Fig. 5: Robustness and generality.

a Designed DCA sequences at 60 mutations from the WT. The histogram shows the activity scores of 991 detected designs (out of 1000 designs) with 22 active sequences beyond the activity threshold (equal to −2.76, red dotted line), corresponding to 2% of the pool. b Comparison of active fraction as a function of the number of mutations for the RUM and DCA-SB models, for different MgCl₂ concentrations: 60 mM as our standard condition (solid lines) and a lower concentration of 5 mM (dashed lines). N per bin per model provided in the Source Data file. Dots are the active fraction. Error bars upper (lower) bound is the active fraction including (excluding) activity scores within the 98.5 percentile of the measurement error distribution around the threshold. c Sequencing score at 37 °C versus 60 °C for the same DCA pool. Due to lower sequencing depth as compared to assays of Fig. 2, a more stringent activity threshold of −1.5 was used (see “Methods” section). The Pearson correlation is 0.64, two-sided p-value = 10⁻¹⁶ (numerical precision). N = 558, comprising 484 active for both conditions (upper-right quadrant), 25 inactive for both (lower-left), 19 active at 60 °C but not 37 °C, 30 active at 37 °C but not 60 °C. d Active fraction as a function of the number of mutations for the DCA model at 37 °C (plain line) and 60 °C (dashed line). Sequences with a score below the activity threshold in at least one condition are in light color. Note that the fraction is computed over all designed sequences, including those that were not active enough to be detected by sequencing (thus counted as inactive). N per bin per model provided in the Source Data file. Dots are the active fraction. Error bars upper (lower) bound of vertical bars is the active fraction including (excluding) activity scores within the 98.5 percentile of the measurement error distribution around the threshold. Source data are provided as a Source Data file.