Maximum Matthews Correlation Coefficient Research Articles

Machine Learning Prediction of Fetal Health Status from Cardiotocography Examination in Developing Healthcare Contexts

Reducing neonatal mortality is a critical global health objective, especially in resource-constrained developing countries. This study employs machine learning (ML) techniques to predict fetal health status based on cardiotocography (CTG) examination findings, utilizing a dataset from the Kaggle repository due to the limited comprehensive healthcare data available in developing nations. Features such as baseline fetal heart rate, uterine contractions, and waveform characteristics were extracted using the RFE wrapper feature engineering technique and scaled with a standard scaler. Six ML models—Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Gradient Boosting (GB), Categorical Boosting (CB), and Extended Gradient Boosting (XGB)—are trained via cross-validation and evaluated using performance metrics. The developed models were trained via cross-validation and evaluated using ML performance metrics. Eight out of the 21 features selected by GB returned their maximum Matthews Correlation Coefficient (MCC) score of 0.6255, while CB, with 20 of the 21 features, returned the maximum and highest MCC score of 0.6321. The study demonstrated the ability of ML models to predict fetal health conditions from CTG exam results, facilitating early identification of high-risk pregnancies and enabling prompt treatment to prevent severe neonatal outcomes.

Globally vs. Locally Trained Machine Learning Models for Landslide Detection: A Case Study of a Glacial Landscape

Landslide risk mitigation is limited by data scarcity; however, this could be improved using continuous landslide detection systems. To investigate which image types and machine learning models are most useful for landslide detection in a Norwegian setting, we compared the performance of five different machine learning models, for the Jølster case study (30 July 2019), in Western Norway. These included three globally pre-trained models; (i) the continuous change detection and classification (CCDC) algorithm, (ii) a combined k-means clustering and random forest classification model, and (iii) a convolutional neural network (CNN), and two locally trained models, including; (iv) classification and regression Trees and (v) a U-net CNN model. Images used included Sentinel-1, Sentinel-2, as well as digital elevation model (DEM) and slope. The globally trained models performed poorly in shadowed areas and were all outperformed by the locally trained models. A maximum Matthew’s correlation coefficient (MCC) score of 89% was achieved with a CNN U-net deep learning model, using combined Sentinel-1 and -2 images as input. This is one of the first attempts to apply deep learning to detect landslides with both Sentinel-1 and -2 images. Using Sentinel-1 images only, the locally-trained deep-learning model significantly outperformed the conventional machine learning model. These findings contribute to developing a national continuous monitoring system for landslides.

The Relative Distance Prediction of Transmembrane Protein Surface Residue Based on Improved Residual Networks

(1) Background: Transmembrane proteins (TMPs) act as gateways connecting the intra- and extra-biomembrane environments, exchanging material and signals crossing the biofilm. Relevant evidence shows that corresponding interactions mostly happen on the TMPs’ surface. Therefore, knowledge of the relative distance among surface residues is critically helpful in discovering the potential local structural characters and setting the foundation for the protein’s interaction with other molecules. However, the prediction of fine-grained distances among residues with sequences remains challenging; (2) Methods: In this study, we proposed a deep-learning method called TMP-SurResD, which capitalized on the combination of the Residual Block (RB) and Squeeze-and-Excitation (SE) for simultaneously predicting the relative distance of functional surface residues based on sequences’ information; (3) Results: The comprehensive evaluation demonstrated that TMP-SurResD could successfully capture the relative distance between residues, with a Pearson Correlation Coefficient (PCC) of 0.7105 and 0.6999 on the validation and independent sets, respectively. In addition, TMP-SurResD outperformed other methods when applied to TMPs surface residue contact prediction, and the maximum Matthews Correlation Coefficient (MCC) reached 0.602 by setting a threshold to the predicted distance of 10; (4) Conclusions: TMP-SurResD can serve as a useful tool in supporting a sequence-based local structural feature construction and exploring the function and biological mechanisms of structure determination in TMPs, which can thus significantly facilitate the research direction of molecular drug action, target design, and disease treatment.

DE-PNN: Differential Evolution-Based Feature Optimization with Probabilistic Neural Network for Imbalanced Arrhythmia Classification

In this research, a heartbeat classification method is presented based on evolutionary feature optimization using differential evolution (DE) and classification using a probabilistic neural network (PNN) to discriminate between normal and arrhythmic heartbeats. The proposed method follows four steps: (1) preprocessing, (2) heartbeat segmentation, (3) DE feature optimization, and (4) PNN classification. In this method, we have employed direct signal amplitude points constituting the heartbeat acquired from the ECG holter device with no secondary feature extraction step usually used in case of hand-crafted, frequency transformation or other features. The heartbeat types include normal, left bundle branch block, right bundle branch block, premature ventricular contraction, atrial premature, ventricular escape, ventricular flutter and paced beat. Using ECG records from the MIT-BIH, heartbeats are identified to start at 250 ms before and end at 450 ms after the respective R-peak positions. In the next step, the DE method is applied to reduce and optimize the direct heartbeat features. Although complex and highly computational ECG heartbeat classification algorithms have been proposed in the literature, they failed to achieve high performance in detecting some minority heartbeat categories, especially for imbalanced datasets. To overcome this challenge, we propose an optimization step for the deep CNN model using a novel classification metric called the Matthews correlation coefficient (MCC). This function focuses on arrhythmia (minority) heartbeat classes by increasing their importance. Maximum MCC is used as a fitness function to identify the optimum combination of features for the uncorrelated and non-uniformly distributed eight beat class samples. The proposed DE-PNN scheme can provide better classification accuracy considering 8 classes with only 36 features optimized from a 253 element feature set implying an 85.77% reduction in direct amplitude features. Our proposed method achieved overall 99.33% accuracy, 94.56% F1, 93.84% sensitivity, and 99.21% specificity.

Using amino acid features to identify the pathogenicity of influenza B virus

BackgroundInfluenza B virus can cause epidemics with high pathogenicity, so it poses a serious threat to public health. A feature representation algorithm is proposed in this paper to identify the pathogenicity phenotype of influenza B virus.MethodsThe dataset included all 11 influenza virus proteins encoded in eight genome segments of 1724 strains. Two types of features were hierarchically used to build the prediction model. Amino acid features were directly delivered from 67 feature descriptors and input into the random forest classifier to output informative features about the class label and probabilistic prediction. The sequential forward search strategy was used to optimize the informative features. The final features for each strain had low dimensions and included knowledge from different perspectives, which were used to build the machine learning model for pathogenicity identification.ResultsThe 40 signature positions were achieved by entropy screening. Mutations at position 135 of the hemagglutinin protein had the highest entropy value (1.06). After the informative features were directly generated from the 67 random forest models, the dimensions for class and probabilistic features were optimized as 4 and 3, respectively. The optimal class features had a maximum accuracy of 94.2% and a maximum Matthews correlation coefficient of 88.4%, while the optimal probabilistic features had a maximum accuracy of 94.1% and a maximum Matthews correlation coefficient of 88.2%. The optimized features outperformed the original informative features and amino acid features from individual descriptors. The sequential forward search strategy had better performance than the classical ensemble method.ConclusionsThe optimized informative features had the best performance and were used to build a predictive model so as to identify the phenotype of influenza B virus with high pathogenicity and provide early risk warning for disease control.Graphical

Impact of Image Resolution on Deep Learning Performance in Endoscopy Image Classification: An Experimental Study Using a Large Dataset of Endoscopic Images.

Recent trials have evaluated the efficacy of deep convolutional neural network (CNN)-based AI systems to improve lesion detection and characterization in endoscopy. Impressive results are achieved, but many medical studies use a very small image resolution to save computing resources at the cost of losing details. Today, no conventions between resolution and performance exist, and monitoring the performance of various CNN architectures as a function of image resolution provides insights into how subtleties of different lesions on endoscopy affect performance. This can help set standards for image or video characteristics for future CNN-based models in gastrointestinal (GI) endoscopy. This study examines the performance of CNNs on the HyperKvasir dataset, consisting of 10,662 images from 23 different findings. We evaluate two CNN models for endoscopic image classification under quality distortions with image resolutions ranging from 32 × 32 to 512 × 512 pixels. The performance is evaluated using two-fold cross-validation and F1-score, maximum Matthews correlation coefficient (MCC), precision, and sensitivity as metrics. Increased performance was observed with higher image resolution for all findings in the dataset. MCC was achieved at image resolutions between 512 × 512 pixels for classification for the entire dataset after including all subclasses. The highest performance was observed with an MCC value of 0.9002 when the models were trained on the highest resolution and tested on the same resolution. Different resolutions and their effect on CNNs are explored. We show that image resolution has a clear influence on the performance which calls for standards in the field in the future.

Benchmarking Automatic Segmentation Algorithms Against Human Interobserver Variability of Prostate and Organs at Risk Delineation on Prostate MRI

<h3>Purpose/Objective(s)</h3> The objective of this study was to benchmark custom automatic segmentation software against human interobserver variability of prostate and OAR segmentation on 3D MRIs for prostate radiotherapy. We hypothesized that automatic segmentation algorithms can exhibit consistency comparable to human readers, including radiation oncologists. <h3>Materials/Methods</h3> Twenty-five patients underwent MRI-based treatment planning and quality assessment for low-dose-rate prostate brachytherapy (LDRPBT). All 25 patients received both treatment planning MRIs and postimplant quality assessment MRIs. The MRIs were acquired on 3T scanners with flexible external array coils and no endorectal coil. The preimplant MRIs were acquired with a 3D T2-weighted TSE sequence. The postimplant MRIs were acquired with a 3D T2/T1-weighted fully balanced SSFP sequence. Five human observers involved in the LDRPBT workflow, including 2 radiation oncologists, delineated the prostate and 4 OARs on the 50 MRIs. The OARs included the external urinary sphincter (EUS), seminal vesicles (SV), rectum, and bladder. Human interobserver variability (IoV) amongst the 5 human readers was quantified in a human observer study. Similarity metrics, including precision (P), recall (R), Jaccard index (JI), and Matthew's correlation coefficient (MCC), were quantified between unique observer pairs for all 5 organ segmentation masks. A custom deep learning-based autosegmentation algorithm was used to segment the prostate and 4 OARs simultaneously on the same 50 MRIs that were annotated by the human readers. P, R, JI, and MCC were evaluated at different probability thresholds. The comparisons were made against reference segmentation masks of the 5 organs computed from the 5 human segmentation masks using the simultaneous truth and performance level estimation algorithm. The human IoV similarity metrics were compared to the computer similarity metrics at the point of maximum MCC using two-tailed nonparametric Wilcoxon rank sum tests. <h3>Results</h3> For the preimplant MRIs, the computer similarity metrics (C) were lower than the human IoV similarity metrics (H) for all organs except the EUS (MCC, C vs. H: prostate – 0.868 vs. 0.903, <i>P</i> < 0.0001; EUS – 0.610 vs. 0.552, <i>P</i> = 0.1690; SV – 0.678 vs. 0.750, <i>P</i> = 0.0094; rectum – 0.790 vs. 0.886, <i>P</i> < 0.0001; bladder – 0.879 vs. 0.941, <i>P</i> < 0.0001). However, the opposite was observed for the postimplant MRIs; C was significantly higher than H for all 5 organs (MCC, C vs. H: prostate – 0.925 vs. 0.892, <i>P</i> < 0.0001; EUS – 0.743 vs. 0.620, <i>P</i> = 0.0003; SV – 0.849 vs. 0.760, <i>P</i> < 0.0001; rectum – 0.925 vs. 0.876, <i>P</i> < 0.0001; bladder – 0.972 vs. 0.938, <i>P</i> < 0.0001). <h3>Conclusion</h3> The autosegmentation algorithm produced high quality segmentation masks for both the preimplant and postimplant MRIs. The human readers were more consistent than the autosegmentation algorithm on the T2-weighted preimplant MRIs, but the autosegmentation algorithm was more consistent than the human readers on the postimplant MRIs.

Weekly dengue forecasts in Iquitos, Peru; San Juan, Puerto Rico; and Singapore.

BackgroundPredictive models can serve as early warning systems and can be used to forecast future risk of various infectious diseases. Conventionally, regression and time series models are used to forecast dengue incidence, using dengue surveillance (e.g., case counts) and weather data. However, these models may be limited in terms of model assumptions and the number of predictors that can be included. Machine learning (ML) methods are designed to work with a large number of predictors and thus offer an appealing alternative. Here, we compared the performance of ML algorithms with that of regression models in predicting dengue cases and outbreaks from 4 to up to 12 weeks in advance. Many countries lack sufficient health surveillance infrastructure, as such we evaluated the contribution of dengue surveillance and weather data on the predictive power of these models.MethodsWe developed ML, regression, and time series models to forecast weekly dengue case counts and outbreaks in Iquitos, Peru; San Juan, Puerto Rico; and Singapore from 1990–2016. Forecasts were generated using available weekly dengue surveillance, and weather data. We evaluated the agreement between model forecasts and actual dengue observations using Mean Absolute Error and Matthew’s Correlation Coefficient (MCC).ResultsFor near term predictions of weekly case counts and when using surveillance data, ML models had 21% and 33% less error than regression and time series models respectively. However, using weather data only, ML models did not demonstrate a practical advantage. When forecasting weekly dengue outbreaks 12 weeks in advance, ML models achieved a maximum MCC of 0.61.ConclusionsOur results identified 2 scenarios when ML models are advantageous over regression model: 1) predicting dengue weekly case counts 4 weeks ahead when dengue surveillance data are available and 2) predicting weekly dengue outbreaks 12 weeks ahead when dengue surveillance data are unavailable. Given the advantages of ML models, dengue early warning systems may be improved by the inclusion of these models.

AntiCP 2.0: an updated model for predicting anticancer peptides.

Increasing use of therapeutic peptides for treating cancer has received considerable attention of the scientific community in the recent years. The present study describes the in silico model developed for predicting and designing anticancer peptides (ACPs). ACPs residue composition analysis show the preference of A, F, K, L and W. Positional preference analysis revealed that residues A, F and K are favored at N-terminus and residues L and K are preferred at C-terminus. Motif analysis revealed the presence of motifs like LAKLA, AKLAK, FAKL and LAKL in ACPs. Machine learning models were developed using various input features and implementing different machine learning classifiers on two datasets main and alternate dataset. In the case of main dataset, dipeptide composition based ETree classifier model achieved maximum Matthews correlation coefficient (MCC) of 0.51 and 0.83 area under receiver operating characteristics (AUROC) on the training dataset. In the case of alternate dataset, amino acid composition based ETree classifier performed best and achieved the highest MCC of 0.80 and AUROC of 0.97 on the training dataset. Five-fold cross-validation technique was implemented for model training and testing, and their performance was also evaluated on the validation dataset. Best models were implemented in the webserver AntiCP 2.0, which is freely available at https://webs.iiitd.edu.in/raghava/anticp2/. The webserver is compatible with multiple screens such as iPhone, iPad, laptop and android phones. The standalone version of the software is available at GitHub; docker-based container also developed.

SAMbinder: A Web Server for Predicting S-Adenosyl-L-Methionine Binding Residues of a Protein From Its Amino Acid Sequence.

MotivationS-adenosyl-L-methionine (SAM) is an essential cofactor present in the biological system and plays a key role in many diseases. There is a need to develop a method for predicting SAM binding sites in a protein for designing drugs against SAM associated disease. To the best of our knowledge, there is no method that can predict the binding site of SAM in a given protein sequence.ResultThis manuscript describes a method SAMbinder, developed for predicting SAM interacting residue in a protein from its primary sequence. All models were trained, tested, and evaluated on 145 SAM binding protein chains where no two chains have more than 40% sequence similarity. Firstly, models were developed using different machine learning techniques on a balanced data set containing 2,188 SAM interacting and an equal number of non-interacting residues. Our random forest based model developed using binary profile feature got maximum Matthews Correlation Coefficient (MCC) 0.42 with area under receiver operating characteristics (AUROC) 0.79 on the validation data set. The performance of our models improved significantly from MCC 0.42 to 0.61, when evolutionary information in the form of the position-specific scoring matrix (PSSM) profile is used as a feature. We also developed models on a realistic data set containing 2,188 SAM interacting and 40,029 non-interacting residues and got maximum MCC 0.61 with AUROC of 0.89. In order to evaluate the performance of our models, we used internal as well as external cross-validation technique.Availability and Implementationhttps://webs.iiitd.edu.in/raghava/sambinder/.

Classification of early and late stage liver hepatocellular carcinoma patients from their genomics and epigenomics profiles.

BackgroundLiver Hepatocellular Carcinoma (LIHC) is one of the major cancers worldwide, responsible for millions of premature deaths every year. Prediction of clinical staging is vital to implement optimal therapeutic strategy and prognostic prediction in cancer patients. However, to date, no method has been developed for predicting the stage of LIHC from the genomic profile of samples.MethodsThe Cancer Genome Atlas (TCGA) dataset of 173 early stage (stage-I), 177 late stage (stage-II, Stage-III and stage-IV) and 50 adjacent normal tissue samples for 60,483 RNA transcripts and 485,577 methylation CpG sites, was extensively analyzed to identify the key transcriptomic expression and methylation-based features using different feature selection techniques. Further, different classification models were developed based on selected key features to categorize different classes of samples implementing different machine learning algorithms.ResultsIn the current study, in silico models have been developed for classifying LIHC patients in the early vs. late stage and cancerous vs. normal samples using RNA expression and DNA methylation data. TCGA datasets were extensively analyzed to identify differentially expressed RNA transcripts and methylated CpG sites that can discriminate early vs. late stages and cancer vs. normal samples of LIHC with high precision. Naive Bayes model developed using 51 features that combine 21 CpG methylation sites and 30 RNA transcripts achieved maximum MCC (Matthew’s correlation coefficient) 0.58 with an accuracy of 78.87% on the validation dataset in discrimination of early and late stage. Additionally, the prediction models developed based on 5 RNA transcripts and 5 CpG sites classify LIHC and normal samples with an accuracy of 96–98% and AUC (Area Under the Receiver Operating Characteristic curve) 0.99. Besides, multiclass models also developed for classifying samples in the normal, early and late stage of cancer and achieved an accuracy of 76.54% and AUC of 0.86.ConclusionOur study reveals stage prediction of LIHC samples with high accuracy based on the genomics and epigenomics profiling is a challenging task in comparison to the classification of cancerous and normal samples. Comprehensive analysis, differentially expressed RNA transcripts, methylated CpG sites in LIHC samples and prediction models are available from CancerLSP (http://webs.iiitd.edu.in/raghava/cancerlsp/).

Detecting Proline and Non-Proline Cis Isomers in Protein Structures from Sequences Using Deep Residual Ensemble Learning.

It has been long established that cis conformations of amino acid residues play many biologically important roles despite their rare occurrence in protein structure. Because of this rarity, few methods have been developed for predicting cis isomers from protein sequences, most of which are based on outdated datasets and lack the means for independent testing. In this work, using a database of >10000 high-resolution protein structures, we update the statistics of cis isomers and develop a sequence-based prediction technique using an ensemble of residual convolutional and long short-term memory bidirectional recurrent neural networks that allow learning from the whole protein sequence. We show that ensembling eight neural network models yields maximum Matthews correlation coefficient values of approximately 0.35 for cis-Pro isomers and 0.1 for cis-nonPro residues. The method should be useful for prioritizing functionally important residues in cis isomers for experimental validations and improving the sampling of rare protein conformations for ab initio protein structure prediction.

LMDIPred: A web-server for prediction of linear peptide sequences binding to SH3, WW and PDZ domains.

Protein-peptide interactions form an important subset of the total protein interaction network in the cell and play key roles in signaling and regulatory networks, and in major biological processes like cellular localization, protein degradation, and immune response. In this work, we have described the LMDIPred web server, an online resource for generalized prediction of linear peptide sequences that may bind to three most prevalent and well-studied peptide recognition modules (PRMs)—SH3, WW and PDZ. We have developed support vector machine (SVM)-based prediction models that achieved maximum Matthews Correlation Coefficient (MCC) of 0.85 with an accuracy of 94.55% for SH3, MCC of 0.90 with an accuracy of 95.82% for WW, and MCC of 0.83 with an accuracy of 92.29% for PDZ binding peptides. LMDIPred output combines predictions from these SVM models with predictions using Position-Specific Scoring Matrices (PSSMs) and string-matching methods using known domain-binding motif instances and regular expressions. All of these methods were evaluated using a five-fold cross-validation technique on both balanced and unbalanced datasets, and also validated on independent datasets. LMDIPred aims to provide a preliminary bioinformatics platform for sequence-based prediction of probable binding sites for SH3, WW or PDZ domains.

CRISPRpred: A flexible and efficient tool for sgRNAs on-target activity prediction in CRISPR/Cas9 systems.

The CRISPR/Cas9-sgRNA system has recently become a popular tool for genome editing and a very hot topic in the field of medical research. In this system, Cas9 protein is directed to a desired location for gene engineering and cleaves target DNA sequence which is complementary to a 20-nucleotide guide sequence found within the sgRNA. A lot of experimental efforts, ranging from in vivo selection to in silico modeling, have been made for efficient designing of sgRNAs in CRISPR/Cas9 system. In this article, we present a novel tool, called CRISPRpred, for efficient in silico prediction of sgRNAs on-target activity which is based on the applications of Support Vector Machine (SVM) model. To conduct experiments, we have used a benchmark dataset of 17 genes and 5310 guide sequences where there are only 20% true values. CRISPRpred achieves Area Under Receiver Operating Characteristics Curve (AUROC-Curve), Area Under Precision Recall Curve (AUPR-Curve) and maximum Matthews Correlation Coefficient (MCC) as 0.85, 0.56 and 0.48, respectively. Our tool shows approximately 5% improvement in AUPR-Curve and after analyzing all evaluation metrics, we find that CRISPRpred is better than the current state-of-the-art. CRISPRpred is enough flexible to extract relevant features and use them in a learning algorithm. The source code of our entire software with relevant dataset can be found in the following link: https://github.com/khaled-buet/CRISPRpred.

AVCpred: an integrated web server for prediction and design of antiviral compounds.

Viral infections constantly jeopardize the global public health due to lack of effective antiviral therapeutics. Therefore, there is an imperative need to speed up the drug discovery process to identify novel and efficient drug candidates. In this study, we have developed quantitative structure–activity relationship (QSAR)‐based models for predicting antiviral compounds (AVCs) against deadly viruses like human immunodeficiency virus (HIV), hepatitis C virus (HCV), hepatitis B virus (HBV), human herpesvirus (HHV) and 26 others using publicly available experimental data from the ChEMBL bioactivity database. Support vector machine (SVM) models achieved a maximum Pearson correlation coefficient of 0.72, 0.74, 0.66, 0.68, and 0.71 in regression mode and a maximum Matthew's correlation coefficient 0.91, 0.93, 0.70, 0.89, and 0.71, respectively, in classification mode during 10‐fold cross‐validation. Furthermore, similar performance was observed on the independent validation sets. We have integrated these models in the AVCpred web server, freely available at http://crdd.osdd.net/servers/avcpred. In addition, the datasets are provided in a searchable format. We hope this web server will assist researchers in the identification of potential antiviral agents. It would also save time and cost by prioritizing new drugs against viruses before their synthesis and experimental testing.

Prediction of anticancer molecules using hybrid model developed on molecules screened against NCI-60 cancer cell lines.

BackgroundIn past, numerous quantitative structure-activity relationship (QSAR) based models have been developed for predicting anticancer activity for a specific class of molecules against different cancer drug targets. In contrast, limited attempt have been made to predict the anticancer activity of a diverse class of chemicals against a wide variety of cancer cell lines. In this study, we described a hybrid method developed on thousands of anticancer and non-anticancer molecules tested against National Cancer Institute (NCI) 60 cancer cell lines.ResultsOur analysis of anticancer molecules revealed that majority of anticancer molecules contains 18–24 carbon atoms and are dominated by functional groups like R2NH, R3N, ROH, RCOR, and ROR. It was also observed that certain substructures (e.g., 1-methoxy-4-methylbenzene, 1-methoxy benzene, Nitrobenzene, Indole, Propenyl benzene) are more abundant in anticancer molecules. Next, we developed anticancer molecule prediction models using various machine-learning techniques and achieved maximum matthews correlation coefficient (MCC) of 0.81 with 90.40 % accuracy using support vector machine (SVM) based models. In another approach, a novel similarity or potency score based method has been developed using selected fragments/fingerprints and achieved maximum MCC of 0.82 with 90.65 % accuracy. Finally, we combined the strength of above methods and developed a hybrid method with maximum MCC of 0.85 with 92.47 % accuracy.ConclusionsWe developed a hybrid method utilizing the best of machine learning and potency score based method. The highly accurate hybrid method can be used for classification of anticancer and non-anticancer molecules. In order to facilitate scientific community working in the field of anticancer drug discovery, we integrate hybrid and potency method in a web server CancerIN. This server provides various facilities that includes; virtual screening of anticancer molecules, analog based drug design, and similarity with known anticancer molecules (http://crdd.osdd.net/oscadd/cancerin).Electronic supplementary materialThe online version of this article (doi:10.1186/s12885-016-2082-y) contains supplementary material, which is available to authorized users.

VaccineDA: Prediction, design and genome-wide screening of oligodeoxynucleotide-based vaccine adjuvants.

Immunomodulatory oligodeoxynucleotides (IMODNs) are the short DNA sequences that activate the innate immune system via toll-like receptor 9. These sequences predominantly contain unmethylated CpG motifs. In this work, we describe VaccineDA (Vaccine DNA adjuvants), a web-based resource developed to design IMODN-based vaccine adjuvants. We collected and analyzed 2193 experimentally validated IMODNs obtained from the literature. Certain types of nucleotides (e.g., T, GT, TC, TT, CGT, TCG, TTT) are dominant in IMODNs. Based on these observations, we developed support vector machine-based models to predict IMODNs using various compositions. The developed models achieved the maximum Matthews Correlation Coefficient (MCC) of 0.75 with an accuracy of 87.57% using the pentanucleotide composition. The integration of motif information further improved the performance of our model from the MCC of 0.75 to 0.77. Similarly, models were developed to predict palindromic IMODNs and attained a maximum MCC of 0.84 with the accuracy of 91.94%. These models were evaluated using a five-fold cross-validation technique as well as validated on an independent dataset. The models developed in this study were integrated into VaccineDA to provide a wide range of services that facilitate the design of DNA-based vaccine adjuvants (http://crdd.osdd.net/raghava/vaccineda/).

Identification of B-cell epitopes in an antigen for inducing specific class of antibodies

BackgroundIn the past, numerous methods have been developed for predicting antigenic regions or B-cell epitopes that can induce B-cell response. To the best of authors’ knowledge, no method has been developed for predicting B-cell epitopes that can induce a specific class of antibody (e.g., IgA, IgG) except allergenic epitopes (IgE). In this study, an attempt has been made to understand the relation between primary sequence of epitopes and the class of antibodies generated.ResultsThe dataset used in this study has been derived from Immune Epitope Database and consists of 14725 B-cell epitopes that include 11981 IgG, 2341 IgE, 403 IgA specific epitopes and 22835 non-B-cell epitopes. In order to understand the preference of residues or motifs in these epitopes, we computed and compared amino acid and dipeptide composition of IgG, IgE, IgA inducing epitopes and non-B-cell epitopes. Differences in composition profiles of different classes of epitopes were observed, and few residues were found to be preferred. Based on these observations, we developed models for predicting antibody class-specific B-cell epitopes using various features like amino acid composition, dipeptide composition, and binary profiles. Among these, dipeptide composition-based support vector machine model achieved maximum Matthews correlation coefficient of 0.44, 0.70 and 0.45 for IgG, IgE and IgA specific epitopes respectively. All models were developed on experimentally validated non-redundant dataset and evaluated using five-fold cross validation. In addition, the performance of dipeptide-based model was also evaluated on independent dataset.ConclusionPresent study utilizes the amino acid sequence information for predicting the tendencies of antigens to induce different classes of antibodies. For the first time, in silico models have been developed for predicting B-cell epitopes, which can induce specific class of antibodies. A web service called IgPred has been developed to serve the scientific community. This server will be useful for researchers working in the field of subunit/epitope/peptide-based vaccines and immunotherapy (http://crdd.osdd.net/raghava/igpred/).ReviewersThis article was reviewed by Dr. M Michael Gromiha, Dr Christopher Langmead (nominated by Dr Robert Murphy) and Dr Lina Ma (nominated by Dr Zhang Zhang).

Designing of inhibitors against drug tolerant Mycobacterium tuberculosis (H37Rv)

BackgroundMycobacterium tuberculosis (M.tb) is the causative agent of tuberculosis, killing ~1.7 million people annually. The remarkable capacity of this pathogen to escape the host immune system for decades and then to cause active tuberculosis disease, makes M.tb a successful pathogen. Currently available anti-mycobacterial therapy has poor compliance due to requirement of prolonged treatment resulting in accelerated emergence of drug resistant strains. Hence, there is an urgent need to identify new chemical entities with novel mechanism of action and potent activity against the drug resistant strains.ResultsThis study describes novel computational models developed for predicting inhibitors against both replicative and non-replicative phase of drug-tolerant M.tb under carbon starvation stage. These models were trained on highly diverse dataset of 2135 compounds using four classes of binary fingerprint namely PubChem, MACCS, EState, SubStructure. We achieved the best performance Matthews correlation coefficient (MCC) of 0.45 using the model based on MACCS fingerprints for replicative phase inhibitor dataset. In case of non-replicative phase, Hybrid model based on PubChem, MACCS, EState, SubStructure fingerprints performed better with maximum MCC value of 0.28. In this study, we have shown that molecular weight, polar surface area and rotatable bond count of inhibitors (replicating and non-replicating phase) are significantly different from non-inhibitors. The fragment analysis suggests that substructures like hetero_N_nonbasic, heterocyclic, carboxylic_ester, and hetero_N_basic_no_H are predominant in replicating phase inhibitors while hetero_O, ketone, secondary_mixed_amine are preferred in the non-replicative phase inhibitors. It was observed that nitro, alkyne, and enamine are important for the molecules inhibiting bacilli residing in both the phases. In this study, we introduced a new algorithm based on Matthews correlation coefficient called MCCA for feature selection and found that this algorithm is better or comparable to frequency based approach.ConclusionIn this study, we have developed computational models to predict phase specific inhibitors against drug resistant strains of M.tb grown under carbon starvation. Based on simple molecular properties, we have derived some rules, which would be useful in robust identification of tuberculosis inhibitors. Based on these observations, we have developed a webserver for predicting inhibitors against drug tolerant M.tb H37Rv available at http://crdd.osdd.net/oscadd/mdri/.

MicroRNA discovery by similarity search to a database of RNA-seq profiles

In silico generated search for microRNAs (miRNAs) has been driven by methods compiling structural features of the miRNA precursor hairpin, as well as to some degree combining this with the analysis of RNA-seq profiles for which the miRNA typically leave the drosha/dicer fingerprint of 1–2 ~22 nt blocks of reads corresponding to the mature and star miRNA. In complement to the previous methods, we present a study where we systematically exploit these patterns of read profiles. We created two datasets comprised of 2540 and 4795 read profiles obtained after preprocessing short RNA-seq data from miRBase and ENCODE, respectively. Out of 4795 ENCODE read profiles, 1361 are annotated as non-coding RNAs (ncRNAs) and of which 285 are further annotated as miRNAs. Using deepBlockAlign (dba), we align ncRNA read profiles from ENCODE against the miRBase read profiles (cleaned for “self-matches”) and are able to separate ENCODE miRNAs from the other ncRNAs by a Matthews Correlation Coefficient (MCC) of 0.8 and obtain an area under the curve of 0.93. Based on the dba score cut-off of 0.7 at which we observed the maximum MCC of 0.8, we predict 523 novel miRNA candidates. An additional RNA secondary structure analysis reveal that 42 of the candidates overlap with predicted conserved secondary structure. Further analysis reveal that the 523 miRNA candidates are located in genomic regions with MAF block (UCSC) fragmentation and poor sequence conservation, which in part might explain why they have been overlooked in previous efforts. We further analyzed known human and mouse miRNA read profiles and found two distinct classes; the first containing two blocks and the second containing >2 blocks of reads. Also the latter class holds read profiles that have less well defined arrangement of reads in comparison to the former class. On comparison of miRNA read profiles from plants and animals, we observed kingdom specific read profiles that are distinct in terms of both length and distribution of reads within the read profiles to each other. All the data, as well as a server to search miRBase read profiles by uploading a BED file, is available at http://rth.dk/resources/mirdba.