- Article
Background: Drug stability is a fundamental issue in drug-quality monitoring during storage. Methods: This study proposes a computational framework for degradation and potency estimation of pharmaceuticals using Raman spectroscopy. The absence of complete longitudinal Raman degradation data with experimental potencies led to the use of real Raman spectra as baseline spectra, which were then converted using physics-informed spectral degradation to create four stages (T0–T3). A modified 1D-Convolutional Neural Network (1D-CNN) comprising multi-scale convolution, residual learning, and spectral attention was designed to simultaneously classify the degradation stage and predict the remaining Active Pharmaceutical Ingredient potency. External computational testing of an independent set of commercial pharmaceutical dosage forms was performed using Raman data. The predicted potency profiles were further analyzed with an approximate zero-order kinetic model to estimate the degradation rate constant and the computational shelf-life . Results: The proposed modified 1D-CNN achieved better performance and an value of 99.23 ± 0.02 for potency prediction, outperforming Support Vector Machine, Random Forest, and XGBoost models. From kinetic analysis, the value was 0.98, the degradation rate constant was 5.78, and the calculated shelf-life was 1.76 normalized time units. Conclusions: The proposed framework can thus be considered as a computational proof-of-concept for Raman-based pharmaceutical degradation and potency assessment.
17 November 2026






![Chemical structure of LUT [17].](https://mdpi-res.com/biophysica/biophysica-06-00092/article_deploy/html/images/biophysica-06-00092-g001-550.jpg)



