Nucleobase discrimination based on terahertz spectroscopy using multi-scale convolutional neural network with convolutional block attention module and long short-term memory

In this study, we propose an integration of terahertz time-domain spectroscopy (THz-TDS) with a multi-scale convolutional neural network incorporating a convolutional block attention module and long short-term memory (MsCNN-CBAM-LSTM) for accurate nucleobase discrimination. The THz-TDS system captures unique spectral fingerprints of nucleobases, which are inherently complex and difficult to distinguish using conventional methods. The proposed MsCNN-CBAM-LSTM algorithm is specifically designed to process these complex THz spectral data, leveraging multi-scale feature extraction, attention mechanisms, and temporal modeling to achieve superior discrimination accuracy. Experimental results demonstrate that the integration of THz-TDS and MsCNNCBAM-LSTM achieves a remarkable accuracy of 99.17 %, outperforming other state-of-the-art models. This work not only highlights the synergy between advanced spectroscopic techniques and deep learning but also provides a robust framework for biochemical analysis with potential applications in diagnostics and molecular sensing.