Terahertz Spectroscopy-Enabled Smart Monitoring of Drug Salt Synthesis: A Case Study on 8-Hydroquinolinium Salicylate

Abstract

While salt formation is a proven strategy to overcome poor aqueous solubility and facilitate the development of active pharmaceutical ingredients, real-time monitoring of this process still lacks molecular-resolution process analytical technology (PAT). Using 8-hydroxyquinolinium salicylate as a model system, this study integrates cryogenic terahertz spectroscopy, Born–Oppenheimer molecular dynamics (BOMD), and interpretable deep learning to develop a terahertz time-domain spectroscopy (THz-TDS) workflow for stage-resolved monitoring of pharmaceutical salt formation. THz measurements (80 ∼300 K) revealed the anharmonic character of the lattice vibrations, whose dominant features were reproduced and assigned by BOMD simulations. Umbrella sampling indicated a nearly barrierless proton-transfer process (ΔG‡ = 1.7 kcal/mol) with picosecond-scale shuttling dynamics. Statistical analysis of ten kinetic models confirms first-order conversion (AIC = −151.2; n = 0.99 ± 0.02) and yielded an Arrhenius activation energy of 45.2 ± 2.3 kJ/mol for predictive process optimization. Dissolution data (n = 3; mean ± SD) showed a statistically significant 2.0–2.6-fold enhancement across pH 1.2, 6.8, and 7.4, as assessed by Welch’s t test and two-way ANOVA with Tukey’s HSD test (p < 0.001). The interpretable CNN achieved 98.2% hold-out test accuracy and used SHAP to link decisions to chemical fingerprint bands. Overfitting was evaluated through spectral dimensionality reduction, dropout, a stratified training/test split, 5-fold cross-validation, leave-one-batch-out cross-validation, and external batches under shifted thermal conditions, with accuracy retained at ≥ 92.7%. These results support a proof-of-concept, interpretable THz-TDS PAT workflow with molecular-level resolution.