Diabetes mellitus detection using deep learning on DNA sequences

Diabetes mellitus, a chronic and incurable disease characterized by insufficient insulin, is a significant global health problem. It is responsible for many deaths yearly, with around 2-5 million people dying from this disease yearly. Detecting and treating diabetes in its early stages is crucial to preventing complications and reducing the risk of severe health problems. This research project aims to develop a method to identify diabetes by analyzing the biomarkers present in DNA sequences. DNA sequences are highly complex and extensive biological data. Therefore, the utilization of Deep Learning methods can effectively process this data due to their ability to comprehend intricate and non-linear patterns. This capability in recognizing intricate patterns assists in extracting features to obtain relevant characteristics, thereby enhancing classification outcomes. In this study, we converted DNA sequences to a numeric value. We used entropy-based mapping techniques to convert DNA sequences into numerical values. This technique involves using the fractional derivative of the Shannon Entropy. We generate spectrogram images as a biomarker on DNA sequences by applying the specgram function from the matplotlib library using the Short-Time Fourier Transform (STFT). These images represent a three-dimensional form namely: time, frequency, and amplitude that represented by a color scale. Futhermore, we extracted the spectrogram image using the Transformer module to obtain the relevant features from the spektogram images of diabetes mellitus patient. The features obtained from the Transformer module are then classified using Support Vector Machine (SVM) to determine whether someone has diabetes mellitus. Our experiments show that SVM achieves an impressive accuracy rate of 99.89% in classifying 100 features that represent DNA sequences. Overall, using this method shows the potential of using deep learning techniques, especially the Transformer module, in identifying diabetes based on DNA biomarkers. This approach involves transforming DNA sequences into spectrogram images and using advanced classification algorithms.