The GPS for drug development: navigating with evidence from human populations

More than 50% of drug programs fail due to lack of efficacy. While the reasons are manifold, the root causes are simple. First, modulating the target has no effect on the disease. Second, biomarkers are either not available or unsuitable to guide clinical development. Third, therefore, the trial populations are inappropriate to demonstrate a clinically relevant benefit. In other words, stakeholders lack critical evidence to align drug targets, biomarkers, and disease populations to navigate the best development paths. Consequently, many programs fail, value is lost, and diseases are left untreated. There has never been a better time to fix this: the explosion of biomedical data of millions of humans linked with genetic information and clinical outcomes has sparked a paradigm shift in the industry. Evidence from human populations has already helped to uncover critical pathways, illuminate disease mechanisms, and now underlies an increasing number of drug programs. Targets with human genetic support have seen a two- to threefold higher likelihood of success in the past, and additional data are emerging every week. Similarly, scalable assays of molecular processes of human biology, such as proteomics and metabolomics, facilitate the identification of novel biomarkers from human populations, and these almost double the likelihood of success when used to guide clinical development. Companies at the forefront of drug development have recognized the value of human evidence already: Alnylam, an RNAi therapeutics company, reported a cumulative success rate of >60% compared with the industry-wide rate of 10% (https://www.alnylam.com/sites/default/files/pdfs/2022%20Alnylam%20Corporate%20Responsibility%20Report.pdf). Pfizer, a large pharmaceutical company, has achieved a tenfold increase in success rate from 2% in 2010 to 21% in 2020, and human evidence has been central to the improvements. They are not alone: Regeneron has built the Regeneron Genetics Center with the central aim to 'harness the power of human genetics to discover important new medicines, validate existing research programs and optimize clinical trials'; the Novo Nordisk Foundation has recently opened up a new research center at the Broad Institute to 'explore human gene regulation in connection with common complex diseases'. Along with others, they invested in large human population cohorts, including the UK Biobank and FinnGen, to prioritize targets with human genetic support, identify suitable biomarkers, and guide clinical development. The question is, how can human evidence become the industry norm? The truth is that navigating drug development with human evidence is difficult. Many companies now want to base decisions on evidence from large human populations, but the accessible evidence is too sparse, difficult to interpret, and often falls short of value inflection. The core reason is that, although we have begun to understand the architecture of many human diseases in the past two decades, our concept of diseaseitselfremains arcane. Our current representations of disease are sparse collections of signposts rather than continuous coordinates: only if a patient hits a known symptom (e.g., fatigue) and ends up with an established measurement (e.g., antibodies against double-stranded DNA), can we assign a diagnosis – in this case, systemic lupus erythematosus (SLE). The space in between is uncharted, confining our ability to navigate from drug targets over disease biomarkers to trial populations to very few well-threaded paths. Today, we lack a map to navigate drug development with evidence from large human populations. This map is what we are building at Pheiron: our platform uses artificial intelligence (AI) to learn quantifiable representations of disease from human multi-omic data that systematically reflect clinical phenotypes and their trajectories over time. Our models integrate the whole range of available complex biomedical data from established clinical markers (e.g., clinical symptoms or blood counts), broad multi-omic assays (e.g., blood transcriptomics, proteomics, and metabolomics), to organ-specific imaging and tests (e.g., heart ultrasound scans and electrocardiography, lung spirometry, and vision and hearing tests). The resulting disease representations capture unique aspects of human health, together forming continuous coordinates that connect the sparse signposts (e.g., 'fatigue for three weeks') to a high-resolution map (e.g., 'the proteomic risk of developing SLE or risk of subsequent terminal kidney failure'). Each new way to probe human biology, physiology, and pathology (such as a new molecular assay or a novel imaging technique) helps us to refine it. Our map enables us to navigate drug discovery with human evidence: we build human (genetic) evidence to establish links between targets and precise clinical phenotypes (based on the map coordinates). With our map, we identify clear paths for indications to go for, safety signals to consider, and suitable biomarkers to measure. Positioning individuals on our map, we inform the selection of the right individuals and design of the trials to demonstrate efficacy in clinical development. In short, our platform identifies safe and effective targets for the disease and suitable biomarkers to select the right people to run the trial – before a single patient is recruited. The R by contrast, Alnylam's success rate is >60%. More than 50% of clinical-stage failures are efficacy-related, and >25% are safety-related. Targets supported by genetics have had a 2.6-fold higher likelihood of success in the past. The amount of available genomes is doubling every year. Programs supported by biomarkers have had an almost twofold greater probability of success. T.B. and J.S. have a >5% ownership stake in Pheiron, Inc.