In the high-stakes world of pharmaceutical development, formulation scientists are often caught in a "try, test, and make again" cycle. The quest to determine how a drug releases its active ingredient—the fundamental measure of efficacy—has long been a slow, iterative, and resource-heavy process. For developers of complex therapeutics like long-acting injectables and specialized implants, a single dissolution study can stretch between 30 and 90 days. If the results fall short, the entire process resets, leading to costly delays that can push back critical regulatory filings by months or even years.
However, a shift is underway. By moving beyond traditional bulk measurement and tapping into the "hidden" architecture of the drug itself, companies like DigiM Solution are ushering in an era of predictive drug development. By transforming complex microstructures into actionable data, researchers can now forecast drug-release profiles before a single batch is manufactured.
The Invisible Variable: Why Bulk Testing Fails
To the naked eye, or even under standard scanning electron microscopy (SEM), two drug formulations may appear identical. Yet, in practice, their performance can diverge drastically. This paradox lies in the "hidden variable"—the internal spatial distribution of the active pharmaceutical ingredient (API), excipients, and pore spaces.
Shawn Zhang, Ph.D., co-founder and CEO of DigiM Solution, explains that conventional methods like laser diffraction often mask these nuances. "If you use conventional laser diffraction, the particle sizes appear identical," Zhang says. "If you look at them by external scanning electron microscopy, they look identical. Yet the release profiles differed."
The culprit is often the manufacturing process itself. In a collaborative study with the University of Connecticut and the FDA, DigiM demonstrated that simple variations in stir rates or the viscosity of silicone oil during the production of PLGA microspheres could create vastly different internal structures. While one batch exhibited a uniform distribution of API and porosity, another was "patchy," containing voids that fundamentally altered how the drug would eventually be released into the body.
Chronology of a Paradigm Shift
The evolution of microstructural intelligence did not happen overnight. Since its founding in 2014, DigiM has spent over a decade refining a methodology that prioritizes data integrity and scientific reproducibility.
- 2014–2020: The formative years. DigiM focused on establishing a robust digital infrastructure, creating standardized protocols for sample preparation and quality control. This period was defined by the realization that raw images are merely "pretty pictures" unless they are converted into quantitative, analytical data.
- 2021: The pivot to Generative AI. Recognizing the need to solve complex development challenges without constant, expensive wet-lab iterations, the team began developing generative models capable of simulating virtual microstructures.
- 2024: Validation through peer-reviewed research. A landmark study published in Nature Communications, co-authored by researchers from Genentech and Merck, provided a high-profile validation of DigiM’s approach. By using generative AI to predict the "percolation threshold"—the point at which API particles form a continuous path for release—the team proved that they could accurately model complex behaviors without physical synthesis.
- 2025–2026: Regulatory integration. The FDA began incorporating microstructural characterization into draft guidance for generic drug development, specifically for complex products like minocycline dental inserts and dexamethasone ophthalmic implants.
The Technical Backbone: From Grayscale to Insight
The core of DigiM’s technology is a sophisticated pipeline that converts grayscale microscopy images into a digital map of the dosage form. Because standard microscopy does not inherently identify material phases, the company utilizes semantic segmentation.
In this process, a domain expert manually labels a small, representative sample of a 2D image. These labeled sets are then used to train convolutional neural networks (CNNs), which can subsequently assign numerical material labels to every pixel in a 2D image or every voxel in a 3D scan.
"Converting those images into what you see as color is a digital transformation," says Zhang. "We are transforming grayscale pictures into actual material phases, particle sizes, volume fractions, mass fractions, and concentration gradients. Those are the critical quality attributes that development and regulatory decisions can be based on."
This workflow is supported by a massive network of over 50 labs and instruments, including advanced synchrotron facilities. By outsourcing the imaging to high-precision hardware while maintaining strict control over the digital processing, DigiM ensures that the final data is not just descriptive, but predictive.
Supporting Data: Efficiency and Precision
The economic and operational implications of this digital-first approach are profound. Conventional dissolution studies are notoriously wasteful, often consuming hundreds of grams of expensive, sometimes "gold-priced" API early in the development cycle.
In contrast, DigiM’s imaging workflow operates on just a few milligrams of material—a single tablet, a few hundred microspheres, or a single granule. By using the ivisLab platform, formulators can model how the product evolves in a release environment, accounting for:
- Water uptake: Tracking how the polymer interacts with the environment.
- API evolution: Monitoring how the drug network changes as the matrix dissolves.
- Mass-transport properties: Calculating effective diffusivity based on the physical structure of the dose.
The result is a directional release prediction that functions as a "digital twin" of the physical formulation. As Zhang notes, "The major difference is that this platform does not require any prior testing data. You get a prediction before you manufacture the batch."
Generative AI: Working Backwards
Perhaps the most ambitious aspect of this technology is its ability to work in reverse. If a company has a target release profile, DigiM’s generative AI can propose the internal architecture required to achieve it.
Using the analogy of chocolate chip cookies, Zhang explains: "If you want to ensure every bite of a cookie has the same amount of chocolate, you need to know how many chips to add so they connect. For a drug product, the ‘chips’ are the API. If the API is connected into a network, it can find its way out through the dissolved phase—a process called percolation. If the particles are isolated, the release is significantly slower."
By digitally "adding" or "shifting" these particles within a virtual model, scientists can identify the exact threshold for release without running a single physical manufacturing campaign. This capability was famously demonstrated in the Nature Communications study, where the team accurately predicted the behavior of microcrystalline cellulose networks within tablets.
Regulatory Implications and Future Outlook
The FDA’s recent interest in microstructure characterization marks a significant turning point. In draft guidances for generic dental and ophthalmic products, the agency has explicitly listed porosity, particle size, and spatial distribution as key requirements for bioequivalence. By citing research papers that utilized DigiM’s methodology, the FDA is signaling that the industry is ready for a more nuanced, data-driven approach to quality.
While these guidances are currently nonbinding, they represent a fundamental shift in the regulatory landscape. The industry is moving toward a future where "the totality of scientific evidence" includes high-resolution digital imaging and predictive modeling as core pillars of approval.
As Shawn Zhang prepares to address these topics at the upcoming PharmSci 360 conference in New Orleans, the message is clear: the era of blind iteration is ending. By "seeing inside the dose," pharmaceutical companies can reduce the uncertainty of development, accelerate the path to market, and ensure that the next generation of complex medicines is as safe and effective as possible.
For developers and regulators alike, the "hidden" microstructures that once obscured the path to success are now becoming the very tools that define it. The future of drug development is not just about making more, but about knowing more—down to the last pixel.
About the Author
