Data-Driven Classification of Solubility Space in Deep Eutectic Solvents: Deciphering Driving Forces Using PCA and K-Means Clustering

This study presents a robust, data-driven framework for classifying and predicting drug solubility in deep eutectic solvents (DESs), moving beyond empirical approaches to enable rational formulation design. By analyzing 2010 solubility measurements of 21 diverse pharmaceutical compounds across numerous choline chloride, betaine, and menthol-based DESs, we employed Principal Component Analysis to reduce 16 COSMO-RS-derived descriptors into four chemically interpretable dimensions explaining 86.7% of the total variance. Persistence analysis confirmed component stability, revealing two key factors: PC1 (global solvation propensity, i.e., the overall capacity of the solvent to stabilize solutes through all interaction types) and PC2 (specific interaction complementarity, i.e., the degree of matching between solute and solvent hydrogen-bonding/polarity features). K-means clustering identified four distinct solubility regimes: high-solubility DES-optimized systems (Cluster 1), reliable moderate performers (Cluster 0), intermediate candidates for optimization (Cluster 3), and fundamentally challenging combinations (Cluster 2). Comparative analysis demonstrated choline chloride's broad utility while revealing specialized roles for menthol and betaine in specific chemical spaces. Case studies of Sulfasalazine and Caffeine illustrated how multi-cluster distributions guide formulation strategies, distinguishing precision-requiring from forgiving compounds. This taxonomy provides formulation scientists with a rational framework for DES selection, emphasizing aqueous modification, HBD and HBA diversity, and balanced solvation-interaction optimization. The integrated PCA-clustering approach transforms DES development from trial-and-error screening to targeted design, offering fundamental insights into solubility mechanisms while accelerating sustainable pharmaceutical formulation.