Fig. 4: Chemical diversity of predicted binders, selected from Broad Compound Collection (Broad CC) for experimental validation, and confirmed binders in biophysical assay in a dose-dependent manner.
From: Evaluation of DNA encoded library and machine learning model combinations for hit discovery
Each panel shows the output of t-distributed stochastic neighbor embedding (t-SNE) analysis for the blind assessment set (Broad CC) used to discover hits, with predicted binders selected for experimental validation in (a) and binders confirmed in biophysical assay in (b) highlighted in colors. The plots are separately colored by the DELs the ML models are trained on (left) and the ML models (right) predicted the compound as a binder.
