Iterative Stochastic Elimination Algorithm for drug discovery by targeting and multitargeting

Bio: Prof. (Emeritus) Goldblum is head of Molecular Modeling and Drug Discovery at the Institute for Drug Research, School of Pharmacy. Goldblum studied Chemistry and Physics and received his PhD in Organic Chemistry at the Hebrew University, followed by Postdoctoral studies of Quantum Biochemistry at CNRS Paris, QSAR at Pomona College California and Computational Chemistry at Stanford. Prof. Goldblum joined the Medicinal Chemistry department of the Hebrew University in 1979 and developed and applies since then theoretical approaches and algorithms for drug discovery. In 2000, Goldblum won the first award of the ACS Computers in Chemistry division in a contest of “Emerging technologies” for his novel algorithm to solve extremely complex combinatorial problems.

Abstract: Our heuristic algorithm, Iterative Stochastic Elimination (ISE), produces models which are large sets of excellent solutions to highly complex combinatorial problems with very many variables that have, each, many values. This generic algorithm has already been applied to many problems in protein and peptide conformations, to ligand docking, and has recently been applied mainly for molecular discovery in medicinal chemistry related projects.

A model of molecular activity is normally constructed of many filters of physico-chemical property ranges, and is useful for virtual screening of huge molecular libraries from commercial sources or from virtual combinatorial libraries. Each screened molecule is scored as it passes (or fails to pass) filters successfully. Top scored molecules compose our “discovery set”, usually a few dozens and sometimes more, depending on our experimental collaborators. Our models are highly enriched and result in the discovery of novel and diverse molecular scaffolds, amenable to apply for Intellectual Property for all types of bioactivity for more than 95% of the active molecules discovered.

It is easy to combine models for different targets as one can combine filters. Multitargeted molecules are those that get higher scores in more than a single model. We focus also on modeling of anti-targets for our screenings. In addition to ligand based modeling by ISE, which is our main method, structure based modeling (docking) is used if crystal structures are available. Although data collection and curation constitute a bottleneck of the process, model building with ISE is a matter of a few hours, and virtual screening of huge databases requires less than a few days.