There's a metric space where distance predicts phase behavior.
Intrinsically disordered proteins (IDRs) drive biomolecular condensate formation — the liquid-liquid phase separation that organizes cellular interiors without membranes. Which IDRs condense with which others, and in what proportions, depends on sequence in ways that molecular simulations can compute but take enormous resources to explore systematically.
Tesei et al. (arXiv:2603.08300) introduce a thermodynamic model that learns low-dimensional, context-independent representations of IDR sequences. These representations define a metric space where the distance between any two IDRs corresponds directly to their thermodynamic compatibility — how similarly they behave in mixtures.
The model predicts multicomponent phase diagrams in quantitative agreement with molecular simulations, without having been trained on free-energy or phase-coexistence data. It was trained only on pairwise interaction parameters, yet it generalizes to arbitrary mixtures and concentrations. The representations are context-independent but combine to produce context-dependent interactions — the same IDR representation means different things depending on what it's mixed with, but the representation itself doesn't change.
Partitioning, condensation, and mutational effects all become geometric problems in this space. Whether a mutant IDR preferentially enters a condensate can be read off as a distance. Whether two IDRs are compatible in a mixture is a triangle inequality.
The broader move: sequence-dependent thermodynamics, which lives in an astronomically high-dimensional sequence space, can be compressed to a metric space where the relevant physics is distance. The compression works because what matters for phase behavior is a low-dimensional summary of each sequence, not the sequence itself. Biology wrote the function in 20 letters; physics reads it in a handful of numbers.
Tesei et al., "A thermodynamic metric quantitatively predicts disordered protein partitioning and multicomponent phase behavior," arXiv:2603.08300 (2026).