Reducing the Number of False Positive Hits

Hard filters are the most stringent in-silico filters and are used to shape the property profile of screening or combinatorial libraries and to prune hit lists from primary screening. They are derived from ID and 2D molecular properties (molecular weight, number of H-bond donors/acceptors, number of rotatable bonds, and so forth) and, as is described below, they are commonly used to reduce the number of false positive hits and to favor lead-like or drug-like chemotypes. 

Reactive, unstable compounds, as well as covalent binders, can be removed from screening collections by substructure searches [21, 22]. At Roche, a global team of experienced medicinal chemists has defined more than 100 functionalities which are reviewed at regular intervals. This list has been augmented by unwanted features (e.g., polyacids, alkyl aldehydes, polyhalogenated phenols, etc.) which are chemically unattractive starting points for a hit-to-lead optimization, because they often result in non-optimizable SAR patterns. These chemotypes have been coded into Markush-type substructures for automated detection and removal of unwanted compounds. However, we need to stress that these filters are fully customizable, and removed chemotypes can be restored if required. 

Another source of false positive hits is promiscuous binders, which were shown by McGovern et al. [23-26], via dynamic light scattering, to form aggregates above 

Based on this in-house dataset, an in-silico prediction model [27] (three-layered neural network, Ghose and Crippen [28,29] descriptors) was constructed to evaluate the frequent hitter potential before compound libraries are purchased or synthesized. This model was validated with a dataset of the above-mentioned promiscuous ligands published by McGovern et al. [26], in which 25 out of 31 compounds were correctly recognized. 

---