Fig. 5.

Distributions of amino acid types across predicted, training, and natural datasets. (a) The relative frequency of each amino acid predicted with a confidence of 80–100%. (b) The relative frequency of each amino acid type in the CNN training data. The correlation between predicted and training frequencies was 0.658 ($p=0.002$) (c) The distribution of highly conserved amino acids. These amino acids are found at frequencies $>80\%$ at individual positions in the MSA. The correlation between predicted and conserved amino acid frequencies was 0.826 ($p=0.002$)