Poster Abstract - Sergey Sosnin & Gerhard Ecker
Analysis of the applicability domain and a model uncertainty plays a crucial role in reliable QSAR modeling. We present a method and tool for visually analyzing a model’s uncertainty on the chemical space of the model.
Our tool utilizes the parametric t-SNE model, which was pretrained on the entire ChEMBL database, to visualize large databases of organic compounds based on compounds’ structural similarity. We demonstrate that this method can have several promising applications: investigation of the applicability domains of QSAR models, search for activity cliffs, and outliers detection.
