We first examined all in silico evolved CRMs for all 28 target patterns and noted that CRMs associated with similar expression patterns are closer to each other than distinctly expressed CRMs (supplementary fig. S2, Supplementary Material online), as expected.
Among the three strategies considered, we demonstrate that the Lyapunov Guidance Vector Field strategy has the best performance for all target motion patterns.
This architecture is reconfigurable for any target pattern and any error tolerance.
> -wrap-foot> We next asked if in silico evolved CRMs for a target pattern are similar to the D. melanogaster CRM (henceforth, "real" CRM) associated with that pattern.
We first tested for a correlation between time-to-evolve for a target pattern and each TF's binding site count (or estimated occupancy) in the real CRM associated with that pattern.
For the target patterns 'one point' and 'half', using 4 regular elements shows good performance, while for the target pattern 'xor', using 6 singular elements yields the best results.
As shown in figure 4, each of these TFs reduces the median time-to-evolve for several target patterns, with the effect of ZLD being clearly more prominent.
If the initial random sequences have a higher fitness value for certain target patterns, perhaps due to a greater frequency of random occurrence of certain binding sites necessary for that pattern, then such patterns may be quicker to evolve.
We interpret these observations to suggest that the number of HB sites in the initial random sequences influences the fitness of those sequences for certain target patterns, and therefore their time-to-evolve estimates.
In particular, reducing 2 Nµ from 0.01 to 0.001 (a factor of 10) results in average time-to-evolve increasing about 7-fold from approximately 2.1 to approximately 18 Myr, with estimates for individual target patterns ranging between 2.1 and 25 Myr.
