A Novel Feature Extraction Method with Feature Selection to Identify Golgi-Resident Protein Types from Imbalanced Data

A method for predicting linear and conformational B-cell epitopes in an antigen from its primary sequence

B-cell is an essential component of the immune system that plays a vital role in providing the immune response against any pathogenic infection by producing antibodies. Existing methods either predict linear or conformational B-cell epitopes in an antigen. In this study, a single method was developed for predicting both types (linear/conformational) of B-cell epitopes. The dataset used in this study contains 3875 B-cell epitopes and 3996 non-B-cell epitopes, where B-cell epitopes consist of both linear and conformational B-cell epitopes. Our primary analysis indicates that certain residues (like Asp, Glu, Lys, and Asn) are more prominent in B-cell epitopes. We developed machine-learning based methods using different types of sequence composition and achieved the highest AUROC of 0.80 using dipeptide composition. In addition, models were developed on selected features, but no further improvement was observed. Our similarity-based method implemented using BLAST shows a high probability of correct prediction with poor sensitivity. Finally, we developed a hybrid model that combines alignment-free (dipeptide based random forest model) and alignment-based (BLAST-based similarity) models. Our hybrid model attained a maximum AUROC of 0.83 with an MCC of 0.49 on the independent dataset. Our hybrid model performs better than existing methods on an independent dataset used in this study. All models were trained and tested on 80 % of the data using a cross-validation technique, and the final model was evaluated on 20 % of the data, called an independent or validation dataset. A webserver and standalone package named "CLBTope" has been developed for predicting, designing, and scanning B-cell epitopes in an antigen sequence available at (https://webs.iiitd.edu.in/raghava/clbtope/).

A machine learning approach reveals features related to clinicians' diagnosis of clinically relevant knee osteoarthritis

OBJECTIVES

To identify highly ranked features related to clinicians' diagnosis of clinically relevant knee OA.

METHODS

General practitioners (GPs) and secondary care physicians (SPs) were recruited to evaluate 5-10 years follow-up clinical and radiographic data of knees from the CHECK cohort for the presence of clinically relevant OA. GPs and SPs were gathered in pairs; each pair consisted of one GP and one SP, and the paired clinicians independently evaluated the same subset of knees. A diagnosis was made for each knee by the GP and SP before and after viewing radiographic data. Nested 5-fold cross-validation enhanced random forest models were built to identify the top 10 features related to the diagnosis.

RESULTS

Seventeen clinician pairs evaluated 1106 knees with 139 clinical and 36 radiographic features. GPs diagnosed clinically relevant OA in 42% and 43% knees, before and after viewing radiographic data, respectively. SPs diagnosed in 43% and 51% knees, respectively. Models containing top 10 features had good performance for explaining clinicians' diagnosis with area under the curve ranging from 0.76-0.83. Before viewing radiographic data, quantitative symptomatic features (i.e. WOMAC scores) were the most important ones related to the diagnosis of both GPs and SPs; after viewing radiographic data, radiographic features appeared in the top lists for both, but seemed to be more important for SPs than GPs.

CONCLUSIONS

Random forest models presented good performance in explaining clinicians' diagnosis, which helped to reveal typical features of patients recognized as clinically relevant knee OA by clinicians from two different care settings.

DMPPred: a tool for identification of antigenic regions responsible for inducing type 1 diabetes mellitus

There are a number of antigens that induce autoimmune response against β-cells, leading to type 1 diabetes mellitus (T1DM). Recently, several antigen-specific immunotherapies have been developed to treat T1DM. Thus, identification of T1DM associated peptides with antigenic regions or epitopes is important for peptide based-therapeutics (e.g. immunotherapeutic). In this study, for the first time, an attempt has been made to develop a method for predicting, designing, and scanning of T1DM associated peptides with high precision. We analysed 815 T1DM associated peptides and observed that these peptides are not associated with a specific class of HLA alleles. Thus, HLA binder prediction methods are not suitable for predicting T1DM associated peptides. First, we developed a similarity/alignment based method using Basic Local Alignment Search Tool and achieved a high probability of correct hits with poor coverage. Second, we developed an alignment-free method using machine learning techniques and got a maximum AUROC of 0.89 using dipeptide composition. Finally, we developed a hybrid method that combines the strength of both alignment free and alignment-based methods and achieves maximum area under the receiver operating characteristic of 0.95 with Matthew's correlation coefficient of 0.81 on an independent dataset. We developed a web server 'DMPPred' and stand-alone server for predicting, designing and scanning T1DM associated peptides (https://webs.iiitd.edu.in/raghava/dmppred/).

Effective multi-sensor data fusion for chatter detection in milling process

This paper introduces a newly developed multi-sensor data fusion for the milling chatter detection with a cheap and easy implementation compared with traditional chatter detection schemes. The proposed multi-sensor data fusion utilizes microphone and accelerometer sensors to measure the occurrence of chatter during the milling process. It has the advantageous over the dynamometer in terms of easy installation and low cost. In this paper, the wavelet packet decomposition is adopted to analyze both measured sound and vibration signals. However, the parameters of the wavelet packet decomposition require fine-tuning to provide good performance. Hence the result of the developed scheme has been improved by optimizing the selection of the wavelet packet decomposition parameters including the mother wavelet and the decomposition level based on the kurtosis and crest factors. Furthermore, the important chatter features are selected using the recursive feature elimination method, and its performance is compared with metaheuristic algorithms. Finally, several machine learning techniques have been adopted to classify the cutting stabilities based on the selected features. The results confirm that the proposed multi-sensor data fusion scheme can provide an effective chatter detection under industrial conditions, and it has higher accuracy than the traditional schemes.

Identification of Sub-Golgi protein localization by use of deep representation learning features

Motivation

The Golgi apparatus has a key functional role in protein biosynthesis within the eukaryotic cell with malfunction resulting in various neurodegenerative diseases. For a better understanding of the Golgi apparatus, it is essential to identification of sub-Golgi protein localization. Although some machine learning methods have been used to identify sub-Golgi localization proteins by sequence representation fusion, more accurate sub-Golgi protein identification is still challenging by existing methodology.

Results

we developed a protein sub-Golgi localization identification protocol using deep representation learning features with 107 dimensions. By this protocol, we demonstrated that instead of multi-type protein sequence feature representation fusion as in previous state-of-the-art sub-Golgi-protein localization classifiers, it is sufficient to exploit only one type of feature representation for more accurately identification of sub-Golgi proteins. Compared with independent testing results for benchmark datasets, our protocol is able to perform generally, reliably and robustly for sub-Golgi protein localization prediction.

Availabilityand implementation

A use-friendly webserver is freely accessible at http://isGP-DRLF.aibiochem.net and the prediction code is accessible at https://github.com/zhibinlv/isGP-DRLF.

Supplementary information

Supplementary data are available at Bioinformatics online.

A Combined Deep Learning and Ensemble Learning Methodology to Avoid Electricity Theft in Smart Grids

Electricity is widely used around 80% of the world. Electricity theft has dangerous effects on utilities in terms of power efficiency and costs billions of dollars per annum. The enhancement of the traditional grids gave rise to smart grids that enable one to resolve the dilemma of electricity theft detection (ETD) using an extensive amount of data formulated by smart meters. This data are used by power utilities to examine the consumption behaviors of consumers and to decide whether the consumer is an electricity thief or benign. However, the traditional data-driven methods for ETD have poor detection performances due to the high-dimensional imbalanced data and their limited ETD capability. In this paper, we present a new class balancing mechanism based on the interquartile minority oversampling technique and a combined ETD model to overcome the shortcomings of conventional approaches. The combined ETD model is composed of long short-term memory (LSTM), UNet and adaptive boosting (Adaboost), and termed LSTM–UNet–Adaboost. In this regard, LSTM–UNet–Adaboost combines the advantages of deep learning (LSTM-UNet) along with ensemble learning (Adaboost) for ETD. Moreover, the performance of the proposed LSTM–UNet–Adaboost scheme was simulated and evaluated over the real-time smart meter dataset given by the State Grid Corporation of China. The simulations were conducted using the most appropriate performance indicators, such as area under the curve, precision, recall and F1 measure. The proposed solution obtained the highest results as compared to the existing benchmark schemes in terms of selected performance measures. More specifically, it achieved the detection rate of 0.92, which was the highest among existing benchmark schemes, such as logistic regression, support vector machine and random under-sampling boosting technique. Therefore, the simulation outcomes validate that the proposed LSTM–UNet–Adaboost model surpasses other traditional methods in terms of ETD and is more acceptable for real-time practices.

A Novel Method to Identify Golgi Protein Types Based on Hybrid Feature and SVM Algorithm

Accurate identification of Golgi protein types can provide useful clues to reveal the correlation between GA dysfunction and disease pathology and improve the ability to develop more effective treatments for the diseases. This paper introduces an effective and robust method to classify Golgi protein type with traditional machine learning algorithms. In which various features such as n-GDip, DCCA, psePSSM were used as training features and SVM with linear kernel was employed as a classifier. To solve the imbalance problem of the benchmark datasets, the oversampling technique SMOTE was adopted. To deal with the huge amount of features, the PCA algorithm and Fisher feature selection method were adopted to reduce feature dimensions and remove redundant features. The experimental results show that the proposed method had a further improvement compared with other traditional machine learning methods in 10-fold cross-validation, Jackknife cross-validation and independent testing, which means a further step for the clinical application of computational methods to predict the Golgi protein types.

Identifying Heat Shock Protein Families from Imbalanced Data by Using Combined Features

Heat shock proteins (HSPs) are ubiquitous in living organisms. HSPs are an essential component for cell growth and survival; the main function of HSPs is controlling the folding and unfolding process of proteins. According to molecular function and mass, HSPs are categorized into six different families: HSP20 (small HSPS), HSP40 (J-proteins), HSP60, HSP70, HSP90, and HSP100. In this paper, improved methods for HSP prediction are proposed—the split amino acid composition (SAAC), the dipeptide composition (DC), the conjoint triad feature (CTF), and the pseudoaverage chemical shift (PseACS) were selected to predict the HSPs with a support vector machine (SVM). In order to overcome the imbalance data classification problems, the syntactic minority oversampling technique (SMOTE) was used to balance the dataset. The overall accuracy was 99.72% with a balanced dataset in the jackknife test by using the optimized combination feature SAAC+DC+CTF+PseACS, which was 4.81% higher than the imbalanced dataset with the same combination feature. The Sn, Sp, Acc, and MCC of HSP families in our predictive model were higher than those in existing methods. This improved method may be helpful for protein function prediction.

Use Chou’s 5-Step Rule to Classify Protein Modification Sites with Neural Network

Lysine malonylation is a novel-type protein post-translational modification and plays essential roles in many biological activities. Having a good knowledge of malonylation sites can provide guidance in many issues, including disease prevention and drug discovery and other related fields. There are several experimental approaches to identify modification sites in the field of biology. However, these methods seem to be expensive. In this study, we proposed malNet, which employed neural network and utilized several novel and effective feature description methods. It was pointed that ANN’s performance is better than other models. Furthermore, we trained the classifiers according to an original crossvalidation method named Split to Equal validation (SEV). The results achieved AUC value of 0.6684, accuracy of 54.93%, and MCC of 0.1045, which showed great improvement than before.

Machine learning techniques for protein function prediction

Proteins play important roles in living organisms, and their function is directly linked with their structure. Due to the growing gap between the number of proteins being discovered and their functional characterization (in particular as a result of experimental limitations), reliable prediction of protein function through computational means has become crucial. This paper reviews the machine learning techniques used in the literature, following their evolution from simple algorithms such as logistic regression to more advanced methods like support vector machines and modern deep neural networks. Hyperparameter optimization methods adopted to boost prediction performance are presented. In parallel, the metamorphosis in the features used by these algorithms from classical physicochemical properties and amino acid composition, up to text-derived features from biomedical literature and learned feature representations using autoencoders, together with feature selection and dimensionality reduction techniques, are also reviewed. The success stories in the application of these techniques to both general and specific protein function prediction are discussed.

A Random Forest Sub-Golgi Protein Classifier Optimized via Dipeptide and Amino Acid Composition Features

To gain insight into the malfunction of the Golgi apparatus and its relationship to various genetic and neurodegenerative diseases, the identification of sub-Golgi proteins, both cis-Golgi and trans-Golgi proteins, is of great significance. In this study, a state-of-art random forests sub-Golgi protein classifier, rfGPT, was developed. The rfGPT used 2-gap dipeptide and split amino acid composition for the feature vectors and was combined with the synthetic minority over-sampling technique (SMOTE) and an analysis of variance (ANOVA) feature selection method. The rfGPT was trained on a sub-Golgi protein sequence data set (137 sequences), with sequence identity less than 25%. For the optimal rfGPT classifier with 93 features, the accuracy (ACC) was 90.5%; the Matthews correlation coefficient (MCC) was 0.811; the sensitivity (Sn) was 92.6%; and the specificity (Sp) was 88.4%. The independent testing scores for the rfGPT were ACC = 90.6%; MCC = 0.696; Sn = 96.1%; and Sp = 69.2%. Although the independent testing accuracy was 4.4% lower than that for the best reported sub-Golgi classifier trained on a data set with 40% sequence identity (304 sequences), the rfGPT is currently the top sub-Golgi protein predictor utilizing feature vectors without any position-specific scoring matrix and its derivative features. Therefore, the rfGPT is a more practical tool, because no sequence alignment is required with tens of millions of protein sequences. To date, the rfGPT is the Golgi classifier with the best independent testing scores, optimized by training on smaller benchmark data sets. Feature importance analysis proves that the non-polar and aliphatic residues composition, the (aromatic residues) + (non-polar, aliphatic residues) dipeptide and aromatic residues composition between NH2-termial and COOH-terminal of protein sequences are the three top biological features for distinguishing the sub-Golgi proteins.

Electricity Theft Detection in Smart Grid Systems: A CNN-LSTM Based Approach

Among an electricity provider’s non-technical losses, electricity theft has the most severe and dangerous effects. Fraudulent electricity consumption decreases the supply quality, increases generation load, causes legitimate consumers to pay excessive electricity bills, and affects the overall economy. The adaptation of smart grids can significantly reduce this loss through data analysis techniques. The smart grid infrastructure generates a massive amount of data, including the power consumption of individual users. Utilizing this data, machine learning and deep learning techniques can accurately identify electricity theft users. In this paper, an electricity theft detection system is proposed based on a combination of a convolutional neural network (CNN) and a long short-term memory (LSTM) architecture. CNN is a widely used technique that automates feature extraction and the classification process. Since the power consumption signature is time-series data, we were led to build a CNN-based LSTM (CNN-LSTM) model for smart grid data classification. In this work, a novel data pre-processing algorithm was also implemented to compute the missing instances in the dataset, based on the local values relative to the missing data point. Furthermore, in this dataset, the count of electricity theft users was relatively low, which could have made the model inefficient at identifying theft users. This class imbalance scenario was addressed through synthetic data generation. Finally, the results obtained indicate the proposed scheme can classify both the majority class (normal users) and the minority class (electricity theft users) with good accuracy.

MinE-RFE: determine the optimal subset from RFE by minimizing the subset-accuracy–defined energy

Recursive feature elimination (RFE), as one of the most popular feature selection algorithms, has been extensively applied to bioinformatics. During the training, a group of candidate subsets are generated by iteratively eliminating the least important features from the original features. However, how to determine the optimal subset from them still remains ambiguous. Among most current studies, either overall accuracy or subset size (SS) is used to select the most predictive features. Using which one or both and how they affect the prediction performance are still open questions. In this study, we proposed MinE-RFE, a novel RFE-based feature selection approach by sufficiently considering the effect of both factors. Subset decision problem was reflected into subset-accuracy space and became an energy-minimization problem. We also provided a mathematical description of the relationship between the overall accuracy and SS using Gaussian Mixture Models together with spline fitting. Besides, we comprehensively reviewed a variety of state-of-the-art applications in bioinformatics using RFE. We compared their approaches of deciding the final subset from all the candidate subsets with MinE-RFE on diverse bioinformatics data sets. Additionally, we also compared MinE-RFE with some well-used feature selection algorithms. The comparative results demonstrate that the proposed approach exhibits the best performance among all the approaches. To facilitate the use of MinE-RFE, we further established a user-friendly web server with the implementation of the proposed approach, which is accessible at http://qgking.wicp.net/MinE/. We expect this web server will be a useful tool for research community.

A Brief Survey of Machine Learning Methods in Protein Sub-Golgi Localization

Background:

The location of proteins in a cell can provide important clues to their functions in various biological processes. Thus, the application of machine learning method in the prediction of protein subcellular localization has become a hotspot in bioinformatics. As one of key organelles, the Golgi apparatus is in charge of protein storage, package, and distribution.

Objective:

The identification of protein location in Golgi apparatus will provide in-depth insights into their functions. Thus, the machine learning-based method of predicting protein location in Golgi apparatus has been extensively explored. The development of protein sub-Golgi apparatus localization prediction should be reviewed for providing a whole background for the fields.

Method:

The benchmark dataset, feature extraction, machine learning method and published results were summarized.

Results:

We briefly introduced the recent progresses in protein sub-Golgi apparatus localization prediction using machine learning methods and discussed their advantages and disadvantages.

Conclusion:

We pointed out the perspective of machine learning methods in protein sub-Golgi localization prediction.

Decision Variants for the Automatic Determination of Optimal Feature Subset in RF-RFE

Feature selection, which identifies a set of most informative features from the original feature space, has been widely used to simplify the predictor. Recursive feature elimination (RFE), as one of the most popular feature selection approaches, is effective in data dimension reduction and efficiency increase. A ranking of features, as well as candidate subsets with the corresponding accuracy, is produced through RFE. The subset with highest accuracy (HA) or a preset number of features (PreNum) are often used as the final subset. However, this may lead to a large number of features being selected, or if there is no prior knowledge about this preset number, it is often ambiguous and subjective regarding final subset selection. A proper decision variant is in high demand to automatically determine the optimal subset. In this study, we conduct pioneering work to explore the decision variant after obtaining a list of candidate subsets from RFE. We provide a detailed analysis and comparison of several decision variants to automatically select the optimal feature subset. Random forest (RF)-recursive feature elimination (RF-RFE) algorithm and a voting strategy are introduced. We validated the variants on two totally different molecular biology datasets, one for a toxicogenomic study and the other one for protein sequence analysis. The study provides an automated way to determine the optimal feature subset when using RF-RFE.

isGPT: An optimized model to identify sub-Golgi protein types using SVM and Random Forest based feature selection

The Golgi Apparatus (GA) is a key organelle for protein synthesis within the eukaryotic cell. The main task of GA is to modify and sort proteins for transport throughout the cell. Proteins permeate through the GA on the ER (Endoplasmic Reticulum) facing side (cis side) and depart on the other side (trans side). Based on this phenomenon, we get two types of GA proteins, namely, cis-Golgi protein and trans-Golgi protein. Any dysfunction of GA proteins can result in congenital glycosylation disorders and some other forms of difficulties that may lead to neurodegenerative and inherited diseases like diabetes, cancer and cystic fibrosis. So, the exact classification of GA proteins may contribute to drug development which will further help in medication. In this paper, we focus on building a new computational model that not only introduces easy ways to extract features from protein sequences but also optimizes classification of trans-Golgi and cis-Golgi proteins. After feature extraction, we have employed Random Forest (RF) model to rank the features based on the importance score obtained from it. After selecting the top ranked features, we have applied Support Vector Machine (SVM) to classify the sub-Golgi proteins. We have trained regression model as well as classification model and found the former to be superior. The model shows improved performance over all previous methods. As the benchmark dataset is significantly imbalanced, we have applied Synthetic Minority Over-sampling Technique (SMOTE) to the dataset to make it balanced and have conducted experiments on both versions. Our method, namely, identification of sub-Golgi Protein Types (isGPT), achieves accuracy values of 95.4%, 95.9% and 95.3% for 10-fold cross-validation test, jackknife test and independent test respectively. According to different performance metrics, isGPT performs better than state-of-the-art techniques. The source code of isGPT, along with relevant dataset and detailed experimental results, can be found at https://github.com/srautonu/isGPT.

Predicting Presynaptic and Postsynaptic Neurotoxins by Developing Feature Selection Technique

Presynaptic and postsynaptic neurotoxins are proteins which act at the presynaptic and postsynaptic membrane. Correctly predicting presynaptic and postsynaptic neurotoxins will provide important clues for drug-target discovery and drug design. In this study, we developed a theoretical method to discriminate presynaptic neurotoxins from postsynaptic neurotoxins. A strict and objective benchmark dataset was constructed to train and test our proposed model. The dipeptide composition was used to formulate neurotoxin samples. The analysis of variance (ANOVA) was proposed to find out the optimal feature set which can produce the maximum accuracy. In the jackknife cross-validation test, the overall accuracy of 94.9% was achieved. We believe that the proposed model will provide important information to study neurotoxins.