Machine Learning-Guided Prediction of Formulation Performance in Inhalable Ciprofloxacin–Bile Acid Dispersions with Antimicrobial and Toxicity Evaluation

Ciprofloxacin (CFX)
is a potent antibiotic for respiratory infections,
but its poor solubility and high crystallinity limit its effectiveness
in dry powder inhaler (DPI) delivery. Although soluble forms such
as CFX hydrochloride are available, their rapid dissolution may lead
to systemic absorption, undermining localized lung targeting. To address
this, we developed solid dispersions of CFX with primary bile acids,
namely, cholic acid (CA) and chenodeoxycholic acid (CDA), using spray
drying and ball milling to enhance solubility in a controlled manner
while maintaining deposition in the lungs. Differential scanning calorimetry
showed glass-transition temperature (Tg) values were elevated for both bile acids, with CA dispersions showing
slightly higher absolute values (114.16–131.77 °C vs 109.13–120.67
°C). However, Fourier transform infrared and dissolution data
indicated that CDA formed stronger directional hydrogen bonding with
CFX. X-ray diffraction confirmed partially amorphous dispersions with
minimal residual crystallinity. Solubility enhancement was observed
for both bile acids, showing slightly higher values with CA dispersions.
Aerodynamic assessments using an Andersen cascade impactor revealed
improved lung deposition with CFX–CDA, with a higher fine particle
fraction (FPF: 30.81%) and lower mass median aerodynamic diameter
(MMAD: 5.89 μm) compared to CFX–CA (FPF: 26.93%, MMAD:
6.19 μm). The emitted dose was highest in CDA with nearly 5
mg compared to CA dispersions (∼3 mg). In vitro antimicrobial
studies showed that dispersions maintained comparable antimicrobial
activity to pure CFX, while in vivo toxicology in rats indicated mild,
dose-dependent hepatic changes. CDA formulations showed AST elevation
at a low dose and ALP increase at a high dose, consistent with the
known hepatic effects of this bile acid, while CA formulations were
broadly comparable to pure CFX. Machine learning algorithms, including
tree-based models and neural networks, were used to predict the formulation
performance and identify critical variables. Feature selection was
achieved using recursive elimination, and permutation analysis showed
that the bile acid type, inlet temperature, and molar ratio were the
most influential predictors of solubility and lung deposition. Models
such as gradient boosting and elastic net showed a high predictive
accuracy (R2 > 0.85). Overall, this
study
highlights the potential of primary bile acid-based DPI formulations
as effective inhalable antibiotic therapies.