Fluorescence dying techniques are widely used in quantitative high-throughput screening (qHTS) as binding or inhibition markers. However, there are some fluorescence-emitting, or autofluorescence compounds producing false fluorescence signals in the qHTS. To avoid the cost on screening compounds with false signals, an in silico autofluorescence prediction model is an efficient tool to remove compounds with autofluorescence potency before the screening process. We used six autofluorescence assays in PubChem Assay Database to train and validate our tree-based models. Most descriptors in the tree were about hetero aromatic ring counts, which are related to the previous knowledge about molecular autofluorescence. There were also some long-chained alkane fragments and cyanide fragments descriptors selected in our model. In addition, consensus modeling methods and applying previous knowledge related filters were helped improving prediction models of highly impaired data.

Classification tree and random forest based prediction models on molecular autofluorescence

245th National Meeting (2013)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.