Deep learning assessment of left ventricular hypertrophy based on electrocardiogram.

Background: Current electrocardiogram (ECG) criteria of left ventricular hypertrophy (LVH) have low sensitivity. Deep learning (DL) techniques have been widely used to detect cardiac diseases due to its ability of automatic feature extraction of ECG. However, DL was rarely applied in LVH diagnosis. Our study aimed to construct a DL model for rapid and effective detection of LVH using 12-lead ECG. Methods: We built a DL model based on convolutional neural network-long short-term memory (CNN-LSTM) to detect LVH using 12-lead ECG. The echocardiogram and ECG of 1,863 patients obtained within 1 week after hospital admission were analyzed. Patients were evenly allocated into 3 sets at 3:1:1 ratio: the training set (n = 1,120), the validation set (n = 371) and the test set 1 (n = 372). In addition, we recruited 453 hospitalized patients into the internal test set 2. Different DL model of each subgroup was developed according to gender and relative wall thickness (RWT). Results: The LVH was predicted by the CNN-LSTM model with an area under the curve (AUC) of 0.62 (sensitivity 68%, specificity 57%) in the test set 1, which outperformed Cornell voltage criteria (AUC: 0.57, sensitivity 48%, specificity 72%) and Sokolow-Lyon voltage (AUC: 0.51, sensitivity 14%, specificity 96%). In the internal test set 2, the CNN-LSTM model had a stable performance in predicting LVH with an AUC of 0.59 (sensitivity 65%, specificity 57%). In the subgroup analysis, the CNN-LSTM model predicted LVH by 12-lead ECG with an AUC of 0.66 (sensitivity 72%, specificity 60%) for male patients, which performed better than that for female patients (AUC: 0.59, sensitivity 50%, specificity 71%). Conclusion: Our study established a CNN-LSTM model to diagnose LVH by 12-lead ECG with higher sensitivity than current ECG diagnostic criteria. This CNN-LSTM model may be a simple and effective screening tool of LVH.

GREMA: modelling of emulated gene regulatory networks with confidence levels based on evolutionary intelligence to cope with the underdetermined problem.

MOTIVATION: Non-linear ordinary differential equation (ODE) models that contain numerous parameters are suitable for inferring an emulated gene regulatory network (eGRN). However, the number of experimental measurements is usually far smaller than the number of parameters of the eGRN model that leads to an underdetermined problem. There is no unique solution to the inference problem for an eGRN using insufficient measurements. RESULTS: This work proposes an evolutionary modelling algorithm (EMA) that is based on evolutionary intelligence to cope with the underdetermined problem. EMA uses an intelligent genetic algorithm to solve the large-scale parameter optimization problem. An EMA-based method, GREMA, infers a novel type of gene regulatory network with confidence levels for every inferred regulation. The higher the confidence level is, the more accurate the inferred regulation is. GREMA gradually determines the regulations of an eGRN with confidence levels in descending order using either an S-system or a Hill function-based ODE model. The experimental results showed that the regulations with high-confidence levels are more accurate and robust than regulations with low-confidence levels. Evolutionary intelligence enhanced the mean accuracy of GREMA by 19.2% when using the S-system model with benchmark datasets. An increase in the number of experimental measurements may increase the mean confidence level of the inferred regulations. GREMA performed well compared with existing methods that have been previously applied to the same S-system, DREAM4 challenge and SOS DNA repair benchmark datasets. AVAILABILITY AND IMPLEMENTATION: All of the datasets that were used and the GREMA-based tool are freely available at https://nctuiclab.github.io/GREMA. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

PredCRP: predicting and analysing the regulatory roles of CRP from its binding sites in Escherichia coli.

Cyclic AMP receptor protein (CRP), a global regulator in Escherichia coli, regulates more than 180 genes via two roles: activation and repression. Few methods are available for predicting the regulatory roles from the binding sites of transcription factors. This work proposes an accurate method PredCRP to derive an optimised model (named PredCRP-model) and a set of four interpretable rules (named PredCRP-ruleset) for predicting and analysing the regulatory roles of CRP from sequences of CRP-binding sites. A dataset consisting of 169 CRP-binding sites with regulatory roles strongly supported by evidence was compiled. The PredCRP-model, using 12 informative features of CRP-binding sites, and cooperating with a support vector machine achieved a training and test accuracy of 0.98 and 0.93, respectively. PredCRP-ruleset has two activation rules and two repression rules derived using the 12 features and the decision tree method C4.5. This work further screened and identified 23 previously unobserved regulatory interactions in Escherichia coli. Using quantitative PCR for validation, PredCRP-model and PredCRP-ruleset achieved a test accuracy of 0.96 (=22/23) and 0.91 (=21/23), respectively. The proposed method is suitable for designing predictors for regulatory roles of all global regulators in Escherichia coli. PredCRP can be accessed at https://github.com/NctuICLab/PredCRP .

ESA-UbiSite: accurate prediction of human ubiquitination sites by identifying a set of effective negatives.

Motivation: Numerous ubiquitination sites remain undiscovered because of the limitations of mass spectrometry-based methods. Existing prediction methods use randomly selected non-validated sites as non-ubiquitination sites to train ubiquitination site prediction models. Results: We propose an evolutionary screening algorithm (ESA) to select effective negatives among non-validated sites and an ESA-based prediction method, ESA-UbiSite, to identify human ubiquitination sites. The ESA selects non-validated sites least likely to be ubiquitination sites as training negatives. Moreover, the ESA and ESA-UbiSite use a set of well-selected physicochemical properties together with a support vector machine for accurate prediction. Experimental results show that ESA-UbiSite with effective negatives achieved 0.92 test accuracy and a Matthews's correlation coefficient of 0.48, better than existing prediction methods. The ESA increased ESA-UbiSite's test accuracy from 0.75 to 0.92 and can improve other post-translational modification site prediction methods. Availability and Implementation: An ESA-UbiSite-based web server has been established at http://iclab.life.nctu.edu.tw/iclab_webtools/ESAUbiSite/ . Contact: syho@mail.nctu.edu.tw. Supplementary information: Supplementary data are available at Bioinformatics online.

Prediction of linear B-cell epitopes of hepatitis C virus for vaccine development.

BACKGROUND: High genetic heterogeneity in the hepatitis C virus (HCV) is the major challenge of the development of an effective vaccine. Existing studies for developing HCV vaccines have mainly focused on T-cell immune response. However, identification of linear B-cell epitopes that can stimulate B-cell response is one of the major tasks of peptide-based vaccine development. Owing to the variability in B-cell epitope length, the prediction of B-cell epitopes is much more complex than that of T-cell epitopes. Furthermore, the motifs of linear B-cell epitopes in different pathogens are quite different (e. g. HCV and hepatitis B virus). To cope with this challenge, this work aims to propose an HCV-customized sequence-based prediction method to identify B-cell epitopes of HCV. RESULTS: This work establishes an experimentally verified dataset comprising the B-cell response of HCV dataset consisting of 774 linear B-cell epitopes and 774 non B-cell epitopes from the Immune Epitope Database. An interpretable rule mining system of B-cell epitopes (IRMS-BE) is proposed to select informative physicochemical properties (PCPs) and then extracts several if-then rule-based knowledge for identifying B-cell epitopes. A web server Bcell-HCV was implemented using an SVM with the 34 informative PCPs, which achieved a training accuracy of 79.7% and test accuracy of 70.7% better than the SVM-based methods for identifying B-cell epitopes of HCV and the two general-purpose methods. This work performs advanced analysis of the 34 informative properties, and the results indicate that the most effective property is the alpha-helix structure of epitopes, which influences the connection between host cells and the E2 proteins of HCV. Furthermore, 12 interpretable rules are acquired from top-five PCPs and achieve a sensitivity of 75.6% and specificity of 71.3%. Finally, a conserved promising vaccine candidate, PDREMVLYQE, is identified for inclusion in a vaccine against HCV. CONCLUSIONS: This work proposes an interpretable rule mining system IRMS-BE for extracting interpretable rules using informative physicochemical properties and a web server Bcell-HCV for predicting linear B-cell epitopes of HCV. IRMS-BE may also apply to predict B-cell epitopes for other viruses, which benefits the improvement of vaccines development of these viruses without significant modification. Bcell-HCV is useful for identifying B-cell epitopes of HCV antigen to help vaccine development, which is available at http://e045.life.nctu.edu.tw/BcellHCV.

Characterizing informative sequence descriptors and predicting binding affinities of heterodimeric protein complexes.

BACKGROUND: Protein-protein interactions (PPIs) are involved in various biological processes, and underlying mechanism of the interactions plays a crucial role in therapeutics and protein engineering. Most machine learning approaches have been developed for predicting the binding affinity of protein-protein complexes based on structure and functional information. This work aims to predict the binding affinity of heterodimeric protein complexes from sequences only. RESULTS: This work proposes a support vector machine (SVM) based binding affinity classifier, called SVM-BAC, to classify heterodimeric protein complexes based on the prediction of their binding affinity. SVM-BAC identified 14 of 580 sequence descriptors (physicochemical, energetic and conformational properties of the 20 amino acids) to classify 216 heterodimeric protein complexes into low and high binding affinity. SVM-BAC yielded the training accuracy, sensitivity, specificity, AUC and test accuracy of 85.80%, 0.89, 0.83, 0.86 and 83.33%, respectively, better than existing machine learning algorithms. The 14 features and support vector regression were further used to estimate the binding affinities (Pkd) of 200 heterodimeric protein complexes. Prediction performance of a Jackknife test was the correlation coefficient of 0.34 and mean absolute error of 1.4. We further analyze three informative physicochemical properties according to their contribution to prediction performance. Results reveal that the following properties are effective in predicting the binding affinity of heterodimeric protein complexes: apparent partition energy based on buried molar fractions, relations between chemical structure and biological activity in principal component analysis IV, and normalized frequency of beta turn. CONCLUSIONS: The proposed sequence-based prediction method SVM-BAC uses an optimal feature selection method to identify 14 informative features to classify and predict binding affinity of heterodimeric protein complexes. The characterization analysis revealed that the average numbers of beta turns and hydrogen bonds at protein-protein interfaces in high binding affinity complexes are more than those in low binding affinity complexes.

R3D-BLAST: a search tool for similar RNA 3D substructures.

R3D-BLAST is a BLAST-like search tool that allows the user to quickly and accurately search against the PDB for RNA structures sharing similar substructures with a specified query RNA structure. The basic idea behind R3D-BLAST is that all the RNA 3D structures deposited in the PDB are first encoded as 1D structural sequences using a structural alphabet of 23 distinct nucleotide conformations, and BLAST is then applied to these 1D structural sequences to search for those RNA substructures whose 1D structural sequences are similar to that of the query RNA substructure. R3D-BLAST takes as input an RNA 3D structure in the PDB format and outputs all substructures of the hits similar to that of the query with a graphical display to show their structural superposition. In addition, each RNA substructure hit found by R3D-BLAST has an associated E-value to measure its statistical significance. R3D-BLAST is now available online at http://genome.cs.nthu.edu.tw/R3D-BLAST/ for public access.