The Dawning Age of Biology

In 2012, Jack Scannell, a pharma industry expert, coined the term “EROOM’s law.” This was a vivid way to contrast the declining rate of productivity in drug discovery with the rapid acceleration in computing power that became known after the 1960s as Moore’s law.

Eroom’s law stated that the number of new drugs approved per billion US dollars spent on R&D has halved roughly every 9 years since 1950, an 80-fold decrease in inflation-adjusted terms.

In short, it has taken more money year-on-year to bring each new drug to market.

We’ve seen extraordinary advances in artificial intelligence and machine learning with massive impact across other sectors of society. Yet, the cost of drug discovery and development continues to be driven primarily by failure, with only ~10% of clinical-stage (in-human) drug candidates receiving regulatory approval. Around 75% of the substantial attrition in Phase 2 trials is because of lack of efficacy or for safety reasons. In short, we do not sufficiently understand human biology when beginning a drug discovery program. Identifying a target remains the first, and often most difficult, decision in drug discovery. 

Finding the Right Target

In 2007, maraviroc, a drug I had worked on at Pfizer, was approved. We took it from concept, through molecule design, clinical trials, and ultimately commercialization. The idea for the drug came from a surprising observation in two separate scientific papers from 1997. They reported that a group of sex workers and gay men avoided AIDS while the disease ravaged others. This protection stemmed from an altered gene for the CCR5 receptor, a receptor on immune cells that HIV uses to enter and infect them. The genetic alteration hindered this process, effectively suppressing viral transmission and pathogenesis.

Thus, from a genetic mutation, we identified a potential drug target. Maraviroc inhibits HIV binding to human cells, protecting patients. It was one of the first drugs for a common condition discovered by understanding how a genetic mutation affects protein function. This demonstrated the power of human genetics in identifying effective drug targets.

Around the same time, therapeutics for orphan diseases also began emerging, driven by the FDA’s Orphan Drug Act (1983). Rare (orphan) genetic disorders became a focus for the industry, as protein dysfunction is often clearly traced to inherited genetic changes. Discovery research in the 1980s pointed to the role of genetic mutations in cancer cells, present across cancer types, which, in turn, enabled more targeted oncology therapies.

These developments underscored how understanding human genetics improves drug survival and helped reverse the decline in R&D productivity. In 2015, my colleague Matt Nelson published a seminal paper showing that approved drugs with a genetic link to the disease were 2–3x more likely to succeed than those without, even among drugs developed before genetic profiling became widespread.

Why Hasn’t This Fully Transformed Drug Discovery?

Common and chronic diseases are often polygenic, driven by many genes with small effect sizes. For example, in osteoporosis, over 1,000 genetic variants may contribute to disease risk. About 90% of these lie in the non-coding genome, the 95% of DNA that regulates the protein-coding genome. Until recently, this ‘dark genome’ has been difficult to study.

The biology here is far more complex than a single gene defect. Over the past decade, technologies have emerged that allow us to visualise biology in high definition, from genome mapping to single-cell expression analysis in healthy and diseased tissues. We can simulate genetic changes with tools like gene editing, all in human cells, not animals.

But the resulting data is vast, far too much for the human brain to comprehend. The human genome contains 3 billion base pairs, encoding over 20,000 proteins, each modifiable in countless ways.

As Sydney Brenner noted in his 2002 Nobel Prize speech: “We are drowning in a sea of data, but starved of knowledge.”

Machine Learning Meets Biology

In the past five years, machine learning (ML) has matured. Biology can be seen as a language, DNA is the language of life, and large language models (LLMs), like those behind ChatGPT, can help interpret the non-coding regulatory genome.

Between any two humans, one in every thousand DNA base pairs diverges. Multiply that over 3 billion base pairs and you get millions of differences. LLMs can help us understand why one person gets a disease while another remains healthy.

This addresses only part of the puzzle. Genetics may hint at causality, signalling dysregulated proteins but not reversibility. Can modulating these targets restore health? Consider Alzheimer’s as billions have been spent on reducing amyloid beta plaques, with minimal efficacy. While genetics may explain plaque development, the disease is multifactorial. A unidimensional view has failed to yield cognitive improvements.

Navigating Complexity

We must address biology’s combinatorial complexity. With over 20,000 genes (and proteins), CRISPR technologies offer >200 million two-gene perturbations, or 1.333 billion for three. We need multi-gene manipulation to understand signaling cascades and cellular behavior.

Even the most advanced experimental systems fall short here. But there may be other approaches enabled by the integration of the new technologies. Instead of performing all experiments simultaneously, we can iterate, using ML techniques like reinforcement learning or active learning to design batches of experiments. AlphaGo followed this same philosophy.

A New Era, A New Convergence

The Nobel Prizes in 2024 signaled why the transformation of drug discovery is now:

• Chemistry: David Baker (University of Washington), Demis Hassabis, and John Jumper (DeepMind, UK) for decoding protein structure, the building blocks of biology.
• Physics: Geoffrey Hinton and John Hopfield for laying the foundations of modern AI.
• Medicine: Victor Ambros and Gary Ruvkun for discovering microRNAs, tiny cellular controllers that regulate gene expression like volume knobs.

These technologies enable discoveries that will transform drug discovery and medicines. We are now working at a new pace, “code to code” where our understanding of human genes and cells will drive the survival of drug targets and candidates.

Moore’s law, once a predictor of chip performance, now underestimates the impact of modern GPUs (e.g., Nvidia), which allow us to train multi-million-parameter models on biological data.

But these models still require carefully constructed datasets, paired clinical phenotypes, healthy vs diseased cells, for sequencing and training. The models of today simply don’t have enough of the right kind of training data.

We are on the cusp of a significant step forward in understanding common, chronic and complex diseases. Realizing that the future will depend on integrating two capabilities: rich measurement of human biology and computation that can turn those measurements into testable hypotheses. Biology is undeniably complex, and no single technology will allow us to understand it. Progress will come from disciplined integration: combining high-resolution human data, careful experimental design, and computational and machine-learning models that are continuously tested against reality. 

Sydney Brenner once said, “Progress in science depends on new techniques, new discoveries, and new ideas, probably in that order.” With advances in wet-lab methods and in silico ML/AI, we now have the opportunity to build discovery systems that reliably connect techniques to discoveries, and discoveries to new ideas that can be pursued as medicines.

At Relation, the company I have the privilege to lead, we are building precisely that capability. We use an engineered Lab-in-the-Loop approach that couples single-cell and multi-omics measurement with functional biology and machine learning, so predictions drive experiments and experimental results systematically refine the models.

Our aim is straightforward: to reduce biological uncertainty earlier in discovery, so that the medicines advancing into development are based on stronger evidence. It is an ambitious goal, rooted in our mission to understand biology to cure disease: There’s little doubt in my mind that we will see a Moore’s law-like phenomenon in biology.