(654e) Machine Learning (ML) Model for Nozzle Droplet Prediction in Spray Drying

Approximately 40% of the New Chemical Entities (NCEs) in Pharma are poorly soluble which leads to inadequate bioavailablity^1,2. There are multiple options to resolve this issue including Hot Melt extrusion, Spray Drying, Cocrystal, Milling, etc. Spray drying is one of the very promising technologies for particle engineering of Pharma API to overcome this solubility issue. Spray drying can be used to generate very fine powders for inhalable drugs and to improve the bioavailability of solid dosage drugs using amorphous solid dispersion. However, spray drying is quite a challenging process to optimize and scale up. The droplet size generated from the nozzle has a significant impact on the final product yield in the spray dryer.

The droplet size from a Spray Drying nozzle depends on the liquid flow rate, nozzle tip size, air flow rate, & solvent. It is not feasible to determine the best operating conditions for any new API with experiments on Spray Dryer for each combination. Many correlations are available to predict the droplet size based on the operating conditions for Spray Drying nozzles³. However, each of these correlations results in vastly different results on droplet size distribution and is only suitable for the specific case. Therefore, this work measured the droplet size for different operating conditions and the nozzle specification using laser diffraction. Droplet size was measured for acetone, water, & Amino Methacrylate Copolymer in acetone. A total of 11,390 runs were conducted to evaluate the impact of parameters on the droplet size. In this work Machine Learning (ML) models have been generated to predict the droplet size based on operating conditions and the nozzle design. Simple regression tree, gradient boosted regression tree, Random Forest, & Tree Ensemble were evaluated in this work. The ML model provides a significant competitive advantage to Veranova by significantly reducing the number of droplet size trials required for a new customer product. The best ML models predicted the validation data with 99.4% accuracy for acetone and 92.9% for water solvent respectively.

References:

1. Price, G. & Patel, D. A. Drug Bioavailability. in StatPearls (StatPearls Publishing, Treasure Island (FL), 2022).
2. Savjani, K. T., Gajjar, A. K. & Savjani, J. K. Drug Solubility: Importance and Enhancement Techniques. ISRN Pharm. 2012, 195727 (2012).
3. Hede, P. D., Bach, P. & Jensen, A. D. Two-fluid spray atomisation and pneumatic nozzles for fluid bed coating/agglomeration purposes: A review. Chem. Eng. Sci. 63, 3821–3842 (2008).