Figure 2. Characteristics of LIN28B binding on non-miRNAs.

(A) Changes in read density at any binding site can be plotted as a function of LIN28B expression level, yielding a regression and slope. The slope is correlated to strength of LIN28B binding.

(B) Distribution of slopes (μ), encompassing 11,433 target sites.

(C) Frequency of known LIN28 motifs, ‘‘UGAU,’’ in eCLIP-identified binding sites.

(D) Frequency of known LIN28 motifs, ‘‘GNNG,’’ in CLIP-identified LIN28 binding sites.

(E) Boxplots showing the distribution of GC, GU, AU, and GG pairs found in RNA structures formed in LIN28 binding sites.

(F) Cumulative distribution functions showing the minimum free energy of the ensemble for all higher-affinity binding sites (m < 0, solid red), the 100 highest-affinity binding sites (dashed red), all lower-affinity sites (m > 0, solid green), and the 100 lowest affinity binding sites (dashed green) compared with randomly chosen sequences outside of eCLIP-identified binding sites (gray).

(G) Receiver-operator characteristic for two random forest classifiers that were developed to predict the affinity of a LIN28 binding site (blue) or the location of a LIN28 binding site (yellow). Predictions against data with shuffled identifiers were used as a control (dashed).

(H) Boxplots showing the accuracy of random forest classifier predictions compared with performance on a dataset with shuffled identifiers.