The biopharmaceutical industry is betting its future on AI. With tens of billions invested, a clear consensus has formed that AI will fundamentally reshape how we discover and develop medicines. The initial results are promising: AI-native companies are growing their pipelines by nearly 40% annually, have demonstrated the ability to shorten preclinical timelines from years to just 12-18 months, and are hinting at much higher success rates in early clinical studies.
Great news right? Absolutely! But this progress hides an important paradox. Our advanced algorithms are being starved of their most essential resource: high-quality, context-rich biological data. Computational power and analytical efficiency have dramatically outpaced our ability to generate the meaningful biological data required to power drug discovery. The single greatest bottleneck in the AI-driven future of medicine is not the sophistication of our algorithms, but the scarcity of the right kind of data.
So what is the right kind of data?? It’s my view that AI-ready, spatially resolved biological data is the new gold standard for drug discovery and the key to unlocking the next generation of medical breakthroughs.
Why More Data Isn't Always Better Data
The problem facing AI in drug discovery is not a lack of data, but a crisis of quality, structure, and context. The industry is flooded with genomic and proteomic information, but this data was generated for human interpretation, not for algorithmic analysis. This creates two fundamental problems:
- The 'Structure' Problem: Most biological data lacks "machine-readiness". It is stored in fragmented formats across disconnected systems with inconsistent metadata. As a result, data scientists spend an enormous amount of time and resources on preprocessing, filtering, and standardizing datasets before an AI model can even begin its work. In a recent poll, 71% of biomedical researchers identified finding clean, structured data as the single biggest hurdle to AI adoption. The consequences are high here - I’ve seen first hand how cutting edge models trained on inconsistent data fail to perform at anything better than predicting which lab did the experiment.
- The 'Context' Problem: The most profound flaw in legacy datasets is that they destroy the native spatial context of the tissue. To achieve scale, conventional methods dissociate a tissue sample into a suspension of individual cells, losing context of where each cell was located and how it interacted with its neighbors. This is like trying to understand how a car engine works by studying a disorganized pile of its components. You can identify every part, but without knowing how they fit together, you cannot understand function or diagnose a problem.
This loss of spatial information forces AI models to search for patterns in noise rather than identify causal mechanisms in biology.
The Solution
The solution is a paradigm shift in data generation: spatial biology. These technologies preserve the physical architecture of the tissue, creating a functional "map of the system" instead of a dissociated "list of parts". By preserving the essential information of "where," spatial omics provides the "why" behind biological function.
While several forms of spatial omics exist, high-plex spatial proteomics offers the most direct and actionable map for drug development. Proteins are the functional machinery of cells and the direct targets for the vast majority of modern therapies. At Terrain, we believe that this technology is the definitive solution for multiple use cases, and it forms the core of our approach to solving the industry’s data crisis.
- Use Case 1: Better Target Identification: By visualizing protein interactions in the context of diseased tissue, spatial proteomics helps researchers uncover causal disease mechanisms, not just correlations. Early-stage researchers can pursue novel drug targets with a higher degree of biological validation from the start.
- Use Case 2: Higher Clinical Success Rates: High-plex data can power complex spatial signatures - patterns of protein expression and cellular organization - that are highly predictive of a patient's response to therapy. Improved patient stratification can directly increase clinical trial success rates, saving both development time and cost.
- Use Case 3: Decoding Therapeutic Response and Resistance: By mapping the tumor microenvironment before and after treatment, we can provide definitive answers as to why a drug works in one patient but fails in another. This provides the unambiguous efficacy signals that clinical researchers need to make confident decisions about their drug candidates.
- Use Case 4: Predicting and Avoiding Toxicity: Unforeseen toxicity is a primary cause of late-stage trial failures. By analyzing a drug's effect on healthy tissue, we can identify potential toxicity signatures far earlier than traditional methods, improving the probability of success across both efficacy and safety.
The cumulative impact is profound, with the potential to reduce time and cost in the discovery phase by an estimated 25% to 50%.
The Dawn of the Data-First Era
For decades, drug discovery has been a target-first endeavor. I believe that the future of medicine is data-first. In this new era, leadership will be defined not by the number of assets or targets in a pipeline, but by the quality and predictive power of the data engine that generates them. Solving the industry's data bottleneck is the key to unlocking the promise of AI in medicine. Spatial proteomics provides the definitive solution by reintroducing the essential dimension of physical context.
Please reach out on our website if you’re interested in learning more about how Terrain approaches this challenge!
References:
- Jayatunga, M. K. P, et al. AI in small-molecule drug discovery: a coming wave? Nat. Rev. Drug Discov. 21, 175-176 (2022). doi: 10.1038/d41573-022-00025-1
- Xu, Z. et al. A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: a randomized phase 2a trial. Nat. Med. 31, 2602–2610 (2025). doi: 10.1038/s41591-025-03743-2
- Jayatunga, M. K. P., et al. How successful are AI-discovered drugs in clinical trials? A first analysis and emerging lessons. Drug Discov. Today 29, 104009 (2024). doi: 10.1016/j.drudis.2024.104009