Access to databases, patents, publications, webpages and other sources for biological activities of small molecules now allows the generation of very large sets of data with several billion knowledge triples. In our presentation, we will describe a new computational tool called EDEN that enables the conversion of very large data sets into computational models to predict the biological activity, DMPK and Tox related properties for any given molecule.<br/>Using either 2D structures of active compounds and/ or the 3D protein structures of the target as input, EDEN generates a complex affinity fingerprint. This is then used to predict the activity of existing compounds and to design new compounds de novo with the desired biological activity.<br/>We believe that EDEN delivers candidate compounds with very high accuracy and a low number of false positives because the computational model used to predict activity is more complete than other approaches. It uses both all “good bits” (required for binding) and all “bad bits” (detrimental to binding). This leads to very high enrichment factors. The program is very fast and can presently handle roughly 500 million compounds in one day on a standard notebook.<br/>We will explain how EDEN identifies and addresses key challenges with big data sets such as noisy datasets with many false positives or biased heterogenous datasets with most of the data belonging to one chemical series.<br/>Examples will be presented to illustrate some of the capabilities of EDEN.

Building fast, robust, and reliable prediction models using very large biological data sets

2019 Fall National Meeting

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