The New Bottleneck in AI Drug Discovery: Knowing Which Biology to Trust
QurieGen Weekly Signal Scan No. 19 - AI × Cells
The new bottleneck in drug discovery is not finding more candidates. It is knowing which biology deserves the next expensive step.
This week’s selected signals show a clear shift across AI drug discovery, virtual cell models, lab-in-the-loop workflows and single-cell multiomics: the field is moving from generating more hypotheses toward building better evidence for which hypotheses should be trusted.
Six signals shaping AI × Cells
1. AI may create more candidates, not fewer bottlenecks
In AI Won’t Fix Biotech’s Biggest Problem, Matt Shlosberg argues that AI could accelerate candidate generation without solving the deeper bottleneck: human validation.
That distinction matters. If AI produces more plausible targets, molecules and mechanisms, the industry still needs to decide which of them deserve costly experimental work, clinical translation and ultimately testing in humans. The challenge may shift from scarcity to overload.
2. “Developing a binder” is not the same as “developing a drug”
A recent SynBioBeta reflection from John Cumbers captured an important point: developing a binder is not the same as developing a drug.
Static prediction can be powerful, but biology keeps moving. Drug development depends on efficacy, toxicity, manufacturability, pharmacology, immune interactions and context-specific biological response. In other words, the hard question is not only whether something can bind. It is whether it can work safely and meaningfully in a living biological system.
3. Tempus and Bristol Myers Squibb expand multimodal AI collaboration
Tempus and Bristol Myers Squibb expanded their collaboration to use AI, multimodal real-world data and data science techniques to optimize clinical trial design and improve development decisions.
This is another signal that AI in drug development is moving closer to decision support: pressure-testing assumptions, understanding patient heterogeneity and improving the probability of technical and regulatory success.
4. Mechanisms matter in cellular perturbation biology
A new bioRxiv preprint, Mechanisms Matter: Transportability of Cellular Perturbation Effects, highlights why perturbation effects cannot be treated as simple, static readouts.
Cellular responses can shift across cell types, biological contexts and datasets. That makes mechanistic understanding essential, especially when models are expected to generalize beyond the setting in which they were trained.
5. Rancho BioSciences launches OmicsHQ for AI/ML drug discovery
Rancho BioSciences launched OmicsHQ, a curated multi-omics data platform built to support AI and machine learning workflows in drug discovery.
This reflects a broader trend: AI performance depends heavily on the quality, structure and biological relevance of the data beneath it. More data alone is not enough. Curated, interpretable and context-rich biological data is becoming a core infrastructure layer for drug discovery.
6. Stanford highlights AI, proteome mapping and perturbation-led virtual cells
Stanford recently highlighted work on AI, proteome mapping and the future of virtual cell models. The article points toward an important direction: mapping proteins in space and time, with perturbation measurements as a next step toward more predictive virtual cell models.
For virtual cells to become useful in real drug discovery decisions, they will need more than static maps. They will need dynamic biological response data that captures how cells change when perturbed.
There is a clear pattern
The field is moving from: “Can AI generate something?” to: “How do we know what deserves trust and should be moved toward humans?”
This is where experimental biology becomes more important, not less. Public datasets, curated multi-omics platforms, spatial maps and real-world data all matter. But the hardest decisions still need fresh experimental evidence: how cells respond, which pathways move first, which intracellular signals change, and which targets look causal after perturbation.
This is the layer QurieGen focuses on: temporally resolved cellular response data across RNA, proteins and phosphoproteins to support stronger target discovery and validation decisions.
By combining single-cell multiomics, intracellular protein and phosphoprotein measurements, and lab-in-the-loop experimental design, QurieGen aims to help move drug discovery from more predictions toward more decision-grade biology.
Learn more
Explore how QurieGen is building human-relevant, temporally resolved single-cell multiomics for target discovery, validation and AI-enabled drug discovery:
www.quriegen.com