Virtual screening with accuracy comparable to experimental High-Throughput Screening (HTS) has long been a goal of computational chemists. Protein Family Virtual Screening (PFVS) is finally achieving this for kinases and several other protein families. The three PFVS methods achieve unprecedented speed and accuracy by including massive amounts of IC50 and structural data from previous targets into models for each new family member: the 2D �Profile-QSAR� meta-QSAR, the Kinase-Kernel chemogenomic model, and the 3D Surrogate AutoShim docking method. Between the methods, 2 billion activity predictions have been made for 4 million internal and commercial compounds across 500+ kinases, so initial kinase virtual screening is now a table lookup. The methods have been applied to over 4 dozen active Novartis projects, with external R2=0.35�0.7 and enrichments of 20x�60x. PFVS has recently been extended to non-kinase adenosine-binding proteins, GPCRs, and to serine and cysteine proteases, in all now covering half of the estimated druggable genome.The methodologies will be described and examples will be presented from all stages along the drug discovery pipeline, from finding tool compounds for early target validation to finding backup chemistries for successful projects going into the clinic.

Protein-family virtual screening for drug discovery: Accurate enzymatic and cellular activities and selectivities � without crystal structures, and with or without training data

246th National Meeting (2013)

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