Cross-species transferability of an ultrastructure-to-dynamics compiler

Determine whether an ultrastructure-to-dynamics compiler trained on data from one species can transfer with high predictive accuracy to another species.

Background

Domain shift across species is a core risk for learned mappings from structure to dynamics. Even if many molecular functions are conserved, structural and physiological differences may hinder transfer.

Establishing whether and when a compiler trained in one species generalizes to another is essential for broad applicability and avoids overfitting to species-specific features.

References

Although many molecular functions are conserved across species, it remains unclear whether a compiler trained in one species would transfer well to another.

— Compiling molecular ultrastructure into neural dynamics  (2603.25713 - Kording et al., 26 Mar 2026) in Appendix 1: potential limitations — Data and measurement limits (Generalization risk across brain regions, species, and conditions)

In biology, whether the predictive architecture generalizes to uncharacterized genetic circuits or alternative bacterial species remains unknown; flagellar-driven collective motility involves complex hydrodynamic and surfactant interactions~\citep{kearns2010field, shaw2026engineered} that can produce emergent non-linear behaviors.

— Accelerating Scientific Research with Gemini in the Real-World  (2608.26701 - Schmidgall et al., 27 Aug 2026) in Limitations and failure modes, Section 2.3, paragraph “Generalization boundaries”