Latent Space Interviews Xaira's Bo Wang and Ci Chu: 'Causal Models Need Causal Data' for Drug Discovery

Latent Space·medium signalXBlueskyLinkedInCopy link

Latent Space published a new episode with Bo Wang (Chief Discovery Officer) and Ci Chu (Chief AI Scientist) of Xaira Therapeutics on the X-Cell model, arguing that biology foundation models trained on scraped observational data cannot support causal inference — so Xaira is vertically integrating its own perturbation data generation instead of scaling on public corpora. The thesis is a direct rebuttal to the 'just scale on existing data' playbook that worked for text. For builders, the transferable lesson is that in domains where your model must answer interventional questions ('what happens if I change X'), owning the data-generation loop is the moat, not the model architecture.