Source-linked AI summary
Accelerating the Development of PLGA In Situ Forming Depots Through AI-Driven Multi-Objective Optimization
Pauric Bannigan, Siddarth Chandrasekaran, Brigitte A. G. Lamers, Inge Hermsen, Gary Tom, Riley J. Hickman, Bahar Yeniad, Morgan Fox, Christine Allen
TL;DR
Balancing release, loading, viscosity, and injectability across a large formulation space is challenging. The study combined PURASORB polymers with automated, AI-guided experimentation to optimize peptide depots. Intermediate-molecular-weight polymers offered a favorable release-viscosity balance, and simulated campaigns found ANDROMEDA 1 identified leads more often than random search or DoE.
Problem
In situ forming depots involve a large design space with interacting objectives, including sustained release, burst control, viscosity, and injectability.
Source: PURASORB Polymer Toolbox · p. 2, VALIANT™ for Multi-Objective… · p. 3Method
The study combined Corbion’s commercially available and developmental PURASORB polymers with automated formulation experiments and ANDROMEDA 1 model-guided optimization.
Source: INTRODUCTION · p. 2, VALIANT™ for Multi-Objective… · p. 3Results
Intermediate-molecular-weight polymers balanced sustained release and viscosity; ANDROMEDA 1 identified a lead in 100% of simulated campaigns, versus 63% for random search and 57% for DoE.
Source: RESULTS · pp. 5, 7Takeaways & Limitations
The case study illustrates that combining polymer science with AI-guided experimentation can accelerate LAI development while generating formulation understanding for robust, scalable products.
Source: INTRODUCTION · p. 2The comparison of ANDROMEDA 1 with random search and DoE was retrospective computational analysis, not a prospective head-to-head experimental study.
Source: RESULTS · p. 7
Abstract
from arXiv · showhide
Developing long-acting injectable formulations requires the simultaneous optimization of drug loading, release kinetics, viscosity, injectability, stability and other objectives. To navigate this multidimensional space, Corbion and Intrepid combined Corbion's diverse PURASORB bioresorbable polymer library with Intrepid Labs' proprietary AI algorithm (ANDROMEDA 1) to develop in situ forming depots for a therapeutic peptide. Over approximately 15 weeks, 181 unique formulations spanning drug loadings of 6-12% w/w were prepared and characterized through broad design-space mapping and targeted multi-objective optimization. Four lead candidate formulations were identified at 6%, 9%, and 12% w/w drug loading. Each met the predefined viscosity and injectability criteria while providing distinct 30-day in vitro release profiles. The study evaluated polymers spanning a broad range of molecular weights, including commercially available PURASORB grades and new polymers under development by Corbion to expand its polymer toolbox. ANDROMEDA 1 identified that polymers with intermediate molecular weights provided a favorable balance between sustained release and solution viscosity. Together, these findings demonstrate how integrated polymer expertise and AI-driven optimization can rapidly identify differentiated formulation candidates, focus the development space, and establish a strong data-driven foundation for further optimization and in vivo evaluation.
INTRODUCTION
In situ forming depots may avoid multistep microsphere manufacture, but their formulations must balance release and drug loading against practical administration constraints. The study combines an expanded PURASORB polymer library with automated, AI-guided multi-objective development.
INTRODUCTION
- In situ forming depots form a drug-containing semi-solid after injection and may avoid the multistep manufacture of clinically validated PLGA microparticles or microspheres.Existing products also illustrate practical constraints: some require repeated mixing before administration, while ready-to-use products still have temperature and viscosity handling requirements.Source: INTRODUCTION · p. 2
- Depot formulation must balance release duration and burst, drug loading, viscosity, injectability, stability, manufacturability, and cost across a large, interacting design space.Improving release control can raise viscosity or compromise injectability, so a feasible region must satisfy these objectives together rather than optimize one endpoint.Source: INTRODUCTION · p. 2, VALIANT™ for Multi-Objective… · pp. 3–4
- The study combines commercially available and developmental PURASORB polymers with VALIANT™ experiments and ANDROMEDA 1 to optimize a one-month peptide depot.Formulations were assessed against a predefined target product profile rather than release behavior alone.Source: INTRODUCTION · p. 2, VALIANT™ for Multi-Objective… · p. 3
- The polymer library spans 50:50 and 75:25 PLGA copolymers and PDL homopolymers, enabling evaluation of composition and material properties against release and viscosity.Higher molecular weight can support sustained release but typically raises viscosity; copolymer ratio influences hydration and degradation, while end-group chemistry did not meaningfully differentiate materials in this campaign.Source: PURASORB Polymer Toolbox · p. 3
- The target product profile includes release duration and trajectory, 24-hour burst, drug loading, viscosity, injectability, and physical feasibility.Physical feasibility includes homogeneous formulations, reproducible depot formation, and no phase separation; injectability is assessed through preparation and passage through a 25 G needle.Source: VALIANT™ for Multi-Objective… · p. 3
- ANDROMEDA 1 learns composition-performance relationships, estimates performance and uncertainty across the remaining design space, and prioritizes subsequent experiments.New experimental data are returned to the model to refine the landscape and concentrate testing in regions more likely to satisfy the target product profile.Source: INTRODUCTION · p. 2, VALIANT™ for Multi-Objective… · p. 3
RESULTS
The study identified lead formulations balancing release control with practical viscosity, and assessed how polymer attributes and formulation composition relate to performance. Retrospective simulations found ANDROMEDA 1 more efficient and consistent at identifying leads than random search or DoE.
RESULTS
- Increasing polymer molecular weight and loading generally increased viscosity while influencing release; chain entanglement, phase inversion, and water ingress contribute to these coupled effects.The resulting trade-off is between early drug retention, sustained release, and practical administration.Source: RESULTS · p. 4
- Selected leads occupied the preferred design region, balancing controlled early release with practical viscosity across drug loadings of 6–12% w/w.The preferred region was defined as viscosity ≤3,000 cP and approximately 6–7% 24-hour burst release.Source: RESULTS · p. 4, Figure 2 · p. 4
- Innovative polymers populated the preferred region and comprised three of four leads; intermediate-molecular-weight polymers best balanced sustained release and practical viscosity.The reported viscosity was suitable for administration through a 25 G needle.Source: RESULTS · p. 5
- Polymer loading dominated burst release, while solvent fraction dominated viscosity; higher residual monomer and PLGA molecular weight were associated with increased burst.The expanded analysis found polymer loading and solvent fraction important across multiple objectives, with some nonlinear relationships.Source: RESULTS · pp. 5–6, Figure 3 · p. 5
- ANDROMEDA 1 found a lead in 100% of simulated campaigns after approximately 35 experiments, versus 63% after approximately 121 for random search and 57% after approximately 120 for DoE.Campaigns were scored on sampling any of the four experimental leads.Source: Figure 4 · p. 6, RESULTS · p. 7
- Approximately 92 g mean material and 7.9 g peptide were used by ANDROMEDA 1 to identify a lead, versus 141 g and >13.0 g for random search and 134 g and 11.9 g for DoE.This was a retrospective computational comparison, not a prospective head-to-head experimental study.Source: RESULTS · p. 7
Supplementary information
Supplementary analyses describe modelled attribute–objective relationships and compare how often search strategies discover lead formulations in simulated campaigns.
Supplementary information
- Figure S1 plots partial-dependence responses for the three most influential attributes across zero-order release deviation and the multi-objective score.Attributes vary from their 5th to 95th percentile; lower objective values are preferred, so descending curves indicate improvement.Source: Figure S1 · p. 9
- ANDROMEDA 1 sampled F148 in 28 of 30 campaigns, versus 6 for random search and 3 for DoE.It sampled at least one lead formulation in all 30 campaigns.Source: Table S1 · p. 10
- Table S1 reports the number of campaigns, out of 30 and within a 160-experiment budget, that sampled each lead formulation.“At least one lead” means one or more of F136, F139, F148, or F162 was sampled.Source: Table S1 · p. 10