#8 · Systems biology

Priors and models fuse heterogeneous measurements to constrain biological mechanism

The shared thread is the integration of heterogeneous measurements with structured priors or mechanistic models to constrain inference about biological states. CircuitATLAS couples a systems neuroscience knowledge graph with an agentic LLM workflow, nominating ATP1A3 as a circuitopathy target that was validated in vivo. Cross-species representation learning applies dual-rule contrastive learning to align mouse and human electrophysiological states, and the alignment tracks clinical drug efficacy across model-drug combinations. Uncertainty-aware soft-constraint integration of deuterium metabolic imaging with genome-scale metabolic models reveals diet-associated remodeling of feasible solution space in mouse skeletal muscle. What is new across these items is the explicit encoding of uncertainty, cross-species correspondence, or knowledge structure as constraints within a single modeling framework, rather than analysis of each modality in isolation.