In drug development, new molecular
entities get all the press. That’s largely justified – new molecules mean new medicines and the process of proving one safe and effective for treating a disease is notoriously laborious and expensive. But spare a thought for the life of a molecule after it’s approved as a drug. Additional work is sometimes put in to formulate that new medicine to give it new properties – for example, to make it more stable in transport and storage, or to give it time-release properties so patients have an easier time taking their medicine as prescribed. This additional work would arguably happen even more often if it didn’t require trained scientists to spend months-to-years working through the trial-and-error process of finding exactly the right formulation for the job.
Karthik Raman and Chris Shelner, co-founders of Persist, bring a unique mix of expertise in materials science, AI, and lab automation to the problem. Their approach holds the promise of designing, building, and validating extended-release formulations applicable to a wide range of medicines in a fraction of the time currently required. With their potential to widen access to the benefits of advanced formulation and even open the possibility of extended release generics, Persist well embodies Innospark’s belief in AI for good.
The Problem of Formulation Design
The idea behind the extended-release formulation of a drug is straightforward. The therapeutic molecule stays the same, but instead of administering the therapeutic in solution, it is encapsulated within a bead of material that slowly breaks down to release it – like sugar cube melting into tea. The material typically used is called PLGA, and has been used in long-acting depot formulations since 19891. PLGA stands for poly(lactic-co-glycolic-acid). Up close, it’s a connected chain of lactic acid and glycolic acid molecules – the same lactic acid your muscles make when they’re tired, and the same glycolic acid used in skin care products. PLGA is a convenient material for time-releasing drugs because its naturally broken down and absorbed by the body.
For a material with simple components, the properties of PLGA can vary dramatically based on how the material is prepared. By varying parameters like the density, the degree to which PLGA develops crystal structure, or the number and size of pores in the material, a chemist can dramatically change the rate at which drug molecules encased in PLGA escape. Complicating matters, the rate at which a drug molecule encased in PLGA escapes also depends on properties of the drug itself. Parameters like a molecule’s weight, charge, and hydrophilicity all come into play. In total, about 300 parameters of the drug-PLGA formulation’s construction can be altered. Problematically for formula design, the parameters are highly interrelated – adding to the number of pores decreases density, and so on. Understanding this large number of formulation parameters, which influence each other in sometimes unexpected ways, you can begin to understand why producing a precise time-release profile is a challenging and iterative process even for trained chemists.
So, if time-release formulations are time-consuming, labor-intensive investments, who’s willing to invest in making them? Very few people. Historically these sorts of investments are made almost exclusively by large pharmaceutical companies to create new renditions of well-performing, often lucrative, drugs. The pharma company is sometimes able to create new intellectual property around the time-release formulation and FDA provides a special pathway, the 505(b)(2), that allows new drug formulations to be
approved more quickly. However, in many circumstances the investment in developing an extended-release formulation is too large and risky to justify.
The Persist Solution
We’re excited about Persist because they combine multiple innovative technologies to make time-release formulations faster and more accessible. The company innovates at each phase of the design-build-test cycle.
First, Persist deploys well-chosen ML methods to predict the extended-release formulation that will take a given drug and release it at a desired rate over a desired length of time. These algorithms leverage both the limited amount of publicly available formulation data as well as a collection of data developed internally.
Second, Persist rapidly builds and tests the formulation predicted by its AI, and collects data on its performance. The speed and scale of this process is accelerated by custom-designed lab automation implementations.
With each iteration, Persist improves two things - the accuracy of the formulation for the specific drug being designed for and the power of their internal dataset to predict good initial formulations of similar drugs in the future.
Into the Future of AI-enabled Drug Formulation
With their unique combination of AI and automation, we hope and expect Persist will be capable of producing well-performing formulations for a wide range of medicines, at a speed and cost that will make them much more available. In the short term, they will enable the best efficacy of some types of therapies, like those that benefit from maintaining a specific blood concentration. In the longer-term, we believe Persist can help better deliver therapies across several modalities, and even serve customers outside the pharmaceutical industry. Karthik, Chris, and their team have the right mix of ML, materials, and robotics expertise to bring to life an exciting new platform for drug formulation design. Their solution puts AI to use for good, and we’re excited to support them.
1. Park et al. “Injectable, long-acting PLGA formulations: analyzing PLGA and understanding microparticle formation.” Journal of Controlled Release. 2019