iSmoog : an automated and rational algorithm to find new protein inhibitors Nicolas Chéron,1 Naveen Jasty,1 Eugene I. Shakhnovich1 1Department of Chemistry and Chemical Biology, Harvard University Contact:
[email protected];
[email protected] Context: All the commercially available molecules will soon have been tested (experimentally or computationally) against all the known target proteins. Moreover, there are only ≈25 new molecular entities approved every year by the FDA (on average). Thus, we must develop new methods to find drugs.
Growing molecules
Goal
• Molecules are grown in the active site by connecting fragments (≈SMoG: JACS 1996, 118, 11733-11744). • For each new fragment, a rotameric search is performed: the one with the lowest energy (and no clashes) is kept. • Fragments are accepted or rejected with a Metropolis criterion. • The growth is stopped when a given treshold is reached (molecular weight, size, number of fragments).
• Develop an open-source, easy-to-use program which can find de novo molecules. • Ligands must be easy to synthesize, stable, soluble in water, non-toxic, with good ADME properties. • Protein flexibility must be accounted for. • Accurate binding energy prediction. iSmoog = inside Small molecule optimized growth
Connecting fragments
# Freq.
Fragments cannot be randomly chosen (otherwise, they would be unstable, toxic or non-soluble): 1. We constructed a drug database. 2. It was analyzed, and information were extracted from it. For example, what is the probability that a phenyl group is connected to an aldehyde? 3. New fragments are chosen according to these probabilities. ≈ FOG: J. Chem. Inf. Mod. 2009, 49, 1630-1642
• •
•
9%
2 18% 3 27% 1
9%
1
9%
2 18% 1
9%
Computing the energy
Searching for 3mer
J. Med. Chem. 2002, 45, 2770-2780
Increases the synthetic accessibility and the “druglikeness” of new ligands.
Accounting for protein flexibility •
1
Ligands are grown simultaneously in several conformations of the protein. The program can first use rotamers, and then conformers (e.g. snapshots from MD). Energy is Boltzmann-averaged: −𝛽𝑆 𝑆∗𝑒 𝐸= , 𝑆 = 𝑆𝑐𝑜𝑟𝑒 −𝛽𝑆 𝑒 Can be used with different protein mutants.
Choosing fragments (GUI)
Application: HIV-1 Protease • • •
DGBinding of 8 FDA-approved HIV-1 protease inhibitors
5000 molecules were regularly grown. 3000 molecules were grown in Restart mode. For both cases, the top 10 were kept.
In the Restart mode, the user defines a starting fragment and which atoms can grow new ligands. This allows to find variations around a known core/anchor.
More accurate calculations of DGBinding were made on the best candidates. The best two are shown below:
Regular41_18
Restart31_40
Roadmap • • • • •
Development of a new scoring function Covalent ligand Toxicity predictions Synthetic accessibility score Implementation on GPU
The authors thank DARPA (Contract HR001111-C-0093) for funding.
