Efficient DNA Mutation Classification Using Deep Active Learning Techniques

Abstract

Deep neural networks have shown impressive performance in a variety of fields, but the challenge in the training procedure for these networks may be time-consuming, particularly when working with huge datasets. Active learning is a potentially useful method that may speed up the training process by prioritizing the selection of samples that include the most relevant information for annotation. In this study, we propose an architecture for deep neural networks that combines active learning with Long Short-Term Memory (LSTM) in order to improve the effectiveness of the training process for DNA mutation categorization which is our contribution. We test the suggested method on the Cancer Cell Lines Encyclopedia (CCLE) dataset and assess its performance in comparison to a deep network and an LSTM network that do not use active learning. According to our experiments, the proposed approach is more accurate and improves efficiency by significantly reducing training time compared to other methods. When comparing deep learning models with deep active models, we found that the former averages a much higher training time of 4859.678 seconds while the latter averaged a substantially lower first training epoch of 978.8522 seconds, making total prediction and training time approximately 1474.385 seconds for the entire first phase. Focusing on some aspects outside of model creation as well, our research demonstrates the promise active learning holds in speeding up the training process for deep networks in context of classifying Deoxyribonucleic acid DNA mutations and emphasizes important aspects needed in constructing effective deep learning models.