*Corresponding Author: Dr. Altaf O. Mulani, Professor, Department of Electronics & Telecommunication Engineering, SKN Sinhgad College of Engineering, Korti, Pandharpur, Maharashtra, India.
Curcumin, a phenolic pigment isolated from Curcuma longa, holds well-documented therapeutic relevance across oncology, neurology, and inflammatory disease. Yet its journey from laboratory promise to clinical reality has been repeatedly derailed by an extremely low oral bioavailability that barely exceeds one percent, a consequence of negligible aqueous solubility, fast metabolic conjugation, and limited intestinal permeability. Encapsulating curcumin within chitosan-tripolyphosphate (CS-TPP) nanoparticles prepared through ionic gelation has long been recognised as a viable strategy to surmount these barriers, but identifying the precise combination of chitosan concentration, drug payload, and crosslinker ratio that simultaneously satisfies targets for particle size, colloidal stability, and encapsulation remains a formidable multivariate challenge when addressed through conventional trial-and-error experimentation.
The present work therefore embedded artificial intelligence at the centre of the formulation development workflow. A multilayer perceptron (MLP) network trained on 120 systematically generated batches learned the complex, nonlinear landscape connecting three process inputs to four physicochemical outputs with a test-set R² of 0.9943 for particle size — markedly better than a Box-Behnken polynomial model applied to the same dataset. Bayesian optimisation with a Gaussian Process surrogate and Expected Improvement acquisition subsequently located the global optimum in 31 guided iterations rather than the eighty-plus experiments a full factorial screen would have demanded.
The AI-recommended formulation — 0.30% w/v chitosan, 5.5 mg curcumin, 1:5 TPP ratio — was synthesised and characterised independently. Particle size settled at 182 ± 9 nm, PDI at 0.17 ± 0.02, zeta potential at +33.6 ± 1.7 mV, and encapsulation efficiency at 86.9 ± 2.1%, each within 6% of the model prediction. Release in phosphate-buffered saline followed anomalous (non-Fickian) kinetics with 79.8% cumulative delivery at 24 h. Apparent permeability across Caco-2 monolayers was 3.9-fold higher than that of unformulated curcumin, and cytotoxic potency against HCT-116 colon carcinoma cells improved 5.7-fold. Beyond the formulation outcomes, the study documents a reproducible AI-pharmaceutical template that shortened the optimization timeline by roughly two-thirds relative to conventional design strategies.
Keywords: Artificial intelligence, chitosan nanoparticles, curcumin oral delivery, Bayesian optimization, multilayer perceptron, bioavailability enhancement.
How to cite this article: Chaudhari M, Dhake TP, Gajare M, Chopade DP, Gawande P, Rana M, Mulani AO. Harnessing artificial intelligence for the rational design and optimization of chitosan-tripolyphosphate nanoparticles loaded with curcumin: a deep learning and Bayesian optimization approach to oral bioavailability enhancement. Int J Drug Deliv Technol. 2026;16(3s): 998-1006; DOI: 10.25258/ijddt.16.3s.120
Source of support: None.
Conflict of interest: None