In the high-stakes world of pharmaceutical formulation, time is the ultimate currency. Formulation scientists are perpetually caught in a high-pressure cycle of "try, test, and make again," where a single dissolution study for long-acting injectables or complex implants can span 90 days or more. Any failure—any deviation in the expected release profile—forces the team to return to the drawing board, effectively restarting the clock and risking millions in development costs and postponed regulatory filings.
Shawn Zhang, Ph.D., co-founder and CEO of DigiM Solution, believes the industry has been looking at the problem through a narrow lens. As he prepares to deliver his keynote, "Seeing Inside the Dose: Predictive Drug Development with Microstructure Intelligence," at the upcoming PharmSci 360 conference in New Orleans this October, Zhang is advocating for a shift from trial-and-error physical testing to a predictive, data-driven paradigm powered by advanced imaging and machine learning.
The Invisible Variable: Why External Looks Can Be Deceiving
The fundamental challenge in modern drug development lies in the "hidden variable" of internal architecture. Conventional assay and bulk measurements—such as laser diffraction or standard light microscopy—often provide an incomplete picture. They measure external dimensions, but they fail to map the internal spatial distribution of the active pharmaceutical ingredient (API), excipients, and pore spaces.
"From the outside, two products can look the same," Zhang explains. "But once you open them up, you see the primary particles are deposited differently because of different process conditions."
This discrepancy was vividly illustrated in a joint study conducted by DigiM, the University of Connecticut, and the FDA. Researchers examined four different PLGA microsphere formulations that differed only in their stir rates and the viscosity of the silicone oil used during manufacturing. While the formulations were externally indistinguishable, their release profiles varied drastically.
One batch displayed a uniform distribution of API and porosity throughout the sphere. The other was "patchy," characterized by regions entirely void of the active ingredient. Standard laser diffraction would have labeled these microspheres identical; however, the internal microstructural variance dictated a completely different clinical performance.
A Chronology of Innovation: The Rise of Microstructure Analytics
DigiM’s journey began in 2014, long before the current AI boom brought generative models into the mainstream. The company’s trajectory has been defined by a pragmatic, steady integration of imaging and computational power:
- 2014: DigiM establishes its core methodology, focusing on high-resolution imaging and semantic segmentation to quantify internal drug structures.
- 2021: The team initiates the development of generative AI models, aiming to predict the impact of formulation changes without physical synthesis.
- 2024: A landmark Nature Communications paper, co-authored with industry giants Genentech and Merck, validates the use of generative AI in creating virtual tablet microstructures to identify critical percolation thresholds.
- 2025–2026: Regulatory bodies, including the FDA, begin incorporating microstructural imaging requirements into draft guidance for complex generics, such as minocycline dental powder and dexamethasone ophthalmic inserts.
The Technological Infrastructure: Turning "Pretty Pictures" into Data
To make these predictions, DigiM has constructed a sprawling network of over 50 labs and instrument facilities, including high-end synchrotron facilities. This infrastructure allows the company to focus on its true value proposition: the digital processing of complex imaging data.
The workflow begins with high-resolution computed tomography (CT) and Focused Ion Beam Scanning Electron Microscopy (FIB-SEM). While a standard medical CT scan looks at anatomy, these instruments resolve micro-scale particles and granules at three orders of magnitude higher resolution.
However, raw images are essentially "pretty pictures." To turn them into actionable data, DigiM utilizes its proprietary digiM I2S software. Through supervised machine learning, domain experts label specific regions of grayscale images, teaching the neural network to differentiate between API, polymer, and pore space. This "semantic segmentation" converts visual data into numerical material maps—volume fractions, mass fractions, and concentration gradients—which then become the basis for regulatory and development decision-making.
Supporting Data: Simulation vs. Traditional Dissolution
Perhaps the most significant differentiator for DigiM’s ivisLab platform is its ability to bypass the "fit-to-curve" trap. Traditional pharmacokinetic platforms often rely on historical dissolution data to calibrate models. DigiM’s approach is fundamentally different: it uses the measured, physical microstructure as an initial condition to simulate the mass-transport properties of the dose.
"The major difference is that our simulation does not require any prior testing data," Zhang notes. By modeling how the product evolves—tracking water uptake, polymer degradation, and the shifting API network simultaneously—the software can predict a dissolution curve based on the physical chemistry of the dose itself.
This is particularly transformative for early-stage development, where the API is often "literally more expensive than gold." Where a conventional dissolution study might consume hundreds of grams of precious material, the imaging workflow requires only a few milligrams.
Generative AI: Designing from the Inside Out
The most forward-looking aspect of DigiM’s work is its use of generative AI to work backward from a desired release target. By using existing microstructures as a baseline, the company can generate virtual variants, changing one attribute while keeping the rest of the morphology constant.
The "chocolate-chip cookie" analogy serves as the foundation for this work. In a tablet, the API particles act like chocolate chips. If the particles are isolated, the API release is slow and diffusion-dependent. If the particles form a connected, percolated network, the drug can "find its way out" through the continuous channel as it dissolves.
Using generative AI, DigiM can predict the precise threshold at which these particles connect—a "percolation threshold"—without manufacturing a single physical tablet. This allows formulators to optimize drug loading and porosity mathematically, drastically reducing the number of manufacturing campaigns required.
Regulatory Implications: A New Era for Bioequivalence
The scientific community is no longer viewing microstructure characterization as an academic exercise; it is becoming a regulatory necessity. The FDA’s recent draft guidances for generic dental and ophthalmic products explicitly list microstructure imaging as a key requirement for establishing bioequivalence.
Specifically, the agency is looking for comparative characterization of:
- Porosity: The void space within the dosage form.
- API Particle Size: Distribution within the internal matrix.
- Spatial Distribution: Where the drug resides relative to the excipient structure.
By citing FIB-SEM papers co-authored by DigiM, the FDA is signaling that the era of "black box" formulations is coming to a close. The regulator is increasingly demanding evidence based on the actual, physical architecture of the product.
Implications for the Future of Drug Delivery
As the pharmaceutical industry faces increasing pressure to reduce costs, shorten development timelines, and improve the predictability of complex drug-delivery systems, the transition toward "digital twins" of dosage forms appears inevitable.
DigiM’s approach represents a broader movement toward high-fidelity digital transformation in life sciences. By replacing slow, iterative physical testing with rapid, precise, and regulatory-compliant digital modeling, developers can move away from the "try, test, and make again" cycle.
As Zhang prepares for his presentation at PharmSci 360, the message is clear: the future of drug development will not be written in a laboratory notebook alone, but in the voxels of a 3D scan. By mastering the internal architecture of the dose, the industry is not just optimizing release profiles; it is fundamentally redefining the speed and efficiency of bringing life-saving medications to market.
About the Author
