A Novel Semi-Automated Pipeline for Optimizing 3D-Printed Drug Formulations

Abstract

3D printing offers a promising approach to creating personalized medicines. However, costly, expertise-dependent trial-and-error methods hinder efficient drug formulation, posing challenges for tailoring treatments to individual patients. To address this, a novel pipeline is developed for 3D printing using selective laser sintering (SLS), replacing laborious steps with advanced computational methods. A differential evolution-based optimizer generates formulations for the desired drugs, while a deep learning ensemble predicts the optimal printing parameters along with associated confidence intervals. Manual handling is only required for the final formulation preparation and printing processes. The pipeline successfully generates diverse formulations, composed of a wide variety of materials and with high printability probabilities. This was validated by successfully printing 80% of the generated drug formulations and achieving 92% accuracy in predicting printing parameters. Notably, the time required to develop and print a new drug formulation is decreased to a single day. This study is the first to demonstrate a semiautomated, 3D printing drug formulation design and printing parameter selection pipeline. Furthermore, the pipeline is not limited to SLS printing but can also be adapted for the optimization of other 3D printing technologies or formulation platforms.

Introduction

3D printing, also known as additive manufacturing, encompasses a range of technologies used to fabricate 3D objects based on digital designs.[1] Originally developed for engineering, this technology is now being explored for manufacturing medicinal products to address patient heterogeneity and the health disparities stemming from the current one-size-fits-all treatment approach.[2] 3D printing enables the production of small batches of medicines tailored to individual patients, facilitating the production of personalized medications, something conventional technologieswhich were designed for large-scale manufacturing,struggle to achieve. Subsequently, recent clinical trials have been conducted to investigate personalized 3D-printed dosage forms.[3]

Selective laser sintering (SLS) is a powder bed fusion 3D printing technology that primarily utilizes carbon dioxide lasers to fuse powder particles.[4] SLS has proven highly effective for 3D printing medicines due to its simplicity, versatility in producing various drug delivery systems, and its suitability for large-scale production.[5] Its success is further attributed to its ability to create complex 3D objects without the need for support structures and using powder feedstock materials without solvents.[5] As a result, SLS has been successfully applied in the development of various drug delivery systems and has recently been trialed in humans for the first time.[6] Despite its advantages and ability to outperform conventional drug manufacturing methods,[7] translating 3D printing technologies into widespread pharmaceutical use has been slow. This is partly due to the challenges in formulating medicines compatible with 3D printing technologies, which were not initially designed for this purpose. Consequently, the current development of 3D-printable medicines relies on a trial-and-error approach, dependent on user expertise. This method is iterative, time-consuming, expensive, and wasteful.[8]

Machine learning (ML), which leverages data for learning rather than relying on explicit programming, has gained significant interest in pharmaceutical manufacturing, enabling the development and optimization of complex drug delivery systems.[9] In the context of 3D printing, ML has shown remarkable success, facilitating the prediction of whether drug formulations can be 3D printed,[10] optimizing printing parameters,[11] and determining the properties of 3D-printed medicines.10, 12 For SLS printing, previous studies have demonstrated the ability to predict whether formulations can be successfully SLS printed.[13] Notably, we have developed a deep learning (DL) ensemble model capable of predicting printability with over 90% accuracy.13 However, no studies have explored the prediction of optimal printing parameters for SLS printing of medicines. Two key parameters- the printing temperature and laser scanning speed – must be optimized during SLS printing. As a result, even if a formulation is printable, parameter optimization remains iterative and resource-intensive. Moreover, designing formulations to predict their printability relies heavily on scientific expertise. If a formulation is unprintable, scientists must use their heuristic knowledge to develop a new formulation, requiring significant experience to do so accurately and often leading to inefficient experimental loops of refinement and testing. Currently, no robust methods exist for generating drug formulations for 3D printing. Due to the complexity of drug formulations and the limitations of available data, only one study to date has attempted to generate 3D-printed drug formulations, though with limited exploration and success.[14] These challenges must be addressed before the full potential of SLS printing in pharmaceutical applications can be realized.

To address these challenges, building on our previous work,13 we developed a novel DL and differential evolution (DE)-based pipeline to automate the SLS drug formulation design and printing parameter selection process. To overcome formulation design and optimization issues, we propose a system that allows researchers to input a nonprintable formulation or an unformulated drug into the algorithm, which then generates an optimal formulation for SLS printing. To address the challenge of determining printing parameters, we developed DL models capable of predicting these parameters with associated confidence intervals. By eliminating human intervention in the initial trial-and-error loop and automating the iterative process of designing formulations, while only requiring human input in the final formulation preparation and printing stage, our pipeline demonstrates the ability to both generate new formulations and optimize nonprintable formulations to make them printable. We validated this approach by successfully printing 80% of the generated formulations and achieving 92% accuracy in predicting printing parameters. This demonstrates the first optimized and automated drug formulation process. This technology has broad applicability, extending to other 3D printing technologies and other drug formulation challenges, and presents the first step towards more efficient drug design and development.

Download the full article as PDF here A Novel Semi-Automated Pipeline for Optimizing 3D-Printed Drug Formulations

or read it here

Materials

The materials used as part of this study to make the drug formulations were AQOAT AS-HG (Shin-Etsu, Tokyo, Japan); Aqualon EC-N50, Aqualon EC-N7, Klucel hydroxypropyl cellulose EF, and Plasdone S-630 (Ashland, Schaffhausen, Switzerland); Candurin gold sheen and polyvinyl alcohol (PVA) 87%–90% hydrolyzed (Merck Life Science Limited, Dorset, UK); chitosan medium molecular weight, magnesium stearate, mannitol, paracetamol, polyethylene oxide (PEO) 1M, polyvinylpyrrolidone (PVP) 40000, stearamide (N,N-ethylenbis), and xylitol (Sigma Aldrich, Gillingham, UK); ethyl cellulose CP 10 (Fisher Scientific Ltd., Loughborough, UK); Eudragit L100-55 and Eudragit RL PO (Evonik, Darmstadt, Germany); Kollicoat Protect, Kollidon CL-M, Kollidon SR, and Soluplus (BASF, Ludwigshafen, Germany); Opadry AMB II: high-performance moisture barrier film coating yellow (Colorcon, Dartford, UK); and PEO) 7M (The Dow Chemical Company, Midland, USA).

Youssef Abdalla, Martin Ferianc, Haya Alfassam, Atheer Awad, Ruochen Qiao, Miguel Rodrigues, Mine Orlu, Abdul W. Basit, David Shorthouse, A Novel Semi-Automated Pipeline for Optimizing 3D-Printed Drug Formulations, Advanced Intelligent Systems published by Wiley-VCH GmbH, First published: 15 May 2025, https://doi.org/10.1002/aisy.202401112