Fig. 3: Consistent convergence of phase mapping across replicates.

A Schematic illustration of the experimental workflow used to produce replicated phase diagrams. The initial set of experimental samples is selected by farthest point sampling, resulting in different starting points for each replicate. Each subsequent cycle then follows a unique path to reach the same phase diagram. Reproducibility across replicates is expected only if the machine learning, formulation, and analysis steps are consistent. B Phase diagrams, showing phase separation in blue and no separation in red, with probability prediction fits (background), validated points (colored dots), and new sample selections (white dots) for three experimental replicates of poly-L-(lysine)₁₀₀ and poly-L-(aspartic acid)₂₀₀ condensates. A total of 72 data points is experimentally validated across 9 cycles. Representative cycles are shown, remaining cycles and entropy maps are reported in Supplementary Figs. 4, 5 and Supplementary Figs. 10–13. C Balanced accuracy plot showing the accuracy on the prediction for each successive cycle with respect to a “ground truth” phase diagram. Cycle 0 represents the balanced accuracy computed with respect of a randomly generated phase diagram as a baseline comparison. D Average Jensen-Shannon Divergence plot illustrating within-experiment divergence by comparing consecutive cycles for each replicate. This reflects the progressive convergence toward the final phase diagram for each replicate. Cycle 0 is a random phase diagram included as a reference for low similarity. E Average Jensen-Shannon Divergence plot comparing divergence across replicates at each cycle, highlighting inter-experiment variation. Cycle 0 compares three random phase diagrams and is included as a reference for low similarity. Phase boundaries are indicated by dotted lines; in some cases, these may be partially obscured by overlapping contour lines. Polypeptide consumption per phase diagram: 4.0 mg poly-L-(aspartic acid)₂₀₀ and 4.2–4.8 mg poly-L-(lysine)₁₀₀.