Background: One of the most prevalent epigenetic alterations in all three kingdoms of life is 5mC, which plays a part in a wide range of biological functions. Although in-vitro techniques are more effective in detecting epigenetic alterations, they are time and money-intensive. Artificial intelligence-based in silico approaches have been used to overcome these obstacles. background: One of the most prevalent epigenetic alterations in all three kingdoms of life is 5mC, which plays a part in a wide range of biological functions. Although in-vitro techniques are more effective in detecting epigenetic alterations, they are time and money intensive. Artificial intelligence-based in silico approaches have been used to overcome these obstacles. Aim: This study aimed to develop an ML-based predictor for the detection of 5mC sites in Poaceae. Objective: The objective of this study was the evaluation of machine learning and deep learning models for the prediction of 5mC sites in rice. Method: In this study, the vectorization of DNA sequences has been performed using three distinct feature sets- Oligo Nucleotide Frequencies (k = 2), Mono-nucleotide Binary Encoding, and Chemical Properties of Nucleotides. Two deep learning models, long short-term memory (LSTM) and Bidirectional LSTM (Bi-LSTM), as well as nine machine learning models, including random forest, gradient boosting, naïve bayes, regression tree, k-Nearest neighbour, support vector machine, adaboost, multiple logistic regression, and artificial neural network, were investigated. Also, bootstrap resampling was used to build more efficient models along with a hybrid feature selection module for dimensional reduction and removal of irrelevant features of the vector space. Result: Random Forest gains the maximum accuracy, specificity and MCC, i.e., 92.6%, 86.41% and 0.84. Gradient Boosting obtained the maximum sensitivity, i.e., 96.85%. The Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) technique showed that the best three models were Random Forest, Gradient Boosting, and Support Vector Machine in terms of accurate prediction of 5mC sites in rice. We developed an R-package, ‘GB5mCPred,’ and it is available in CRAN (https://cran.r-project.org/web/packages/GB5mcPred/index.html). Also, a user-friendly prediction server was made based on this algorithm (http://cabgrid.res.in:5474/). Conclusion: With nearly equal TOPSIS scores, Random Forest, Gradient Boosting, and Support Vector Machine ended up being the best three models. The major rationale may be found in their architectural design since they are gradual learning models that can capture the 5mC sites more correctly than other learning models.
Read full abstract