Extended Data Fig. 6: Features and predictions of repressive post-transcriptional disordered regions.
From: Deciphering disordered regions controlling mRNA decay in high-throughput
a, Correlation between number of negative changes and average activity scores (n = 2). b, Correlation between number of aromatic residues and average activity scores (n = 2). c, Prediction probabilities of the motif-dependent classifier to predict the composition-driven repressors (red) and the composition-driven classifier to predict motif-dependent repressors (blue). d, Correlation between activity score and Kyte-Doolittle hydrophobicity (n = 2). e, Coefficients for dipeptides in the logistic regression classifier for composition-driven repressors. The model includes both single-amino acid and dipeptides as variables. f, Correlation between average activity scores for composition-driven repressors and the spacing of aromatic amino acids as defined by the parameter Ωaro (n = 2)33. Activity score is shown for each wildtype and stable scrambled composition-driven peptide, determined from the mutational scanning screen. g-i, Light attention model predictions for three composition-driven repressors using their ESM-1b representations. Peptide examined is displayed in panel legend. j, Correlation between predicted and measured activity scores for Mrn1(61–110) variants generated from the light attention model (n = 3). k, Correlation between predicted score in the final linear layer of our light attention model and number of di-aromatic residues (bottom) with density shown (top). The parent peptide used for 10,000 random permutations is indicated above. l, Correlation between dipeptide content and measured activity of composition-driven repressors and their scrambles. The Pearson correlation coefficient is displayed in panels when appropriate.
