An Advanced Cognitive Approach for Personalized Drug Formulation in Type 2 Diabetes Using Deep Neural Networks and Feature-based Fusion

Personalized medicine is an innovative approach toward achieving improved therapeutic outcomes, notably in chronic diseases like Type 2 Diabetes (T2D), where individualised drug formulations represent the crux of treatment. Conventional methods of drug prescription tend to focus on generic approaches without considering the specific genetic, physiological, and lifestyle characteristics of individuals. This research fills this gap by developing an advanced deep learning-based method for personalised drug formulation in T2D. It integrates diverse patient data, including demographics, medical history, clinical measurements, genomic information, pharmacokinetic information, real-time monitoring data, and responses to previous treatments. For optimal drug formulation, it makes use of a combination of deep learning models, such as Recurrent Neural Network (RNN) for time-series prediction, Convolutional Neural Network (CNN) for genomic analysis, GAN for producing synthetic data, and auto-encoder for feature extraction. The dataset consists of 2700 patient records collected over one year, wherein the pre-processing phase includes handling missing values, normalising continuous variables, and encoding categorical features. Trained models were assessed based on precision, recall, F1-score, and AUC ROC levels; the best performance came from the RNN model with a maximum accuracy of 97.35%. CNN, GAN, and Autoencoder also obtained similar competitive results. However, the RNN-based model received the highest score in recommending the appropriate drugs. The application of this research can contribute to the broader discipline of precision medicine.