#39: Bending the Drug-Development Curve
Making it faster, better, and cheaper to get life-saving medicines to patients.
Tech-Bio
“Tech-bio” has become a popular and buzzy phrase. I find myself using it often, so (mainly for my own benefit) I thought it’d be productive to articulate how I’ve been thinking about the opportunity set.
The argument for investing in tech-bio seems simple enough: It can take more than a decade and billions of dollars to get a drug from discovery to market. If a drug does reach blockbuster status, the investment is still worth it—ten drugs generated more than $10B each last year. Still, too many therapeutic pipelines are cut short because of this “all-or-nothing” approach, leaving patients without treatments that could improve their lives. Pharma and biotechs should therefore have a high willingness to pay for technologies that take substantial time and cost out of the drug-development process. Practically, we are seeing this play out.
Lab equipment and general computing have come down cost curves, giving rise to technologies addressing major bottlenecks in the drug-discovery and development process. These technologies are making it faster, better, and cheaper to develop new therapeutics—helping to get potentially life-saving treatments to patients sooner.
Faster, Better, Cheaper
Here’s a basic step-by-step of the drug development process, as well as emerging technologies addressing major bottlenecks.
Scientists begin by hypothesizing which protein/gene/pathway to target for a desired therapeutic effect. They then screen molecules against potential targets, which is where the first bottleneck lies. Scientists are limited by the speed at which they can screen molecules, the supply of reagents, and their ability to capture and analyze data, making technologies that speed up the pace of discovering new targets and drugs valuable.
High-throughput screening automation using robotics speeds up the process and generates more data faster.
Miniaturization via microfluidics requires less starting reagent to test the effects of a drug.
AI-enabled in silico modeling of drug-target interactions provides a scale of screening that isn’t possible in the physical world.
Once a target is identified, a drug needs to be tested for initial toxicity. This is usually screened alongside the therapeutic effect. Traditional assays that are cell-based, such as 2D cell cultures, are used to model toxicity but lack biological realism and therefore have limited predictive power. This is a major pain point and drives much of the downstream cost of clinical trials, since toxicity is among the biggest reasons trials fail.
Human organoids and organ-on-a-chip models better mimic biology (though still not at a point where these can replace animal models).
AI/ML platforms are predicting off-target binding to improve translation to human trials.
The FDA requires safety data from two mammalian species (one rodent and one non-rodent)—though they’ve publicly signaled a desire to move away from animal testing and may begin accepting non-animal data. Animal testing is time-consuming, costly, and not as effective as it should be; data capture and general maintenance of animals remain largely manual, leading to poor reproducibility and translation to humans.
Non-animal models (lab-grown organs, organ-on-a-chip, digital twins) are improving the translation of in vitro testing to humans.
Automated and continuous data capture for animal facilities standardizes data collection and analysis, while improving animal welfare.
Drugs are packaged into a formulation that determines the stability, solubility, and half-life of active ingredients. Formulation development begins early because of implications on dosage, efficacy, safety, and patient convenience. Today, scientists hypothesize and test formulations manually, where each test can last months.
The convergence of robotics, AI, and microfluidics has led to automated formulation labs that accelerate formulation timelines from months to weeks.
Human clinical trials are the most costly part of the drug development process, sometimes costing >$100k per participant. Across the three phases of trials, common bottlenecks include the cost of testing equipment, finding and enrolling eligible patients, matching trials to sites, ineffective endpoints, and poor patient adherence.
AI platforms that screen health records for eligible patients make recruitment easier.
Clinical trial-site matching technology better qualifies potential collaborators to reduce startup time.
Novel endpoints (some AI-based) are proving to be more effective than existing ones.
After human trials, a robust data report must be sent to the FDA, which evaluates quality, trial conduct, and manufacturing feasibility. Generating the hundreds of thousands of pages needed for FDA review is a time-intensive bottleneck.
AI agents help produce necessary documentation faster than human-only approaches.
Early Innings
Certain development bottlenecks are especially prohibitive in newer therapeutic categories like cell and gene therapies, of which <50 are FDA-approved. In general, there are fewer eligible patients for clinical trials, making them slow and costly. Every “batch” must also be personalized to a patient, making manufacturing complicated. And despite efficacy, the one-time nature of cell and gene therapies makes payers less willing to reimburse and hospitals less likely to pay for the equipment and expertise needed to deliver treatment.
Much of my understanding of major drug-development pain points comes from anecdotal conversations with startups. I’m certain I’m missing others (such as scale-up and manufacturing), but I’m always looking to hear from other folks thinking about or working directly on technologies that are making it faster and cheaper to bring life-saving treatments to patients. It feels like we are in the early innings of seeing how pharma and biotechs adopt these time- and cost-saving technologies.
What I’ve been reading this week:
Toward Same-Day Genome Sequencing in the Critical Care Setting, NEJM
Role of Circadian Health in Cardiometabolic Health and Disease Risk: A Scientific Statement From the American Heart Association, AHA
Why I Run, The Atlantic
Eroom's law is hard to break. I hope the new tech will help rather than adding to the bloat.
IMHO we need to move away from brute force approach and focus on understanding the underlying biological process.