Eleonore Fröhlich1, 2, Noon Sharafeldin2, Stefan Mitsche3,4, Hartmuth Schröttner3,4 & Sarah Zellnitz-Neugebauer2
1Center for Medical Research, Medical University of Graz, Stiftingtalstr. 24, Graz, 8010, Austria
2RCPE GmbH, Inffeldgasse 13, Graz, 8010, Austria
3Institute of Electron Microscopy and Nanoanalysis (FELMI), Graz University of Technology, Steyrergasse 17, Graz, 8010, Austria
4 Graz Centre for Electron Microscopy (ZFE), 8010 Graz, Steyrergasse 17, Austria
Summary
Machine learning (ML) offers a promising strategy for accelerating pharmaceutical development by enabling data-driven prediction of viable formulations. In this study, we expand on prior work by applying ML to predict co-amorphous systems (COAMS) suitable for inhalation-based treatment of tuberculosis (TB). COAMS, consisting of two active pharmaceutical ingredients (APIs) in a single amorphous phase, can enhance solubility, stability, and dose consistency—key factors in pulmonary drug delivery. Rifampicin-Ethambutol (RIF-ETH) and Rifampicin-Moxifloxacin (RIF-MOX) were selected based on ML predictions. Both combinations were prepared via co-milling and spray-drying at 1:1 molar ratios and adjusted to therapeutically relevant ratios (1:6 for RIF-ETH; 1:1.25 for RIF-MOX) to reflect inhalation dosing requirements. X-ray diffraction confirmed amorphous character for RIF-MOX across all conditions, while RIF-ETH showed incomplete amorphization at 1:6. Particle size analysis and SEM imaging indicated spray-drying produced smaller, more spherical particles, though RIF-ETH (1:6) spray-dried particles exhibited poor separation. Aerodynamic assessment of RIF-MOX (1:1.25) formulations demonstrated superior performance for the spray-dried system, with an emitted fraction (Ef) of 85.2% for MOX and 77.5% for RIF and a fine particle fraction (FPF) of 78.2% for MOX and 90.0% for RIF, outperforming both co-milled formulation and physical mixture. Future work will focus on evaluating dissolution behaviour for the COAMS.