A Comparative Analysis of Machine Learning and Deep Learning Approaches for Circular RNA Classification

Circular RNAs (circRNAs) have emerged as crucial regulators of gene expression and potential biomarkers in various biological processes and diseases. However, accurately classifying and analyzing circRNAs necessitates effective representation of their sequences, a challenge that has led to significant research gaps. To address this, the present research project aims to explore and compare the performance of different methods and models, encompassing both machine learning algorithms and deep learning architectures, for sequence classification tasks. The project employs the k-mer extraction technique and the DNA2Vec algorithm to achieve vector representation or sequence embedding. These techniques enable the conversion of DNA sequences into numerical representations suitable for processing by machine learning and deep learning algorithms. A publicly available dataset is utilized for training the models, and their performance is assessed using metrics such as accuracy, precision, recall, and F1 score. Several machine learning algorithms, including K Nearest Neighbour, Random Forest, Logistic Regression, Naive Bayes, SVM (Linear), and SVM (Polynomial), are tested for classifying circRNA and IncRNA sequences. These algorithms exhibit varying levels of performance in accurately categorizing the sequences. Notably, the SVM (Polynomial) algorithm achieves the highest accuracy of 89.4%, demonstrating its proficiency in classifying the sequences with precision. In contrast, the deep learning algorithm ACNN-BLSTM, which combines a convolutional neural network with a recurrent neural network, surpasses the baseline models with an accuracy of 94.0%. This highlights the importance of incorporating both local and sequential information for accurate circRNA prediction. The findings shed light on the strengths and limitations of each model, providing valuable insights into their suitability for circRNA classification and functional analysis. The study emphasizes the significance of selecting appropriate sequence embedding techniques based on the dataset's characteristics and the specific requirements of the analysis task.