A Review of Deep Learning-Based Approaches for Detection and Diagnosis of Diverse Classes of Drugs
Kumar, Ashish,Kuriakose, Jeril,Kumar, Yogesh
DOI: https://doi.org/10.1007/s11831-023-09936-7
IF: 9.7
2023-05-10
Archives of Computational Methods in Engineering
Abstract:Artificial intelligence-based drug discovery has gained attention lately since it drastically cuts the time and money needed to produce new treatments. In recent years, a vast quantity of data in various formats has been made accessible in the medical field to analyse different health complications. Drug discovery aims to uncover possible novel medications using a multidisciplinary approach that includes biology, chemistry, and pharmacology. Traditional sentiment analysis methods count or repeat words in a text assigned sentiment ratings by an expert. Several outdated, ineffective old methodologies are utilized to forecast drug design and discovery. However, with the development of DL (deep learning), the traditional drug discovery method has been further simplified. In this work, we applied deep learning models, such as LSTM (Long short-term memory), GRU (Gated recurrent units), Bidirectional LSTM (BiLSTM), Bidirectional GRU(BiGRU), SimpleRNN, embedding + LSTM, embedding + GRU, embedding + GRU + dropout, embedding + conv1d + LSTM, and Embedding + Conv1d + GRU on a dataset of drug reviews. Furthermore, we used Adam and RMSprop, two optimizers, for each model, for increased optimization. This research focuses on categorizing medication reviews into positive and negative categories. The effectiveness of the different deep learning models was assessed using a wide range of performance measures. Experiments demonstrated that the GRU (Gated Recurrent Unit) generated exceptional validation dataset results. In addition, this study emphasizes the relevance of deep learning methods over traditional learning approaches in categorization.
computer science, interdisciplinary applications,engineering, multidisciplinary,mathematics