RMSD-BASED CLUSTERING AND QR FACTORIZATION. Rob Swift, UCI Tuesday, August 2, 2011

Transcription

1 RMSD-BASED CLUSTERING AND QR FACTORIZATION Rob Swift, UCI Tuesday, August 2, 2011

2 Outline Biomolecular flexibility, a few examples CADD pipeline overview RMSD clustering QR factorization

3 Our evolving ideas of molecular recognition Rigid lock-and-key (~1900) Induced fit (~ 1958) Lock (protein) Receptor exists in a population of energetically lowlying sub-states, or a conformational ensemble Key (ligand) Conformational ensembles (~ 2000) E Ligand binds to one of these sub-states and shifts the population toward the favorable bound conformation

4 Conformational ensembles and population shift Swift R.V., McCammon, JACS, 2009, 131:

5 HIV Integrase pocket expansion crystal like pose Schames, & McCammon, J. Med. Chem. (2004) New pocket

6 HIV Integrase Butterfly compounds Schames, & McCammon, J. Med. Chem. (2004) Open conformation Closed conformation Crystal conformation

7 CADD pipeline workflow

8 Processing large ensembles Long simulations leads to lots of structures (>10,000) Many of these are redundant or very similar Utilizing all of these structures is computationally intractable RMSD-based clustering QR-factorization

9 RMSD-based clustering Algorithm 1. Least-squares alignment to a common reference frame 2. RMSD values are calculated between all unique conformer pairs 3. Each conformation is assigned neighbors Number of neighbors equals the number of conformations whose RMSD is less than a user-defined cutoff. 4. Clusters with decreasing conformer membership are generated Result A set of clusters is generated Each cluster has fewer members than the last Each cluster has a central member, or centroid

10 Choosing a cutoff Calculate the number of clusters at increasing cutoff values Reducing the computational effort is the goal There is no one right answer Look at the binding site, use your chemical common sense.

11 Clustering distills essential information Extracted snapshots from 4 chains explicit 40 ns simulations (160 ns for both apo & holo) Alignment based on Cα atoms Computed RMSD distance matrix using subset of 62 residues (sidechains included) lining the binding pocket

12 RMSD clustering using the CADD pipeline

13 Reducing the conformational ensemble with QR-factorization Description QR-factorization orders a set of conformations in increasing order of linear dependence An RMSD cutoff determines the number of clusters in the representative ensemble. Choosing a cutoff Selected in the same way a RMSD clustering cutoff is selected. Amaro, Schnaufer, Interthal, Hol, Stuart, McCammon., PNAS (2008) *O Donoghue & Luthey-Schulten, MMBR (2003) Conformational dendogram

14 QR-factorization using the CADD pipeline

15 QR-factorization and RMSD clustering are different RMSD clustering provides population information Estimate free energy differences between clusters $!G( 1" 2) = #RT ln f ' 2 & ) % ( Weighted average binding free energy estimates n clusters "!G bind = f i!g i i=1 QR factorization does not Unweighted average binding free energy estimates n clusters "!G bind =!G i i=1 f 1

16 Conclusions Biomolecules are flexible MD simulations typically generate too many conformations to tractable analyze during a virtual screen RMSD clustering and QR-factorization are reliable ways to reduce the conformational ensemble RMSD clustering gives population information, QR factorization does not

17 New class of HIV Drugs: Merck & Co. Discovery of unexpected binding site in HIV-1 Integrase using MD and AutoDock: Schames, & McCammon, J. Med. Chem. (released on web, early 2004) Exploration of the structural basis for this unexpected result suggests an approach to the development of integrase inhibitors with unique resistance profiles. D. Hazuda et al., Proc. Natl. Acad. Sci. USA (Aug. 2004), refers to Schames, et al. (2004). L MK-0518 February, 2006 Phase III Clinical Trials February, 2007 Name announced: Isentress (raltegravir) October, 2007 FDA fast track approval