Extended Data Fig. 2: State space optimization, design of perineurial glia enhancers and modification of genomic sequences toward KC enhancers.
a, Prediction score distribution for 3 million sequences generated by selecting the top 20 best mutations for 5 incremental mutational steps. Blue line, score of EFS-4 from the greedy algorithm. b, Zoomed-in version of panel a to the sequences that have higher prediction score than 0.25. c, Prediction score of evolved sequences by greedy algorithm (EFS-4) vs the best of 3 million sequences on each mutational step. d, Nucleotide contribution score of the original and evolved sequences as well as delta prediction score of in silico saturation mutagenesis for EFS-4 (top) and the top scoring sequence (bottom) e, Prediction scores of 6 selected PNG sequences at each mutational step for PNG model (left) and KC model (right). The selected iteration (15th mutation) is indicated with a dashed line. f, In vivo enhancer activity of the cloned PNG sequences with no enhancer activity. g, Comparison between γ-KC prediction score and mean γ-KC accessibility for the binned fly genome regions. The selected regions with high prediction and low accessibility are highlighted with blue, orange, green and red dots. h, γ-KC ATAC-seq profile of the four selected regions. The exact ___location of the regions is indicated with dashed lines. i, Prediction scores of 4 selected KC near-enhancer sequences at each mutational step for KC model. The selected iteration (6th mutation) is indicated with a dashed line. After the 6th mutation, 4 more mutations are performed in FP3 to improve prediction score while 7 or 8 mutations are performed in the three other sequences to generate repressor sites. j, In vivo enhancer activity of the cloned WT genomic “near-enhancer” sequences with no enhancer activity. The expected ___location of KC is shown with dashed circles. Scale bars, 100 µm.
