How We Choose Drug Target is Changing - Fast
I’m fascinated by how we go from “what could go wrong in disease?” to “this protein is my arrow.” Today’s target-selection methods feel like hunting in the dark: one-off guesses in dishes, genetic clues here and there. That’s slowly giving way to something wild, and I think it’s really cool, especially what Gordian Biotechnology is doing.
What we did (traditionally)
Target selection used to mean:
Picking a gene or protein based on fragmented clues, like genetics, mouse models, or scaffolding from known drugs.
Testing compounds in dishes. For example, see if X molecule flips Y cellular switch.
Relying on animal models one molecule at a time.
This method works, but it’s slow, expensive, and often misleading. Many targets fail once they hit patients, because dish and mini-mammal models barely reflect the complexity of real, aged human biology.
Living animals as discovery engines
Gordian does something daring: they don't wait to optimise a therapy before testing it. They test hundreds of gene therapies in one living animal. Audacious and necessary.
Here’s how:
Patient Avatars: They use animals—with diseases that humans also experience, like horses with osteoarthritis or primates modeling metabolic syndromes. These aren’t lab rats—they’re living, complex mammals.
Mosaic Screening: Hundreds of gene-therapy variants, each with unique barcodes, are delivered into a single animal—so every barrelled cell tests a different hypothesis .
Pythia: AI that reads single-cell sequencing data and links each barcode to cellular reactions in vivo .
They’ve dubbed it “stepwise different” in practice .
To improve efficiency and accuracy in predicting drug targets, the trend leans towards using AI and single-cell tech or focus on better target prediction. Being bold to integrate these elements into in vivo high-throughput screening is a fresh approach, and a quite ambitious one. The statistics delineates a quite positive picture, with 80% accuracy prediction of cartilage regeneration in horses with osteoarthritis, and matching 13 out of 16 known clinical outcomes in a mouse model of metabolic-associated steatohepatitis (MASH). It worked, but why?
Why did it work?
Biology in real context
Gordian’s core strength is testing therapies inside living, aged animals that naturally develop diseases similar to humans, like horses with osteoarthritis or primates with metabolic syndrome. This is a game-changer compared to traditional cell culture or young rodent models, which often fail to replicate the complexity and chronic nature of human diseases.
Speed!
Instead of testing one drug candidate or one gene target at a time, Gordian tests hundreds simultaneously inside one animal. They barcode each gene therapy and use single-cell sequencing to track which therapies trigger beneficial effects. This massively accelerates discovery and reduces cost, allowing them to explore a huge landscape of possibilities quickly.
AI…of course
Their AI, Pythia, learns from the complex single-cell data generated inside these animals to identify promising targets with higher predictive power than traditional approaches. It cuts through noisy biology to find signals that could translate to human therapies.
Drug testing in complex mammals, or even straight in human organs, seems to be a catching on a trend, with reason. Ochre Bio, an EIT Health-supported start-up that signed $1BN+ deal with Boehringer Ingelheim, models chronic liver disease using real human livers. Don’t panic.
“Our Liver ICU takes donor livers that can’t be used for transplant because they are slightly diseased or old or fatty,” said O’Meara. “We keep them alive on OrganOx perfusion machines for five days plus, while we study the effects of our RNA therapies on the liver physiology.”
The strategy dramatically improved the development timeline, shortening it to days instead of months, even years.
Something to learn from these strategies:
Start with biology, not hypotheses
Scale smarter
Combine tech
Don’t simplify complexity sometimes
Be ambitious
Some educated prediction for drug discoveries in the next 5 - 10 years
Target discovery will shift from hypotheses to system-level data mining
Drug targets will increasingly be discovered using multi-omics, single-cell atlases, and AI-driven causal inference, rather than relying on traditional disease hypotheses. Human disease is more complex than single-gene or single-protein dysfunction. The shift toward understanding whole-cell states and regulatory networks will allow for more robust, translatable targets. This can be supported by Nature Medicine (2021), which Highlights the importance of single-cell technologies in reclassifying diseases, as well as Gordian Biotechnology.
In vivo platforms will replace simplified disease models
High-throughput in vivo systems - like mosaic screening in aged animalsm - will become a gold standard for early validation of therapies. Lab models (e.g., immortalised cell lines, young mice) often don’t reflect real disease biology, especially for aging-related conditions. Scannell et al., 2012 highlights translational failure due to unrealistic preclinical models, and Elliot Hershberg’s Century of Biology Newsletter (2024) on CRISPR describes how context-rich systems improve gene therapy precision. Gordian’s success shows early validation in naturally diseased animals.
Therapies will target cellular programmes, not just genes or proteins
Drugs will increasingly be designed to modulate entire cellular states, such as senescence, inflammation, or regeneration readiness, rather than single targets. Complex diseases (e.g. cancer, Alzheimer’s, fibrosis) are driven by dysregulated cell fates, not just broken genes. Rood et al., 2024 introduces the Human Cell Atlas and emphasizes cell-state–based understanding, startups like Unity Biotechnology develop drugs that target senescent cell states, not just individual pathways, and the establishment of research labs like Altos and Calico focusing on cell rejuvenation programmes.
Generative AI will be used to design novel drugs and cell behaviors
AI tools will not only assist in drug selection but generate new molecular designs and potentially new therapeutic modalities (e.g., synthetic circuits, protein scaffolds). Foundation models trained on chemistry, protein folding, and cell states can generate novel designs beyond human intuition. Isomorphic Labs (DeepMind) uses AlphaFold-derived models for structure-based drug design, Salesforce’s ProGen demonstrated de novo protein generation using large language models.
Aging will become a treatable indication category
Aging itself will gain recognition as a modifiable and regulated target category, leading to approved geroprotective drugs. The FDA has already allowed clinical trials with aging-relevant endpoints (e.g., TAME trial), and biotech firms are advancing senolytics, NAD boosters, and cellular reprogramming. TAME Trial (AFAR) first attempt to test metformin for delaying multiple age-related diseases, and the emergence of startups like Loyal and Gordian signifies an early shift in perspective in viewing aging as treatable system dysfunction.
Drug discovery will become modular, distributed, and faster
With cloud labs, robotic platforms, and digital twins, the time from idea to IND (Investigational New Drug) will shrink dramatically. Companies like Emerald Cloud Lab and Strateos allow remote-controlled, fully automated biology experiments. Combined with LLMs for scientific planning, the R&D cycle is accelerating. Focuses of investors, such as Andreesson Horowitz’s Bio Builders, sescribes shift to API-first biology and cloud-native drug R&D.
See ya next time:)
