PRESa2i: incremental decision trees for prediction of Adenosine to Inosine RNA editing sites

Alif Choyon; Ashiqur Rahman; Md. Hasanuzzaman; Dewan Md Farid; Swakkhar Shatabda

doi:10.12688/f1000research.22823.1

Home Browse PRESa2i: incremental decision trees for prediction of Adenosine to...

ALL Metrics

-

Views

-

Downloads

Get PDF

Get XML

Export

▬

✚

Method Article

PRESa2i: incremental decision trees for prediction of Adenosine to Inosine RNA editing sites

[version 1; peer review: 1 approved with reservations, 2 not approved]

Alif Choyon¹, Ashiqur Rahman¹, Md. Hasanuzzaman¹, Dewan Md Farid¹, Swakkhar Shatabda¹

Alif Choyon¹, Ashiqur Rahman¹, [...] Md. Hasanuzzaman¹, Dewan Md Farid¹, Swakkhar Shatabda¹

PUBLISHED 16 Apr 2020

Author details Author details

¹ Department of Computer Science and Engineering, United International University, Dhaka, 1212, Bangladesh

Alif Choyon
Roles: Data Curation, Investigation, Methodology, Software, Validation, Visualization, Writing – Original Draft Preparation, Writing – Review & Editing

Ashiqur Rahman
Roles: Methodology, Software, Validation, Visualization, Writing – Original Draft Preparation, Writing – Review & Editing

Md. Hasanuzzaman
Roles: Software, Validation, Visualization, Writing – Original Draft Preparation, Writing – Review & Editing

Dewan Md Farid
Roles: Supervision, Validation, Writing – Original Draft Preparation, Writing – Review & Editing

Swakkhar Shatabda
Roles: Conceptualization, Formal Analysis, Supervision, Writing – Original Draft Preparation, Writing – Review & Editing

OPEN PEER REVIEW

REVIEWER STATUS

Abstract

RNA editing is a very crucial cellular process affecting protein encoding and is sometimes correlated with the cause of fatal diseases, such as cancer. Thus knowledge about RNA editing sites in a RNA sequence is very important. Adenosine to Inosine (A-to-I) is the most common of the RNA editing events. In this paper,we present PRESa2i, a computation prediction tool for identification of A-to-I RNA editing sites in given RNA sequences. PRESa2i uses a simple, yet effective set of sequence based features generated from RNA sequences and a novel feature selection technique. It uses an incremental decision tree algorithm as the classification algorithm. On a standard benchmark dataset and independent set, it achieves 86.48% accuracy and 90.67% sensitivity and significantly outperforms state-of-the-art methods. We have also implemented a web application based on PRESa2i and made it available freely at: http://brl.uiu.ac.bd/presa2i/index.php. The materials for this paper are also available to use from: https://github.com/swakkhar/RNA-Editing/.

Keywords

RNA Editing Sites, Sequence Based Features, Feature Selection, Incremental Decision Tree

Corresponding author: Swakkhar Shatabda

Competing interests: No competing interests were disclosed.

Grant information: The author(s) declared that no grants were involved in supporting this work.

Copyright: © 2020 Choyon A et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

How to cite: Choyon A, Rahman A, Hasanuzzaman M et al. PRESa2i: incremental decision trees for prediction of Adenosine to Inosine RNA editing sites [version 1; peer review: 1 approved with reservations, 2 not approved]. F1000Research 2020, 9:262 (https://doi.org/10.12688/f1000research.22823.1) First published: 16 Apr 2020, 9:262 (https://doi.org/10.12688/f1000research.22823.1) Latest published: 16 Apr 2020, 9:262 (https://doi.org/10.12688/f1000research.22823.1)

Introduction

RNA editing is a process by which insertion, deletion or substitution of base nucleotides happen within the RNA. Such editing changes occur within the cell meaning nucleotide sequences are altered. This may cause alteration in the peptide sequence, which could be entirely different from that encoded in the DNA. Adenosine to Inosine (A-to-I) is the most common of all RNA editing events. Apart from making genetic alterations and resulting in codon changes A-to-I is often responsible for several intracellular operations like alternative splicing¹. A-to-I is found to have its effect on RNA secondary structural changes². Moreover, it is found to be in correlation with cancer formation³. Thus, research and identification of RNA editing is very important.

Experimental methods like RNA-Seq⁴ are effective means for identification of A-to-I sites. However, they are time consuming. Moreover, as known RNA-editing sites are available, computational methods for prediction of RNA editing sites are becoming more relevant in this era of knowledge and data discovery. Various computational methods are proposed in the literature for prediction of RNA editing sites, including Genetic Algorithms (GA)⁵, Support Vector Machines (SVM)⁶, Logistic Regression (LR)⁷ and Deep Learning⁸. PREPACT was proposed as a computational tool for C-to-U and U-to-C RNA editing site prediction for plant species⁹. Another method was presented by 10 for prediction of C-to-U editing sites using biochemical and evolutionary information.

Several machine learning based prediction methods are proposed in the literature of A-to-I RNA editing site identification in recent years. PAI was proposed by 11 that used pseudo nucleotide compositions as features with a SVM classification algorithm. They proposed and built two benchmark datasets derived from the Drosophila genome based on the work by 12. In a subsequent work, they proposed iRNA-AI⁶ using general form of pseudo nucleotide composition (PseKNC) with SVM and further improved the previous results on A-to-I datasets. A prediction method iRNA-3typeA¹³ was proposed to detect and identify three types of RNA-modifications at Adenosine sites. In a recent work, auto-encoders were used to develop PAI-SAE⁸ for prediction of A-to-I sites.

In this paper, we propose and present PRESa2i, a novel prediction method for A-to-I RNA Editing sites. In our proposed method, we have used the Hoeffding tree, an incremental decision tree for classification of samples. We have used sequence based features extracted from RNA samples collected as positive and negative instances in the dataset. We have used a novel feature selection method to select only the top 179 features. On a standard benchmark dataset PRESa2i has achieved a significant accuracy of 86.48% and on the independent test set it has a significant sensitivity of 90.67%. We have also made our method freely available for use by researchers at: http://brl.uiu.ac.bd/presa2i/index.php.

Methods

PRESa2i is trained with a dataset that has sequence based features. A novel feature selection algorithm is applied to select the top features and incremental decision tree algorithms are learned. The model is built on a standard benchmark training set and the performance of the model is tested using an independent set. As suggested by 14 and followed by many researchers^15–18, we follow five steps for establishing any good tool for attribute prediction of biological entities: i) selection of benchmark datasets, ii) feature generation, iii) selection of appropriate algorithm, iv) evaluation methods, and v) establishment of a web server.

Datasets

The first step in any machine learning based supervised prediction task is to select a standard set of data. The RNA A-to-I editing site prediction problem is formulated as a binary classification problem, where Adenine in a RNA sequence or subsequence is labeled with two classes: A-to-I or positive samples and non A-to-I or negative samples. A dataset can be formally defined as the following:

S = S^{+} \cup S^{-} (1)

Here, $S$ denotes the dataset and positive and negative datasets are denoted by $S^{+}$ and $S^{-}$ respectively. The datasets that we have used is first used in 11 based on the work in 12. The original dataset constructed in 12 contained 127 A-to-I editing sites with sequences and 127 non A-to-I editing sites with sequences. These sequences were obtained by sequencing wild-type and ADAR-deficient D. melanogaster DNA and RNA. After removing the redundant sequences, a benchmark dataset was constructed that contained 125 positive or A-to-I sites with sequences and 119 negative or non A-to-I sequence sites. Each of these RNA sequences in the dataset are 51 nucleotides long. From herein, we refer to this dataset as benchmark dataset.

An independent test set was constructed by further analyzing the sequences of D. melanogaster by 19. This set contained 51 nucleotide long 300 positive A-to-I sites with sequence. A summary of the datasets are presented in Table 1. Note that the benchmark dataset is balanced, while the independent dataset contains only positive sequences. Both datasets used are available here: https://github.com/swakkhar/RNA-Editing/.

Table 1. Summary of two datasets used in this study.

Dataset	Positive Instances	Negative Instances	Sequence Size
Benchmark Dataset	125	119	51
Independent Test Set	300	–	51

RNA sequence representation

Any sample in the dataset is a sequence of RNAs. These are all 51 nucleotide long RNA strings. All of these strings are formed by taking symbols from the RNA alphabet, Σ = {A, C, G, U}. Formally a RNA sequence string R ∈ $S$ can be formulated as below:

R = N_{1}, N - 2, \dots, N_{L} (2)

Here, N_i is a nucleotide symbol and L = 51 is the length of the RNA sequence. We have extracted three groups of features from each sequence. Thus each sample in the dataset corresponds to a feature vector containing three groups of features, as below:

F (R) = [F_{1} (R) F_{2} (R) F_{3} (R)] (3)

The three groups of features that were considered in the PRESa2i method are: k-mer compositions, gapped k-mer compositions and other statistical features. The first two types of features are widely used in the literature for solving other problems as well.

1. k-mer Compositions: k-mers are sequences of length k drawn from the alphabet Σ. Compositions are calculated as the frequency of the different length k-mers normalized by sequence length. This is a widely used feature in the literature^15,20. We have used k-mers with value of k = 1, 2, 3, 4. Thus the total number of features in this group was 4+16+64+256=340.
2. Gapped k-mers: We have used gapped k-mer compositions as features. These features were previously used in the literature^15,21 and are extended from the ideas of gapped k-mer composition²² or gapped di-peptide composition²³. They are normalized frequency of k-mers with gaps in between them. We have considered gaps, g = 1, 2, 3, · · · , 10 in this paper and considered up to k-mers with k = 2, 3, 4. Here, the total number of features is 160+640+640+2560=4000.
3. Other statistical features: We have used other statistically derived features, ratio of start and end codons and distribution of bases, as features. In total, there were five features from this group. The start codon is AUG and end codons are UGA, UAA and UAG. We have used the ratio of these codons. Also there are four nucleotides or bases: A, U, G, C. We have converted the presence of these bases into a distribution and used that as a feature.

Hybrid feature selection

Total number of features generated using the feature extraction technique was 4345. The feature generation technique is simple since they are extracted directly from RNA sequences. After the features are selected, we have used a hybrid feature selection technique. Our feature hybrid feature selection method is a multi-step step method where the first step is a wrapper method followed by consecutive filter methods. In the first phase of feature selection we have used best first search (BFS). It is an incremental feature selection procedure that adds or deletes one feature at a time and finds the optimal set of features. After the first phase the selected features were then sent for a single pass classification done by a logistic function (LF). For each class labels, the features with positive weights were considered and others were filtered out in this phase. In the last step, we used ensemble of random decision trees and added most significant features found by the randomly generated features which were left in the previous step to get the final set of features. Figure 1 shows the steps of the feature selection technique used by the PRESa2i method.

Figure 1. Steps of hybrid feature selection technique used by the PRESa2i method.

Incremental decision trees

We have used Hoeffding tree²⁴, an incremental decision tree algorithm as the classification engine for PRESa2i. Though its more applicable to data streams, the performance of this classifier is nearly the same as non-incremental learning algorithms. This algorithm starts with a single leaf decision tree and in each iteration for all examples it uses the Hoeffding bound to split the node, recursively selecting one attribute each time based on evaluative functions. The leafs are decided using adaptive naive Bayes technique.

Performance evaluation

We have used four metrics for the binary classification problem. These are Accuracy, Sensitivity (S_n), Specificity (S_p) and Mathew’s Correlation Coefficient (MCC). These metrics were selected to conform with previous methods^8,11. For a binary classification method, let T P denote the number of true positives, T N denote the number of true negatives, F P denote the number of false positives and F N denote the number of false negatives. Now the four metrics are defined as follows:

A c c u r a c y = \frac{T N + T P}{T P + F N + T N + F P} (4)

S_{n} = \frac{T P}{T P + F N} (5)

S_{p} = \frac{T N}{T N + F P} (6)

M C C = \frac{(T P \times T N) - (F P \times F N)}{\sqrt{(T P + F P) (T P + F N) (T N + F P) (T N + F N)}} (7)

Note that, the first three metrics have the values in the range [0,1], where a value equal or near 0 means a bad classifier and a value near 1 means a good classifier. However, MCC have values in range [-1,1]. We have also used Receiver Operating Characteristic (ROC) curve to show the performances of different classifiers. It plots true positive rate against false positive rate at different thresholds of a probabilistic classification method.

Results and discussion

In this section, we present the details of experimental results for our proposed method, PRESa2i. All the algorithms and programs were developed using Java 8 standard edition and Weka library²⁵. All the experiments were done 10 times and only the average results are reported.

Comparison with other methods

In this section, we provide the experimental results for comparisons with other state-of-the-art methods with our method PRESa2i. We have considered two previous state-of-the-art methods for comparison with our method: PAI proposed by 11 and PAI-SAE proposed by 8. We have used accuracy, sensitivity, specificity and MCC to compare the results. We have not run PAI and PAI-SE, rather as all these methods are tested on the same benchmark dataset, we have taken the results as being reported in the respective publications. Table 2 presents the results on the benchmark dataset. In this table, best values in each criteria are shown in bold fonts.

Table 2. Performance comparison of different methods - PRESa2i, PAI¹¹ and PAI-SAE⁸ - on the benchmark dataset.

Method name	Accuracy	MCC	Sensitivity	Specificity
PAI	79.59%	0.0600	0.8560	0. 7311
PAI-SAE	81.97%	0.6414	0.8720	0. 7647
PRESa2i	86.48%	0.7393	0.9520	0.7731

From the values reported in Table 2, we can see that PRESa2i achieves higher values in all the performance metrics compared with PAI and PAI-SAE. In case of accuracy our method is 6.89% and 4.51% more accurate than PAI and PAI-SAE, respectively. It also outperforms both of the methods in terms of MCC and sensitivity. In case of specificity its performance is significantly improved compared to other two methods. To further assess the performance of our method we used an independent test set. Note that, while constructing the independent test set CD-HIT was used to remove sequences with higher similarity. We used the same cutoff as being used in PAI proposed by 11. In the case of the independent test set as there were no negative samples, and the only metric applicable was sensitivity. We report the sensitivity of PAI and PRESa2i in Table 3.

Table 3. Performance comparison of different meth- ods - PRESa2i, PAI¹¹ - on the independent test set.

Method Name	PAI²⁶	PRESa2i
Sensitivity	82.33%	90.67%

Note that, the sensitivity of our method on the benchmark dataset was not a result of overfit and hence 272 samples out of 300 samples were correctly predicted in the independent test set by PRESa2i compared to 247 correctly predicted by PAI. Thus, we can conclude that our method achieved superior performances in both benchmark dataset and independent test set over these current state-of-the-art methods.

Effects of feature selection

We have proposed a multi-stage novel feature selection method. To show the effectiveness of our feature selection method, we have used accuracy, sensitivity and specificity as performance measures. We have compared the performance of the feature selection method with that of ‘Using All Features’ and with that of ‘Reduced Using BFS+LF’. Results are reported in Table 4. We have used six classifiers: Naive Bayes Classifier, AdaBoost Classifier, Hoeffding Tree, Random Forest Classifier, Support Vector Machines and Bagging.

Table 4. Performance of classifiers in different stages of feature selection method on the benchmark dataset.

Classifier	Using All Features			Reduced Using BFS+LF			Final Selected Features
Classifier		Sn(%)	Sp(%)	Acc(%)	Sn(%)	Sp(%)	Acc(%)	Sn(%)	Sp(%)	Acc(%)
Naive Bayes	88.0	69.74	79.09	91.2	75.63	83.60	92.0	78.15	85.24
AdaBoost	92.8	63.02	78.27	94.4	78.15	86.47	96.0	81.51	88.93
Hoeffding Tree	90.4	68.90	79.91	92.8	76.47	84.83	95.2	77.31	86.47
Random Forest	91.2	67.22	79.50	94.4	76.47	85.65	92.8	73.94	83.60
SVM	82.4	77.31	79.91	92.8	88.23	90.57	93.6	89.91	91.80
Bagging	88.0	64.70	76.63	93.6	71.42	82.78	88.0	70.58	79.50

The bold fonts show the best values of accuracy, sensitivity and specificity for each of the classifiers on the different set of features selected at different stages of feature selection. From this, we can note the effectiveness of the over-all feature selection procedure. Here, we can see that, using best first search and logistic function based feature selection in the first phase generates features that improves the performance of all the classifiers compared to no feature selection. On the other hand, when we further refine the features using decision tree based significance test, all the measures improve for four of the classifiers which is an indication of the effectiveness of the later stage of feature selection.

Classifier selection

We can note that among the six classifiers used in our experiments as shown in Table 4, AdaBoost classifier and Hoeffding Tree are the two best classifiers, followed by SVM. In the case of SVM though the accuracy is higher compared to these two algorithms, the sensitivity is not as good as these two. Note that these are the results on the benchmark dataset. We have also shown the receiver operating characteristic curve for each of these classifiers in Figure 2.

Figure 2. Receiver operating characteristic curve for different classifiers on the benchmark dataset.

To make the selection of the final classification algorithm, we also tested them on the independent dataset. In Table 5, the results on the independent test set in terms of sensitivity is presented. We see that, the performance of Hoeffding tree does not overfit the data in benchmark dataset and hence the accuracy is enhanced from 86.47% to 90.67% from benchmark dataset to the independent test set. Thus we chose Hoeffding tree as our final classifier and save the model learned by this algorithm for the prediction tool.

Table 5. Performance of different classifiers in terms of sensitivity on the independent test set.

Algorithm Name	Naive Bayes	SVM	Hoeffding Tree	AdaBoost	Bagging	Random Forest
Sensitivity (%)	86.67	64.67	90.67	85.33	82.67	83.67

Effect of negative test data

We have further investigated the effect of introducing negative instances in the independent test set. For this purpose, we have generated negative instances using a empirical distribution learned from the negative training dataset based on the distribution of the base nucleotides only. The generated negative instance are then added to the positive ones and the algorithms were again tested on this set. The results are reported in Table 6.

Table 6. Experiments on the independent dataset after introduction of negative instances.

Algorithm	S_n (%)	S_p (%)	Accuracy (%)
Naive Bayes	86.66	73.33	80.00
AdaBoost	85.33	51.33	68.33
Hoeffding Tree	90.66	52.66	71.66
Random Forest	83.66	74.33	79.00
SVM	64.66	70.00	67.33
Bagging	82.66	61.66	72.16

From the results reported in Table 6, we note that addition of the negative data in the independent test set does not affect on the effectiveness of the features and the feature selection method and the model trained. Note that, the introduced negative instances are given in the repository at: https://github.com/swakkhar/RNA-Editing/.

Please note that the results presented in Table 6 shows the effect of introducing negative examples in the dataset. Note that these negative datasets are taken randomly and are not verified at all. We note that though the accuracy and sensitivity is still satisfactory, Hoeffding tree and AdaBoost fails in terms of specificity. However, it also indicates that there is scope to improve the performance by adding more negative samples in the training set and also to improve the independent test set by adding real negative examples instead of taking random samples.

Web server

We have also implemented a web server, based on the best classification model learned by the incremental decision tree and final selected set of features on the benchmark dataset. Our website was developed using PHP language on the server side using Java and Weka library to provide the prediction from backend. Our method is readily available for use from: http://brl.uiu.ac.bd/presa2i/ index.php.

Implementation

The web server has two major components: an web interface and a background server application. A small brief on the implementation is given below:

1. Web Interface: The web interface and its backend is implemented using PHP and boostrap. The PHP script takes the input of the RNA sequence and then calls the prediction server which is a Java program and again receives the output from the prediction server and shows them in the web area.
2. Prediction Server: The prediction server is written in Java language. All the experiments are written in java language. The feature generation and selection part is implemented using Java. A program takes the input of the RNA sequence and generates the necessary features and then uses the Weka package for the classification of indentification engine. After the results are processed they are fed back to the Web interface where the results are shown. Interface between the Web Interface and the Java program is made using secured http connection.

Operation

The operation of the webserver is very simple. There are only one text field and two buttons. A screenshot of the developed web application is given in Figure 3. Here is a step by step procedure on thow to use the web application:

Figure 3. Screenshot of the web application developed for PRESa2i.

1. First one has to put the input RNA sequences in the text area. If there is any mistake, pressing the ‘Clear’ button will clear the area.
2. After the input is given in the text area, the predict button should be pressed. Thus a request would be made to the online server.
3. The results for the query sequence will be shown just below the input pane. For the next entry, the process should be followed in the same manner starting from step 1.

Conclusion

In this article, we have proposed PRESa2i, a novel predictor for A-to-I RNA editing sites using sequence based features and incremental decision trees. Our method is based on a number of effective features selected using successive phases of best first search in incremental forward selection and ranking based on random decision trees and logistic regression. We have made all our datasets available online at: https://github.com/swakkhar/RNA-Editing/, and have also made our method available as an web application from: http://brl.uiu.ac.bd/presa2i/index.php. In the future, we would like to explore other RNA editing sites and develop layered or multi-class prediction system. We believe further enhancement of the dataset can improve the performance of the method to a great extent.

Data availability

Datasets for this article available from: https://github.com/swakkhar/RNA-Editing/

Archived datasets as at time of publication: https://zenodo.org/record/3727221²⁷

License: CC0

Software availability

PRESa2i web application: http://brl.uiu.ac.bd/ presa2i/index.php

Source code available from: https://github.com/swakkhar/RNA-Editing

Archived source code as at time of publication: https://zenodo.org/record/3727221²⁷

License: CC0

Faculty Opinions recommended

References

1. Rueter SM, Dawson TR, Emeson RB: Regulation of alternative splicing by RNA editing. Nature. 1999; 399(6731): 75–80. PubMed Abstract | Publisher Full Text
2. Maas Stefan: Gene regulation through RNA editing. Discov Med. 2010; 10(54): 379–386. PubMed Abstract
3. Paz N, Levanon EY, Amariglio N, et al.: Altered adenosine-to-inosine RNA editing in human cancer. Genome Res. 2007; 17(11): 1586–1595. PubMed Abstract | Publisher Full Text | Free Full Text
4. Wang Z, Gerstein M, Snyder M: RNA-Seq: a revolutionary tool for transcriptomics. Nat Rev Genet. 2009; 10(1): 57. PubMed Abstract | Publisher Full Text | Free Full Text
5. Thompson J, Gopal S: Genetic algorithm learning as a robust approach to RNA editing site prediction. BMC Bioinformatics. 2006; 7(1): 145. PubMed Abstract | Publisher Full Text | Free Full Text
6. Chen W, Feng P, Yang H, et al.: iRNA-AI: identifying the adenosine to inosine editing sites in RNA sequences. Oncotarget. 2017; 8(3): 4208–4217. PubMed Abstract | Publisher Full Text | Free Full Text
7. Xiong H, Liu D, Li Q, et al.: RED-ML: a novel, effective RNA editing detection method based on machine learning. GigaScience. 2017; 6(5): 1–8. PubMed Abstract | Publisher Full Text | Free Full Text
8. Xiao X, Wang P, Xu Z, et al.: Pai-sae: Predicting adenosine to inosine editing sites based on hybrid features by using spare autoencoder. In IOP Conference Series: Earth and Environmental Science. 2018; 170(5): 052018. IOP Publishing. Publisher Full Text
9. Lenz H, Knoop V: PREPACT 2.0: Predicting C-to-U and U-to-C RNA Editing in Organelle Genome Sequences with Multiple References and Curated RNA Editing Annotation. Bioinform Biol Insights. 2013; 7: 1–19. PubMed Abstract | Publisher Full Text | Free Full Text
10. Du P, Li Y: Prediction of C-to-U RNA editing sites in plant mitochondria using both biochemical and evolutionary information. J Theor Biol. 2008; 253(3): 579–586. PubMed Abstract | Publisher Full Text
11. Chen W, Feng P, Ding H, et al.: PAI: Predicting adenosine to inosine editing sites by using pseudo nucleotide compositions. Sci Rep. 2016; 6: 35123. PubMed Abstract | Publisher Full Text | Free Full Text
12. St Laurent G, Tackett MR, Nechkin S, et al.: Genome-wide analysis of A-to-I RNA editing by single-molecule sequencing in Drosophila. Nat Struct Mol Biol. 2013; 20(11): 1333–9. PubMed Abstract | Publisher Full Text
13. Chen W, Feng P, Yang H, et al.: iRNA-3typeA: Identifying Three Types of Modification at RNA's Adenosine Sites. Mol Ther Nucleic Acids. 2018; 11: 468–474. PubMed Abstract | Publisher Full Text | Free Full Text
14. Chou KC: Some remarks on protein attribute prediction and pseudo amino acid composition. J Theor Biol. 2011; 273(1): 236–247. PubMed Abstract | Publisher Full Text | Free Full Text
15. Rahman S, Aktar U, Jani R, et al.: iPromoter-FSEn: Identification of bacterial σ⁷⁰ promoter sequences using feature subspace based ensemble classifier. Genomics. 2019; 111(5): 1160–1166. PubMed Abstract | Publisher Full Text
16. Rayhan F, Ahmed S, Shatabda S, et al.: iDTI-ESBoost: Identification of Drug Target Interaction Using Evolutionary and Structural Features with Boosting. Sci Rep. 2017; 7(1): 17731. PubMed Abstract | Publisher Full Text | Free Full Text
17. Chowdhury SY, Shatabda S, Dehzangi A: iDNAProt-ES: Identification of DNA-binding Proteins Using Evolutionary and Structural Features. Sci Rep. 2017; 7(1): 14938. PubMed Abstract | Publisher Full Text | Free Full Text
18. Uddin MR, Sharma A, Farid DM, et al.: EvoStruct-Sub: An accurate Gram-positive protein subcellular localization predictor using evolutionary and structural features. J Theor Biol. 2018; 443: 138–146. PubMed Abstract | Publisher Full Text
19. Yu Y, Zhou H, Kong Y, et al.: The Landscape of A-to-I RNA Editome Is Shaped by Both Positive and Purifying Selection. PLoS Genet. 2016; 12(7): e1006191. PubMed Abstract | Publisher Full Text | Free Full Text
20. Shatabda S, Saha S, Sharma A, et al.: iPHLoc-ES: Identification of bacteriophage protein locations using evolutionary and structural features. J Theor Biol. 2017; 435: 229–237. PubMed Abstract | Publisher Full Text
21. Al Maruf MA, Shatabda S: iRSpot-SF: Prediction of recombination hotspots by incorporating sequence based features into Chou's Pseudo. Genomics. 2019; 111(4): 966–972. PubMed Abstract | Publisher Full Text
22. Ghandi M, Mohammad-Noori M, Beer MA: Robust k-mer frequency estimation using gapped k-mers. J Math Biol. 2014; 69(2): 469–500. PubMed Abstract | Publisher Full Text | Free Full Text
23. Chang J, Su EC, LO A, et al.: PSLDoc: Protein subcellular localization prediction based on gapped-dipeptides and probabilistic latent semantic analysis. Proteins. 2008; 72(2): 693–710. PubMed Abstract | Publisher Full Text
24. Hulten G, Spencer L, Domingos P: Mining time-changing data streams. In ACM SIGKDD Intl. Conf. on Knowledge Discovery and Data Mining. ACM Press. 2001; 97–106. Publisher Full Text
25. Hall M, Frank E, Holmes G, et al.: The weka data mining software: an update. ACM SIGKDD explorations newsletter. 2009; 11(1): 10–18. Publisher Full Text
26. Chen XX, Tang H, Li WC, et al.: Identification of Bacterial Cell Wall Lyases via Pseudo Amino Acid Composition. Biomed Res Int. 2016; 2016: 1654623. PubMed Abstract | Publisher Full Text | Free Full Text
27. swakkhar: swakkhar/RNAEditing:Pres-a2i (Version v1.1). Zenodo. 2020. http://www.doi.org/10.5281/zenodo.3727221

Comments on this article Comments (0)

Version 1

VERSION 1 PUBLISHED 16 Apr 2020

Author details Author details

¹ Department of Computer Science and Engineering, United International University, Dhaka, 1212, Bangladesh

Alif Choyon
Roles: Data Curation, Investigation, Methodology, Software, Validation, Visualization, Writing – Original Draft Preparation, Writing – Review & Editing

Ashiqur Rahman
Roles: Methodology, Software, Validation, Visualization, Writing – Original Draft Preparation, Writing – Review & Editing

Md. Hasanuzzaman
Roles: Software, Validation, Visualization, Writing – Original Draft Preparation, Writing – Review & Editing

Dewan Md Farid
Roles: Supervision, Validation, Writing – Original Draft Preparation, Writing – Review & Editing

Swakkhar Shatabda
Roles: Conceptualization, Formal Analysis, Supervision, Writing – Original Draft Preparation, Writing – Review & Editing

Competing interests

No competing interests were disclosed.

Grant information

The author(s) declared that no grants were involved in supporting this work.

Article Versions (1)

version 1

Published: 16 Apr 2020, 9:262

https://doi.org/10.12688/f1000research.22823.1

Copyright

© 2020 Choyon A et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Download

Export To

metrics

	Views	Downloads
F1000Research	-	-
PubMed Central Data from PMC are received and updated monthly.	-	-

Citations

0

SEE MORE DETAILS

CITE

how to cite this article

Choyon A, Rahman A, Hasanuzzaman M et al. PRESa2i: incremental decision trees for prediction of Adenosine to Inosine RNA editing sites [version 1; peer review: 1 approved with reservations, 2 not approved]. F1000Research 2020, 9:262 (https://doi.org/10.12688/f1000research.22823.1)

NOTE: If applicable, it is important to ensure the information in square brackets after the title is included in all citations of this article.

track

receive updates on this article

Track an article to receive email alerts on any updates to this article.

Open Peer Review

Current Reviewer Status: ?

Key to Reviewer Statuses VIEW HIDE

ApprovedThe paper is scientifically sound in its current form and only minor, if any, improvements are suggested

Approved with reservations A number of small changes, sometimes more significant revisions are required to address specific details and improve the papers academic merit.

Not approvedFundamental flaws in the paper seriously undermine the findings and conclusions

Version 1

VERSION 1

PUBLISHED 16 Apr 2020

Views

4

Reviewer Report 17 Aug 2021

Xiaoyong Pan, Institute of Image Processing and Pattern Recognition, Shanghai Jiao Tong University, Shanghai, China

Approved with Reservations

https://doi.org/10.5256/f1000research.25198.r91074

The authors present a computational method for predicting A-to-I RNA editing sites using sequence-based features and incremental decision trees, and they show that the proposed method yields better performance. However, there are limited novelties in this manuscript.

The authors present a computational method for predicting A-to-I RNA editing sites using sequence-based features and incremental decision trees, and they show that the proposed method yields better performance. However, there are limited novelties in this manuscript.

How is sequence similarity between the training set and test set? The redundant sequences in the test set should be removed.
Deep learning has been widely used in genomic sequences. The authors should compare the proposed method with the CNN-based method which uses a one-hot encoded sequence as input.
In Table 3, for the independent test set, why is only sensitivity reported? Are other metrics of the proposed method worse than the baseline method? It is bad to only keep positive samples in the independent test set.
In Figure 2, the AUC value should be added.
How is the performance of XGBoost on this task? This classifier has been shown to perform well on many tasks.

Is the rationale for developing the new method (or application) clearly explained?

Partly
Is the description of the method technically sound?

Partly
Are sufficient details provided to allow replication of the method development and its use by others?

Partly
If any results are presented, are all the source data underlying the results available to ensure full reproducibility?

Partly
Are the conclusions about the method and its performance adequately supported by the findings presented in the article?

Partly

Competing Interests: No competing interests were disclosed.

Reviewer Expertise: Bioinformatics, RNA-protein interaction prediction

I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.

CITE

Report a concern

Respond or Comment

Views

5

Reviewer Report 21 Jun 2021

Shizuka Uchida, Center for RNA Medicine, Department of Clinical Medicine, Aalborg University, Aalborg, Denmark

Not Approved

https://doi.org/10.5256/f1000research.25198.r87702

In this manuscript, the authors present a new method to predict A-to-I RNA editing sites. This method is implemented in a web interface so that users can access freely. The method proposed is technically sound and valid, yet the rationale ... Continue reading

In this manuscript, the authors present a new method to predict A-to-I RNA editing sites. This method is implemented in a web interface so that users can access freely. The method proposed is technically sound and valid, yet the rationale for developing such method is hampered by the fact that there are many bioinformatic tools available to detect (not predict) A-to-G changes from RNA sequencing (RNA-seq) data. Thus, the usage of the author’s web interface is not of high priority to researchers; given that the cost for RNA-seq is not high anymore. More specific comments are listed below:

Major points

The authors must benchmark the performance of their tool to the known RNA editing sites deposited in databases, such as DARNED, RADAR, and REDIportal.
The authors must provide the results of other species, as most of the current research in A-to-I RNA editing is focused on human diseases.
This web tool is not functioning. When the known RNA edited sequence, TTCTCAACCTCCTGGGCTCTTGTGATCCTCCTGTCTTA (from PMID: 27595325), was pasted onto the field, Enter the query RNA sequences here, and the [Predict] button is pressed, the following message appears: “Login Failed”. Thus, this reviewer cannot evaluate the actual performance of this tool that the authors developed.

Minor points

(1) From the URL (http://brl.uiu.ac.bd/presa2i/index.php#), nothing happens when “ • Don't have any RNA string click here.” is pressed.

Is the rationale for developing the new method (or application) clearly explained?

Partly
Is the description of the method technically sound?

Yes
Are sufficient details provided to allow replication of the method development and its use by others?

Yes
If any results are presented, are all the source data underlying the results available to ensure full reproducibility?

Yes
Are the conclusions about the method and its performance adequately supported by the findings presented in the article?

Yes

References

1. Stellos K, Gatsiou A, Stamatelopoulos K, Perisic Matic L, et al.: Adenosine-to-inosine RNA editing controls cathepsin S expression in atherosclerosis by enabling HuR-mediated post-transcriptional regulation.Nat Med. 22 (10): 1140-1150 PubMed Abstract | Publisher Full Text

Competing Interests: No competing interests were disclosed.

Reviewer Expertise: bioinformatics, biochemistry, developmental biology, molecular biology, RNA biology, non-coding RNAs, miRNAs, lncRNAs, circRNAs, A-to-I RNA editing, epitranscriptomics, RNA sequencing

I confirm that I have read this submission and believe that I have an appropriate level of expertise to state that I do not consider it to be of an acceptable scientific standard, for reasons outlined above.

CITE

Report a concern

Respond or Comment

Views

23

Reviewer Report 19 May 2020

Avdesh Mishra, Department of Electrical Engineering and Computer Science, Texas A&M University - Kingsville, Kingsville, TX, USA

Not Approved

https://doi.org/10.5256/f1000research.25198.r62452

In this paper, authors present PRESa2i, a computation prediction tool for identification of A-to-I RNA editing sites in given RNA sequences. PRESa2i uses a simple, yet effective set of sequence based features generated from RNA sequences and a novel feature selection ... Continue reading

In this paper, authors present PRESa2i, a computation prediction tool for identification of A-to-I RNA editing sites in given RNA sequences. PRESa2i uses a simple, yet effective set of sequence based features generated from RNA sequences and a novel feature selection technique. It uses an incremental decision tree algorithm as the classification algorithm.

Below are my comments and suggestions:

Authors should specify which method the proposed method outperforms "significantly outperforms state-of-the-art methods" is a very general statement. In near future there will be another method that would outperform this method.
I did not find any example RNA sequence when I clicked on "click here" link on the authors webserver.
The webserver is not working. I tried to run the webserver with an example RNA Sequence. But I got "Login Failed" message.
Does authors method achieves 86.48% accuracy and 90.67% sensitivity on benchmark or independent test data set? This is not clear from the sentence below from the abstract: "On a standard benchmark data set and independent set, it achieves 86.48% accuracy and 90.67% sensitivity and significantly outperforms state-of-the-art methods." Use 'respectively' would help.
In sentence "Another method was presented by 10 for prediction of C-to-U editing sites using biochemical and evolutionary information." does '10' refer to a reference? I found similar issues at other locations in the manuscript. Please review the entire manuscript for reference style.
Correct sentence grammar "The datasets that we have used is first used in 11 based on the work in 12.". Replace 'is' with 'was'.
Is there a specific reason why independent test dataset does not contain negative samples?
Authors should to clearly specify what kind of features were extracted from the distribution of A, C, G, and U. Also how did they obtain the distribution of the A, C, G, and U.
Please check the grammar and remove the redundant words from this sentence: "Our feature hybrid feature selection method is a multi- step step method where the first step is a wrapper method followed by consecutive filter methods."
It will be better if authors put some more light on feature selection approach. For example, how add or delete of one feature at a time works? Also, this sentence "In the last step, we used ensemble of random decision trees and added most significant features found by the randomly generated features which were left in the previous step to get the final set of features." is difficult to read and understand. This will be helpful in replication of this work.
How many features were finally selected after the Hybrid feature selection? Authors need to include this information in the Hybrid Feature Selection section of the manuscript.
What were the value of parameters for the incremental decision tree algorithm? Please specify the values of all the parameters that were used.
Authors are comparing their work with PAI and PAI-SAE which were published in year 2016 and year 2018. Doing a quick google search, I found a method EPAI-NC which was published in year 2019 that utilizes Locally Deep Support Vector Machines and same benchmark and independent test datasets. Compared to EPAI-NC the proposed PRESa2i shows poor performance. Authors need to compare their method with EPAI-NC as this is a method published in the recent past.
Authors should also compare their method with PAI-SAE and EPAI-NC on the independent test dataset as PAI-SAE and EPAI-NC are more recently published method compared to PAI.

Is the rationale for developing the new method (or application) clearly explained?

Yes
Is the description of the method technically sound?

Yes
Are sufficient details provided to allow replication of the method development and its use by others?

Partly
If any results are presented, are all the source data underlying the results available to ensure full reproducibility?

Yes
Are the conclusions about the method and its performance adequately supported by the findings presented in the article?

No

Competing Interests: No competing interests were disclosed.

Reviewer Expertise: Artificial Intelligence and Machine Learning, Data Science, Bioinformatics, Computational and Structural Biology, Biomedical Informatics, Evolutionary Computation, Information Retrieval and Text Mining, and Big Data.

I confirm that I have read this submission and believe that I have an appropriate level of expertise to state that I do not consider it to be of an acceptable scientific standard, for reasons outlined above.

CITE

Report a concern

Respond or Comment

Comments on this article Comments (0)

Version 1

VERSION 1 PUBLISHED 16 Apr 2020

Open Peer Review

Reviewer Status

Reviewer Reports

	Invited Reviewers
	1	2	3
Version 1 16 Apr 20	read	read	read

Avdesh Mishra, Texas A&M University - Kingsville, Kingsville, USA
Shizuka Uchida, Aalborg University, Aalborg, Denmark
Xiaoyong Pan, Shanghai Jiao Tong University, Shanghai, China

Comments on this article

All Comments(0)

Add a comment

Sign up for content alerts

Browse by related subjects

Back to all reports

Reviewer Report

4 Views

17 Aug 2021 | for Version 1

Xiaoyong Pan, Institute of Image Processing and Pattern Recognition, Shanghai Jiao Tong University, Shanghai, China

4 Views Cite this report Responses(0)

Approved With Reservations

The authors present a computational method for predicting A-to-I RNA editing sites using sequence-based features and incremental decision trees, and they show that the proposed method yields better performance. However, there are limited novelties in this manuscript.

How is sequence similarity between the training set and test set? The redundant sequences in the test set should be removed.
Deep learning has been widely used in genomic sequences. The authors should compare the proposed method with the CNN-based method which uses a one-hot encoded sequence as input.
In Table 3, for the independent test set, why is only sensitivity reported? Are other metrics of the proposed method worse than the baseline method? It is bad to only keep positive samples in the independent test set.
In Figure 2, the AUC value should be added.
How is the performance of XGBoost on this task? This classifier has been shown to perform well on many tasks.

Is the rationale for developing the new method (or application) clearly explained?

Partly
Is the description of the method technically sound?

Partly
Are sufficient details provided to allow replication of the method development and its use by others?

Partly
If any results are presented, are all the source data underlying the results available to ensure full reproducibility?

Partly
Are the conclusions about the method and its performance adequately supported by the findings presented in the article?

Partly

Competing Interests

No competing interests were disclosed.

Reviewer Expertise

Bioinformatics, RNA-protein interaction prediction

I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.

Respond to this report

Responses (0)

Back to all reports

Reviewer Report

5 Views

21 Jun 2021 | for Version 1

Shizuka Uchida, Center for RNA Medicine, Department of Clinical Medicine, Aalborg University, Aalborg, Denmark

5 Views Cite this report Responses(0)

Not Approved

In this manuscript, the authors present a new method to predict A-to-I RNA editing sites. This method is implemented in a web interface so that users can access freely. The method proposed is technically sound and valid, yet the rationale for developing such method is hampered by the fact that there are many bioinformatic tools available to detect (not predict) A-to-G changes from RNA sequencing (RNA-seq) data. Thus, the usage of the author’s web interface is not of high priority to researchers; given that the cost for RNA-seq is not high anymore. More specific comments are listed below:

Major points

The authors must benchmark the performance of their tool to the known RNA editing sites deposited in databases, such as DARNED, RADAR, and REDIportal.
The authors must provide the results of other species, as most of the current research in A-to-I RNA editing is focused on human diseases.
This web tool is not functioning. When the known RNA edited sequence, TTCTCAACCTCCTGGGCTCTTGTGATCCTCCTGTCTTA (from PMID: 27595325), was pasted onto the field, Enter the query RNA sequences here, and the [Predict] button is pressed, the following message appears: “Login Failed”. Thus, this reviewer cannot evaluate the actual performance of this tool that the authors developed.

Minor points

(1) From the URL (http://brl.uiu.ac.bd/presa2i/index.php#), nothing happens when “ • Don't have any RNA string click here.” is pressed.

Is the rationale for developing the new method (or application) clearly explained?

Partly
Is the description of the method technically sound?

Yes
Are sufficient details provided to allow replication of the method development and its use by others?

Yes
If any results are presented, are all the source data underlying the results available to ensure full reproducibility?

Yes
Are the conclusions about the method and its performance adequately supported by the findings presented in the article?

Yes

References

1. Stellos K, Gatsiou A, Stamatelopoulos K, Perisic Matic L, et al.: Adenosine-to-inosine RNA editing controls cathepsin S expression in atherosclerosis by enabling HuR-mediated post-transcriptional regulation.Nat Med. 22 (10): 1140-1150 PubMed Abstract | Publisher Full Text

Competing Interests

No competing interests were disclosed.

Reviewer Expertise

bioinformatics, biochemistry, developmental biology, molecular biology, RNA biology, non-coding RNAs, miRNAs, lncRNAs, circRNAs, A-to-I RNA editing, epitranscriptomics, RNA sequencing

I confirm that I have read this submission and believe that I have an appropriate level of expertise to state that I do not consider it to be of an acceptable scientific standard, for reasons outlined above.

Respond to this report

Responses (0)

Back to all reports

Reviewer Report

23 Views

19 May 2020 | for Version 1

Avdesh Mishra, Department of Electrical Engineering and Computer Science, Texas A&M University - Kingsville, Kingsville, TX, USA

23 Views Cite this report Responses(0)

Not Approved

In this paper, authors present PRESa2i, a computation prediction tool for identification of A-to-I RNA editing sites in given RNA sequences. PRESa2i uses a simple, yet effective set of sequence based features generated from RNA sequences and a novel feature selection technique. It uses an incremental decision tree algorithm as the classification algorithm.

Below are my comments and suggestions:

Authors should specify which method the proposed method outperforms "significantly outperforms state-of-the-art methods" is a very general statement. In near future there will be another method that would outperform this method.
I did not find any example RNA sequence when I clicked on "click here" link on the authors webserver.
The webserver is not working. I tried to run the webserver with an example RNA Sequence. But I got "Login Failed" message.
Does authors method achieves 86.48% accuracy and 90.67% sensitivity on benchmark or independent test data set? This is not clear from the sentence below from the abstract: "On a standard benchmark data set and independent set, it achieves 86.48% accuracy and 90.67% sensitivity and significantly outperforms state-of-the-art methods." Use 'respectively' would help.
In sentence "Another method was presented by 10 for prediction of C-to-U editing sites using biochemical and evolutionary information." does '10' refer to a reference? I found similar issues at other locations in the manuscript. Please review the entire manuscript for reference style.
Correct sentence grammar "The datasets that we have used is first used in 11 based on the work in 12.". Replace 'is' with 'was'.
Is there a specific reason why independent test dataset does not contain negative samples?
Authors should to clearly specify what kind of features were extracted from the distribution of A, C, G, and U. Also how did they obtain the distribution of the A, C, G, and U.
Please check the grammar and remove the redundant words from this sentence: "Our feature hybrid feature selection method is a multi- step step method where the first step is a wrapper method followed by consecutive filter methods."
It will be better if authors put some more light on feature selection approach. For example, how add or delete of one feature at a time works? Also, this sentence "In the last step, we used ensemble of random decision trees and added most significant features found by the randomly generated features which were left in the previous step to get the final set of features." is difficult to read and understand. This will be helpful in replication of this work.
How many features were finally selected after the Hybrid feature selection? Authors need to include this information in the Hybrid Feature Selection section of the manuscript.
What were the value of parameters for the incremental decision tree algorithm? Please specify the values of all the parameters that were used.
Authors are comparing their work with PAI and PAI-SAE which were published in year 2016 and year 2018. Doing a quick google search, I found a method EPAI-NC which was published in year 2019 that utilizes Locally Deep Support Vector Machines and same benchmark and independent test datasets. Compared to EPAI-NC the proposed PRESa2i shows poor performance. Authors need to compare their method with EPAI-NC as this is a method published in the recent past.
Authors should also compare their method with PAI-SAE and EPAI-NC on the independent test dataset as PAI-SAE and EPAI-NC are more recently published method compared to PAI.

Is the rationale for developing the new method (or application) clearly explained?

Yes
Is the description of the method technically sound?

Yes
Are sufficient details provided to allow replication of the method development and its use by others?

Partly
If any results are presented, are all the source data underlying the results available to ensure full reproducibility?

Yes
Are the conclusions about the method and its performance adequately supported by the findings presented in the article?

No

Competing Interests

No competing interests were disclosed.

Reviewer Expertise

Artificial Intelligence and Machine Learning, Data Science, Bioinformatics, Computational and Structural Biology, Biomedical Informatics, Evolutionary Computation, Information Retrieval and Text Mining, and Big Data.

I confirm that I have read this submission and believe that I have an appropriate level of expertise to state that I do not consider it to be of an acceptable scientific standard, for reasons outlined above.

Respond to this report

Responses (0)

[1] 1. Rueter SM, Dawson TR, Emeson RB: Regulation of alternative splicing by RNA editing. Nature. 1999; 399(6731): 75–80. PubMed Abstract | Publisher Full Text

[2] 2. Maas Stefan: Gene regulation through RNA editing. Discov Med. 2010; 10(54): 379–386. PubMed Abstract

[3] 3. Paz N, Levanon EY, Amariglio N, et al.: Altered adenosine-to-inosine RNA editing in human cancer. Genome Res. 2007; 17(11): 1586–1595. PubMed Abstract | Publisher Full Text | Free Full Text

[4] 4. Wang Z, Gerstein M, Snyder M: RNA-Seq: a revolutionary tool for transcriptomics. Nat Rev Genet. 2009; 10(1): 57. PubMed Abstract | Publisher Full Text | Free Full Text

[5] 5. Thompson J, Gopal S: Genetic algorithm learning as a robust approach to RNA editing site prediction. BMC Bioinformatics. 2006; 7(1): 145. PubMed Abstract | Publisher Full Text | Free Full Text

[6] 6. Chen W, Feng P, Yang H, et al.: iRNA-AI: identifying the adenosine to inosine editing sites in RNA sequences. Oncotarget. 2017; 8(3): 4208–4217. PubMed Abstract | Publisher Full Text | Free Full Text

[7] 7. Xiong H, Liu D, Li Q, et al.: RED-ML: a novel, effective RNA editing detection method based on machine learning. GigaScience. 2017; 6(5): 1–8. PubMed Abstract | Publisher Full Text | Free Full Text

[8] 8. Xiao X, Wang P, Xu Z, et al.: Pai-sae: Predicting adenosine to inosine editing sites based on hybrid features by using spare autoencoder. In IOP Conference Series: Earth and Environmental Science. 2018; 170(5): 052018. IOP Publishing. Publisher Full Text

[9] 9. Lenz H, Knoop V: PREPACT 2.0: Predicting C-to-U and U-to-C RNA Editing in Organelle Genome Sequences with Multiple References and Curated RNA Editing Annotation. Bioinform Biol Insights. 2013; 7: 1–19. PubMed Abstract | Publisher Full Text | Free Full Text

[10] 10. Du P, Li Y: Prediction of C-to-U RNA editing sites in plant mitochondria using both biochemical and evolutionary information. J Theor Biol. 2008; 253(3): 579–586. PubMed Abstract | Publisher Full Text

[11] 11. Chen W, Feng P, Ding H, et al.: PAI: Predicting adenosine to inosine editing sites by using pseudo nucleotide compositions. Sci Rep. 2016; 6: 35123. PubMed Abstract | Publisher Full Text | Free Full Text

[12] 12. St Laurent G, Tackett MR, Nechkin S, et al.: Genome-wide analysis of A-to-I RNA editing by single-molecule sequencing in Drosophila. Nat Struct Mol Biol. 2013; 20(11): 1333–9. PubMed Abstract | Publisher Full Text

[13] 13. Chen W, Feng P, Yang H, et al.: iRNA-3typeA: Identifying Three Types of Modification at RNA's Adenosine Sites. Mol Ther Nucleic Acids. 2018; 11: 468–474. PubMed Abstract | Publisher Full Text | Free Full Text

[14] 14. Chou KC: Some remarks on protein attribute prediction and pseudo amino acid composition. J Theor Biol. 2011; 273(1): 236–247. PubMed Abstract | Publisher Full Text | Free Full Text

[15] 15. Rahman S, Aktar U, Jani R, et al.: iPromoter-FSEn: Identification of bacterial σ⁷⁰ promoter sequences using feature subspace based ensemble classifier. Genomics. 2019; 111(5): 1160–1166. PubMed Abstract | Publisher Full Text

[16] 16. Rayhan F, Ahmed S, Shatabda S, et al.: iDTI-ESBoost: Identification of Drug Target Interaction Using Evolutionary and Structural Features with Boosting. Sci Rep. 2017; 7(1): 17731. PubMed Abstract | Publisher Full Text | Free Full Text

[17] 17. Chowdhury SY, Shatabda S, Dehzangi A: iDNAProt-ES: Identification of DNA-binding Proteins Using Evolutionary and Structural Features. Sci Rep. 2017; 7(1): 14938. PubMed Abstract | Publisher Full Text | Free Full Text

[18] 18. Uddin MR, Sharma A, Farid DM, et al.: EvoStruct-Sub: An accurate Gram-positive protein subcellular localization predictor using evolutionary and structural features. J Theor Biol. 2018; 443: 138–146. PubMed Abstract | Publisher Full Text

[19] 19. Yu Y, Zhou H, Kong Y, et al.: The Landscape of A-to-I RNA Editome Is Shaped by Both Positive and Purifying Selection. PLoS Genet. 2016; 12(7): e1006191. PubMed Abstract | Publisher Full Text | Free Full Text

[20] 20. Shatabda S, Saha S, Sharma A, et al.: iPHLoc-ES: Identification of bacteriophage protein locations using evolutionary and structural features. J Theor Biol. 2017; 435: 229–237. PubMed Abstract | Publisher Full Text

[21] 21. Al Maruf MA, Shatabda S: iRSpot-SF: Prediction of recombination hotspots by incorporating sequence based features into Chou's Pseudo. Genomics. 2019; 111(4): 966–972. PubMed Abstract | Publisher Full Text

[22] 22. Ghandi M, Mohammad-Noori M, Beer MA: Robust k-mer frequency estimation using gapped k-mers. J Math Biol. 2014; 69(2): 469–500. PubMed Abstract | Publisher Full Text | Free Full Text

[23] 23. Chang J, Su EC, LO A, et al.: PSLDoc: Protein subcellular localization prediction based on gapped-dipeptides and probabilistic latent semantic analysis. Proteins. 2008; 72(2): 693–710. PubMed Abstract | Publisher Full Text

[24] 24. Hulten G, Spencer L, Domingos P: Mining time-changing data streams. In ACM SIGKDD Intl. Conf. on Knowledge Discovery and Data Mining. ACM Press. 2001; 97–106. Publisher Full Text

[25] 25. Hall M, Frank E, Holmes G, et al.: The weka data mining software: an update. ACM SIGKDD explorations newsletter. 2009; 11(1): 10–18. Publisher Full Text

[26] 26. Chen XX, Tang H, Li WC, et al.: Identification of Bacterial Cell Wall Lyases via Pseudo Amino Acid Composition. Biomed Res Int. 2016; 2016: 1654623. PubMed Abstract | Publisher Full Text | Free Full Text

[27] 27. swakkhar: swakkhar/RNAEditing:Pres-a2i (Version v1.1). Zenodo. 2020. http://www.doi.org/10.5281/zenodo.3727221

PRESa2i: incremental decision trees for prediction of Adenosine to Inosine RNA editing sites

Abstract

Keywords

Introduction

Methods

Datasets

Table 1. Summary of two datasets used in this study.

RNA sequence representation

Hybrid feature selection

Figure 1. Steps of hybrid feature selection technique used by the PRESa2i method.

Incremental decision trees

Performance evaluation

Results and discussion

Comparison with other methods

Table 2. Performance comparison of different methods - PRESa2i, PAI11 and PAI-SAE8 - on the benchmark dataset.

Table 3. Performance comparison of different meth- ods - PRESa2i, PAI11 - on the independent test set.

Effects of feature selection

Table 4. Performance of classifiers in different stages of feature selection method on the benchmark dataset.

Classifier selection

Figure 2. Receiver operating characteristic curve for different classifiers on the benchmark dataset.

Table 5. Performance of different classifiers in terms of sensitivity on the independent test set.

Effect of negative test data

Table 6. Experiments on the independent dataset after introduction of negative instances.

Web server

Implementation

Operation

Figure 3. Screenshot of the web application developed for PRESa2i.

Conclusion

Data availability

Software availability

References

Comments on this article Comments (0)

Open Peer Review

Comments on this article Comments (0)

Open Peer Review

Reviewer Status

Reviewer Reports

Comments on this article

Browse by related subjects

Competing Interests Policy

Stay Updated

Table 2. Performance comparison of different methods - PRESa2i, PAI¹¹ and PAI-SAE⁸ - on the benchmark dataset.

Table 3. Performance comparison of different meth- ods - PRESa2i, PAI¹¹ - on the independent test set.