In the latest episode of The Genetics Podcast, Patrick Short sits down with Dr. Brent Richards, CEO and founder of 5 Prime Sciences and Professor of Human Genetics, Epidemiology, and Medicine at McGill University. Brent is an endocrinologist by training who has spent his career using human genetics to work out which risk factors actually cause disease. His argument in this conversation is precise: genetic evidence only de-risks a drug program when teams interrogate its biases, rather than treating "human genetic support" as a box to tick.
The conversation moves from why so many well-supported targets still fail, to the methods that separate a genuine causal signal from a convincing artifact. Below are the takeaways most relevant to biotech and pharma teams making early target decisions.
Brent's central concern is how genetic evidence gets used inside a program. A single confirmation that support exists tends to close the discussion instead of opening it.
As he puts it: "Somebody says, 'Is there human genetic evidence to support this?' 'Yes.' 'Okay, move on to the next question.' That's not a very good place to land. You want to have a really good understanding of what are the sources of bias in this tool set so that you're not just checking a box and moving on to the next question about pharmacokinetics of the target." The cost of that shortcut shows up in the clinic. Brent ties most late-stage failure to efficacy, and efficacy to target choice: teams advance targets that look strong in model systems and observational epidemiology but do not cause the disease in humans.
Brent frames each genetic method as a tool with a known profile. GWAS is highly sensitive. Proteogenomic Mendelian randomization adds a causal biomarker and a direction of effect. Exome-wide association studies are the strongest signal when available, though they rarely are. Confidence comes from convergence across methods whose weaknesses do not overlap.
He describes the logic directly: "All of them have limitations, and all of them have strengths, and the trick is to be able to triangulate across them so that the different types of biases from the different types of study designs become orthogonal to each other. And so if you end up getting the same answer from different points of view despite those orthogonal biases, then you probably have a much more probable drug development program that's going to succeed." An example he mentions is a 2012 analysis of bone mineral density GWAS, which recovered every known osteoporosis drug target. This is evidence that the signal can align with biology when the methods agree.
One of the sharp findings in the episode concerns how targets get chosen in practice. Brent points to the volume of published papers on a target as the dominant factor in whether it attracts money, and argues this reflects herd psychology more than biology.
His numbers make the gap explicit. Paper count on a target is the single strongest driver of whether a sponsor will start a program, carrying an odds ratio of five. However, literature volume has no relationship to whether the resulting drug gets approved. The odds ratio for that is one, with tight confidence intervals around the null. The practical implication is uncomfortable for novel biology. Targets in less-studied areas struggle to attract capital precisely because few peers have moved first, even when the underlying evidence is sound.
Genetic support is not a guarantee of direction. Brent uses the Zeus trial of IL-6 inhibition, run by Novo Nordisk, as a case where genetic backing did not prevent failure. With signaling cytokines, secondary signaling can produce a reverse effect, so even a well-powered causal method can return a directionally wrong answer. Brent is candid that proteogenomic Mendelian randomization, one of his preferred tools, can land directionally opposite to the truth.
Bias in how a trial population is assembled compounds the problem. Selecting patients through two independent risk criteria can induce a spurious negative correlation between them, a pattern known as collider bias, and Brent notes it has distorted reanalyses of the PCSK9 trials FOURIER and ODYSSEY. Randomization does not remove it. That point carries real weight for polygenic-risk-stratified trials, where selection on genetic and non-genetic risk is built into the design.
Brent's framework is demanding by design. He asks teams to name the biases in each method, seek convergence across independent lines of evidence, and question how patients entered a study before trusting the result. His motivation is clinical rather than academic. He splits his time between McGill and 5 Prime Sciences because, in his view, every intervention he can offer a patient draws on both academia and industry, and the gap between them is where good targets get lost.
Listen to the full episode below.