Fig. 3: CodonTransformer generates natural-like codon distributions.
From: CodonTransformer: a multispecies codon optimizer using context-aware neural networks
a Schematic representation of %MinMax and dynamic time warping (DTW). %Minmax represents the proportion of common and rare codons in a sliding window of 18 codons. DTW algorithm computes the minimal distance between two %MinMax profiles by finding the matching positions (“Methods”). b %MinMax profiles for sequences generated by different models for genes yahG (E. coli), SER33 (S. cerevisiae), AT4G12540 (A. thaliana), Csad (M. musculus), ZBTB7C (H. sapiens). c DTW distances between %MinMax profiles of model-generated sequences and their genomic counterparts for 50 random genes selected among the top 10% codon similarity index (CSI). For each organism, the gene for which the %MinMax profiles are represented above (b) is highlighted in gray. Mean DTW distances were compared to the fine-tuned model using a two-sided unpaired t-test (n = 52), with the numerical p-value shown for each. Center line shows the median; box limits represent the 25th (Q1) and 75th (Q3) percentiles; whiskers extend to 1.5× interquartile range (IQR); points are outliers beyond whiskers. d Mean and standard deviation of normalized DTW distances by sequence length between sequences for the 5 organisms (for organism-specific DTW distances, see Supplementary Figs. 18). Data underlying this figure is provided in the Source Data File.
