Background: Patients with heart failure (HF) often display dyspnea associated with pulmonary congestion, along with intravascular congestion, both may result in urgent hospitalization and subsequent death. A combination of radiographic pulmonary congestion and plasma volume might screen patients with a high risk of in-hospital mortality in the emergency department (ED).Methods: In the pathway of dyspneic patients in emergency (PARADISE) cohort, patients admitted for acute HF were stratified into 4 groups based on high or low congestion score index (CSI, ranging from 0 to 3, high value indicating severe congestion) and estimated plasma volume status (ePVS) calculated from hemoglobin/hematocrit.Results: In a total of 252 patients (mean age, 81.9 years; male, 46.8%), CSI and ePVS were not correlated (Spearman rho <0 .10, p > 0.10). High CSI/high ePVS was associated with poorer renal function, but clinical congestion markers (i.e., natriuretic peptide) were comparable across CSI/ePVS categories. High CSI/high ePVS was associated with a four-fold higher risk of in-hospital mortality (adjusted-OR, 95%CI = 4.20, 1.10-19.67) compared with low CSI/low ePVS, whereas neither high CSI nor ePVS alone was associated with poor prognosis (all-p-value > 0.10; Pinteraction = 0.03). High CSI/high ePVS improved a routine risk model (i.e., natriuretic peptide and lactate)(NRI = 46.9%, p = 0.02), resulting in high prediction of risk of in-hospital mortality (AUC = 0.85, 0.82-0.89).Conclusion: In patients hospitalized for acute HF with relatively old age and comorbidity burdens, a combination of CSI and ePVS was associated with a risk of in-hospital death, and improved prognostic performance on top of a conventional risk model.

Personally Tailored Survival Prediction of Patients With Follicular Lymphoma Using Machine Learning Transcriptome-Based Models

Follicular Lymphoma (FL) has a 10-year mortality rate of 20%, and this is mostly related to lymphoma progression and transformation to higher grades. In the era of personalized medicine it has become increasingly important to provide patients with an optimal prediction about their expected outcomes. The objective of this work was to apply machine learning (ML) tools on gene expression data in order to create individualized predictions about survival in patients with FL. Using data from two different studies, we were able to create a model which achieved good prediction accuracies in both cohorts (c-indexes of 0.793 and 0.662 in the training and test sets). Integration of this model with m7-FLIPI and age rendered high prediction accuracies in the test set (cox c-index 0.79), and a simplified approach identified 4 groups with remarkably different outcomes in terms of survival. Importantly, one of the groups comprised 27.35% of patients and had a median survival of 4.64 years. In summary, we have created a gene expression-based individualized predictor of overall survival in FL that can improve the predictions of the m7-FLIPI score.

Disentangling direct from indirect relationships in association networks

Networks are vital tools for understanding and modeling interactions in complex systems in science and engineering, and direct and indirect interactions are pervasive in all types of networks. However, quantitatively disentangling direct and indirect relationships in networks remains a formidable task. Here, we present a framework, called iDIRECT (Inference of Direct and Indirect Relationships with Effective Copula-based Transitivity), for quantitatively inferring direct dependencies in association networks. Using copula-based transitivity, iDIRECT eliminates/ameliorates several challenging mathematical problems, including ill-conditioning, self-looping, and interaction strength overflow. With simulation data as benchmark examples, iDIRECT showed high prediction accuracies. Application of iDIRECT to reconstruct gene regulatory networks in Escherichia coli also revealed considerably higher prediction power than the best-performing approaches in the DREAM5 (Dialogue on Reverse Engineering Assessment and Methods project, #5) Network Inference Challenge. In addition, applying iDIRECT to highly diverse grassland soil microbial communities in response to climate warming showed that the iDIRECT-processed networks were significantly different from the original networks, with considerably fewer nodes, links, and connectivity, but higher relative modularity. Further analysis revealed that the iDIRECT-processed network was more complex under warming than the control and more robust to both random and target species removal (P < 0.001). As a general approach, iDIRECT has great advantages for network inference, and it should be widely applicable to infer direct relationships in association networks across diverse disciplines in science and engineering.

GLAM: An adaptive graph learning method for automated molecular interactions and properties predictions

Improving drug discovery efficiency is a core and long-standing challenge in drug discovery. For this purpose, many graph learning methods have been developed to search potential drug candidates with fast speed and low cost. In fact, the pursuit of high prediction performance on a limited number of datasets has crystallized them, making them lose advantage in repurposing to new data generated in drug discovery. Here we propose a flexible method that can adapt to any dataset and make accurate predictions. The proposed method employs an adaptive pipeline to learn from a dataset and output a predictor. Without any manual intervention, the method achieves far better prediction performance on all tested datasets than traditional methods, which are based on hand-designed neural architectures and other fixed items. In addition, we found that the proposed method is more robust than traditional methods and can provide meaningful interpretability. Given the above, the proposed method can serve as a reliable method to predict molecular interactions and properties with high adaptability, performance, robustness and interpretability. This work would take a solid step forward to the purpose of aiding researchers to design better drugs with high efficiency.

U.S. Unemployment Rate Prediction by Economic Indices in the COVID-19 Pandemic Using Neural Network, Random Forest, and Generalized Linear Regression

Artificial neural network (ANN) has been showing its superior capability of modeling and prediction. Neural network model is capable of incorporating high dimensional data, and the model is significantly complex statistically. Sometimes, the complexity is treated as a Blackbox. However, due to the model complexity, the model is capable of capture and modeling an extensive number of patterns, and the prediction power is much stronger than traditional statistical models. Random forest algorithm is a combination of classification and regression trees, using bootstrap to randomly train the model from a set of data (called training set) and test the prediction by a testing set. Random forest has high prediction speed, moderate variance, and does not require any rescaling or transformation of the dataset. This study validates the relationship between the U.S. unemployment rate and economic indices during the COVID-19 pandemic and constructs three different predictive modeling for unemployment rate by economic indices through neural network, random forest, and generalized linear regression model.

Identification of Metro-Bikeshare Transfer Trip Chains by Matching Docked Bikeshare and Metro Smartcards

Metro-bikeshare integration, an important way of improving the efficiency of public transportation, has grown rapidly during the last decades in many countries. However, most previous analysis of metro-bikeshare transfer trips were based on limited sample size and the number of recognized metro-bikeshare trips were not sufficient. The primary objective of this study is to derive a method to recognize metro-bikeshare transfer trips. The two data sources are provided by Nanjing Metro Company and Nanjing Public Bicycle Company over the same period from 9–29 March 2016. The identifying method includes three steps: (1) Matching Card Pairs (2) Filtering Card Pairs and (3) Identifying Card Pairs. The case study indicates that the Support Vector Classification (SVC) performs best with a high prediction accuracy of 95.9% using seamless smartcards. The identifying method is then used to recognize the transfer trips from other types of cards, resulting in 17,022 valid metro-bikeshare transfer trips made by 2948 travelers. Finally, travel patterns extracted from the two groups of identified transfer trips are analyzed comparatively. The method proposed presents new opportunities for analyzing metro-bikeshare transfer trip characteristics.

Colour Quality Evaluation of Bluish-Green Serpentinite Based on the CIECAM16 Model

The CIECAM16 colour appearance model is currently a model with high prediction accuracy. It can solve the problem of predicting the influence of different observation conditions on the colour of gemstones. In this study, a computer vision system (CVS) was used to measure the colour of 59 bluish-green serpentinite samples, and the tristimulus values were input into the CIECAM16 forward model to calculate the colour appearance parameters of serpentinite under different surrounds, illuminances, and light sources. It was found that the darkening of the surround causes the lightness and brightness to increase. Pearson’s r of brightness and colourfulness with illuminance is 0.885 and 0.332, respectively, which predicts the Stevens and Hunt effects. When the light source changes from D65 to A, the calculated hue angle shifts to the complementary area of the A light source, which is contrary to the CVS measurement result. The D65 light source is more suitable for the colour presentation and classification of bluish-green serpentinite.

Bagging Based Cross-Media Retrieval Algorithm

It is very challenging to propose a strong learning algorithm with high prediction accuracy of cross-media retrieval, while finding a weak learning algorithm which is slightly higher than that of random prediction is very easy. Inspired by this idea, we propose an imaginative Bagging based cross-media retrieval algorithm (called BCMR) in this paper. First, we utilize bootstrap sampling to carry out random sampling of the original training set. The amount of the sample abstracted by bootstrap is set to be same as the original dataset. Second, 50 bootstrap replicates are used for training 50 weak classifiers independently. We take advantage of homogenous individual classifiers and integrate eight different baseline methods in our experiments. Finally, we generate the final strong classifier from the 50 weak classifiers by the integration strategy of sample voting. We use collective wisdom to eliminate bad decisions so that the generalization ability of the integrated model could be greatly enhanced. Extensive experiments performed on three datasets show that BCMR can effectively improve the accuracy of cross-media retrieval.

Crude Oil Price Forecast Based on Deep Transfer Learning: Shanghai Crude Oil as an Example

Crude oil is an important fuel resource for all countries. Accurate predictions of oil prices have important economic and social values. However, the price of crude oil is highly nonlinear under the influence of many factors, so it is very difficult to predict accurately. Shanghai crude oil futures were officially listed in March 2018. It is of great significance to accurately predict the price of Shanghai crude oil futures for guiding China’s domestic production practice. Forecasting the price of Shanghai crude oil futures is even more difficult because of the lack of price data due to the short listing time. In order to solve this problem, this paper proposes using Long Short-Term Memory Network (LSTM) based on transfer learning to predict the price of crude oil in Shanghai. The basic idea is to take advantage of the correlation between Brent crude oil and Shanghai crude oil, use Brent crude oil for training in the early stage, and then use Shanghai crude oil to fine-tune the network. The empirical results show that the LSTM model based on transfer learning has strong generalization ability and high prediction accuracy.