Big Picture Bio launches with £2.2M to design cancer combinations

Why it matters: Combination therapies are central to oncology, but choosing which drugs to combine — and in what dose, sequence and patient population — creates a huge experimental search space. Big Picture Bio wants to narrow it computationally before combinations reach the lab.

Backstory: The company was founded by former Deep Science Ventures executives Kerstin Papenfuss and Mark Hammond. It raised £1.5 million in a pre-seed round co-led by Kadmos Capital and Exceptional Ventures, with participation from Gloucester Ventures and angel investor John White, alongside £700,000 in non-dilutive funding from Innovate UK.

How it works: Rather than training a conventional machine-learning model to spot patterns in existing datasets, Big Picture Bio says it has built a system around human-defined principles governing cancer biology, the immune system, drug resistance and delivery.

• The platform works through the causal chain behind a treatment response and produces an explanation alongside its prediction. “Every prediction is a readable causal chain, not a score,” Papenfuss and Hammond told European Biotechnology. “We give you the argument behind it, so the people making decisions can review it, understand it and challenge it.”
• The system draws on single-cell and proteomics datasets, CRISPR maps, scientific literature and clinical trial results. The company says its own wet-lab data will increasingly be used to fill gaps in the model.

Testing the model: Big Picture Bio has also used clinical trial outcomes as a stress test. The founders said the system prospectively called five of six selected ASCO trial readouts correctly and currently has an overall prospective accuracy of about 86%. “What we’re most proud of is that all predictions above >65% confidence have all been correct so far so we’ve learnt that we can really trust the confidence values and be much more cautious in the lower ranges,” added the founders.

• One example was Regeneron’s fianlimab melanoma study. Big Picture Bio says it predicted before the readout that both tested doses would miss their primary endpoint. For the high-dose arm, it predicted a progression-free survival hazard ratio of 0.83 to 0.90; the reported result was 0.845.

Yes, but: Predicting trials is not the end product. “Trial prediction isn’t our main goal,” the founders said. “We want to design new combinations that unlock biological synergy and produce much bigger jumps in response.”

What’s next: The first combinations are now moving into wet-lab testing and mostly involve approved or clinical-stage drugs. Big Picture Bio plans initially to focus on solid tumors where extensive single-cell datasets are available. “We can see a route to 50%-200% OS (overall survival) improvements with existing drugs in the right combination for some of the hardest cases. This will begin to prove out that synergistic effects are predictable,” said the founders.

• The startup’s early business model could also give failed drugs another shot. The company plans to look particularly at parked or deprioritized assets, license individual drugs where necessary, and build new intellectual property around the combination, dosing, sequencing and patient selection.
• As the founders put it: “We take drugs that are cheap because they failed alone, use the model to design the combination that makes them work, and hold the IP on that combination.”