1: Techbios x Africa : The manifesto part 1 6523
Closer to Humans: The Next Big Opportunity in TechBio:
Hitting Eroom's law in translating assets to clinics
If Moore's law promises exponential gains from technology, Eroom's law (Moore spelled backwards) reminds us that drug discovery has stubbornly resisted that curve. For decades, the cost of bringing a new drug to market has roughly doubled every nine years, even as compute and data scaled exponentially. AI-driven TechBios were supposed to break this trend and accelerate discovery, lower costs, and flood the pipeline with new medicines. In its early day's, Recursion was going with something like a 100 drugs in 10 years.
And to some extent, they have delivered. Programs from Insilico or Recursion show how AI can compress preclinical timelines from five years down to 18–30 months. Costs are lower, throughput is higher, and in silico tools have expanded the space of molecules Pharma can explore.
But reality is that most AI-first drugs are still aimed at well-known targets, and once they reach the clinic, they face the same bottlenecks as traditionally developed drugs. Phase II proof-of-concept success rates hover at ~40%, unchanged.
Back to Eroom's law in action, the bottleneck has shifted downstream. The graph from speed invest tells the story nicely.
Early discovery (target validation, compound screening, lead optimization) accounts for ~25% of costs, the bulk of time and money is lost in Phase II and Phase III, where failure rates spike and costs per molecule can exceed 20–25% of the total.
Functional Data Is the Missing Piece
Why? Because our translational models are still inadequate proxies for human biology. Drugs fail not because they weren't optimized enough in silico, but because they don't behave as expected in humans, showing weak efficacy, unexpected toxicity, or adverse effects that outweigh benefits.
Conversely regulators are now pushing for more personalized approaches: genotyping, deeper disease phenotyping, and companion biomarkers to better stratify patients.
That means the next opportunity isn't about yet another molecule generator. It's about building the translation layer: generating functional, human-relevant data at scale.
Two pillars stand out:
Bench side. New experimental systems like organoids and organ-on-a-chip can capture human biology more faithfully than animal models, giving us early readouts of drug response in tissue that resembles real patients. it can be high-dimensional functional data (cells content imaging)
Bedside. Richer molecular profiling of patients to capture complete responses to interventions across all biological layers. The omics data, reflects physiological responses from the gene expressed to the protein inhibited till the end metabolite produced.
This is the frontier TechBios have yet to tackle.
Proprietary datasets from in vitro, in vivo, or in silico work aren't enough, because by design they remain at a distance from real human complexity.
Reminder, the demand is still there as the patent cliffs of 2030 are not going anywhere.
The Funding Gap: Bench Traction, Bedside Wide Open
The common denominator in TechBio is always the same: proprietary datasets.
On the bench side, we're already seeing how this can play out. Just last month, Parallel Bio raised $21 million to push forward a new model for immune drug discovery.
Their platform combines organoids and AI is set to generate massive proprietary datasets of immune responses. This 'Immune system in a dish' allows simulate how drugs behave across populations and verify candidates in vitro before they ever enter the clinic. The company dates back to 2021, but recent series A show their gearing up for growth and points to serious answers to the translation problems from Capital Interest.
The story on the bedside is very different. Here, the prerequisite is well-characterized patient data of omics like genomics, proteomics, metabolomics, deep clinical phenotyping. Not really the type of data you can engineer in your lab with enough wetware and hardware.
Pharma companies guard their clinical trials data as part of their asset. Biobanks have the scale needed but primarily share it with research partners and academics or monetizes them directly, selling access to screened samples and metadata at high prices. Their funded by goverments and charitable organizations around projects with defined partners within a consortio that have their for privilege access.
Hospitals typically generate only small, fragmented cohorts a few hundred patients, often disease-specific and far from the scale needed to train robust models.
And once TechBios push into later stages like preclinical or Phase I, costs spike: recruiting patients, managing trial sites, and running protocols and more tailored to big Pharma economics.
In the West, shrinking patient pools for many chronic diseases add yet another barrier driving the cost further up.
This imbalance explains why most visible TechBio innovation so far has come from the bench. Benchside players like Parallel Bio are proving you can generate your own data and own the feedback loop.
On the bedside, by contrast, barriers remain high and that leaves the space wide open.
The real question is not if bedside innovation will emerge, but where.
And it may well be that the answer lies outside the traditional Pharma hubs
You are leaving Bluwr.
We cannot guarantee what's on the other side of this link:
