Hebrew University of Jerusale

Prof. Amiram Goldblum's

School of Pharmacy
Molecular Modeling and Drug Design Group


Iterative Stochastic Elimination

Our group developed in the last 10 years a general optimization algorithm*, Iterative Stochastic Elimination (ISE)**, for finding a very large set of optimal solutions to highly complex combinatorial problems. It finds global and local minima and produces large sets of best solutions ("ensembles"). The algorithm has been mainly applied to problems in structural biology and bioinformatics, such as solving proton positions in crystal structures, comparative modeling of side chains, loops and multiple loops in proteins, molecular conformational ensembles and the associated Boltzmann-weighted Molecular properties, flexible ligand docking ("ISE-dock") and protein-protein interactions, protein design and chemoinformatics problems of ligand design and discovery, optimization of complex QSAR equations, variable selection, focused molecular libraries and related issues.

The latest applications of ISE in cheminformatics are:
  1. Producing "focused libraries" of molecules with the highest chance to be active on specific targets ("Molecular Bioactivity Index" or MBI)
  2. "Drug Like Index" (DLI) for each molecule by optimizing the distinction between drugs and non-drugs on the basis of their properties ("descriptors")
  3. Prioritization of scaffolds for general drug likeness or for specific targets
  4. Combinations of the above with each other or with docking ("double focusing")
  5. Indexes for selectivity, Differential Toxicity, h-Erg toxicity, solubility, BBB transport and more
  6. Polypharmacology, or multitargeted drugs: single molecules that are active on two or more targets

*The algorithm won the first prize in the Emerging Technologies Symposium of the Computers in Chemistry Division, The American Chemical Society, Washington D.C. August 2000.

**ISE may be adapted to any highly complex problem which can be expressed in terms of variables and discrete variable values. Another requirement is to have (or to formulate) a precise scoring function for evaluating each and any configuration of the studied system. Finally, variables should “interact” so that the optimal solution is not simply the combination of the best scores of each separate variable and may thus be found by a “self consistent” search.


1. A stochastoc method to determine, in Silico, the drug like character of molecules (Rayan & Goldblum) www.wipo.org/pctdb/en/wo.jsp?wo=2005022111

2. System and Method for searching a combinatorial space (Glick & Goldblum) www.wipo.org/pctdb/en/wo.jsp?wo=2001039098

Last Update : January 19, 2011 | Copyright ©, 2009, The Hebrew University of Jerusalem. All Rights Reserved.