Diapositivo 1

A-MOP
UTAP-EXPL/QEQ-COM/0019/2014
Algorithms for Macro-Molecular
Pocket Detection
Joaquim Jorge
INESC-ID
A-MOP
Emerging Technologies
Exploratory project
12 months
Institutional Partners
Project Members
Chandrajit Bajaj
Joaquim Jorge
Daniel Simões Lopes
João Madeiras Pereira
Abel Gomes
Tiago Simões
Motivation
Structure-Based Drug Design Challenges:
How to correctly predict which small molecules can bind to a specific protein?
How to assess their impact on protein function?
Major Issues:
size of proteins that current approaches can handle
time required to find cavities and rendering
Desiderata:
More efficient algorithms for detecting pockets on the surface of large proteins
(> 500K atoms)
Goals
(A) to develop new and efficient geometric algorithms to determine pockets and
other cavities in macromolecules
(B) to develop computational methods to tackle the problem of scalability with
number of atoms (e.g. millions of atoms)
Goal (A)
Geometry-based pocket prediction method:
explore critical points to quickly find pockets on the protein surface, for mesh- or
meshless-based methods
no need to explicitly evaluate the whole surface of the molecule
Gomes 2014, Comp Graph 38, 365–373
Goal (B)
GPU Parallelization:
Proposed methods are both localized and decoupled it is seemingly possible to
take advantage of parallel computation using multiple CPU/GPU cores
Dias & Gomes 2007
Plan and Methods
Hypothesis: by resorting on local molecular information, it is possible to develop
more efficient techniques to find and classify pocket sites of large proteins
Research Plan:
A - Geometric modeling of molecular entities
B - Location of pocket sites
C - Classification of pocket sites
D - Visualization of pocket sites / Molecular Visualization
E - Validation of computational procedures
F - Algorithm benchmarking
Geometric modeling of molecular entities
explore implicit surface representations
rely on the critical point theory that uniquely associates critical points of an implicit
smooth surface to molecular biology meaning:
a pocket must be a critical point of the implicit function
Use Gaussian functions to represent molecular surface
models electron density
is always smooth)
Locating pocket sites
For atoms on the periphery of the molecule find mimimization/maximation paths
(of the partial derivatives)
find 2-saddle points in the domain of the Gaussian function without spatial
enumeration or voxelization of the domain.
2D algorithm: ACM Trans Comp Graph 38, 2014
Triangulating molecular surfaces over a LAN of
GPU-enabled computers, Parallel Computing
2015
Multi-GPU-based detection of protein cavities
using critical points, SE Dias, QT Nguyen, JA
Jorge , AP Gomes, Future Generation Computer
Systems, 2016
Output Indicators
International journal Papers
5
International Conference papers
1PhD Thesis
1
1 MSc Thesis
Organization of seminars and conferences
1
Ongoing / Future Work
Novel protein docking methods
Database databases of protein-ligand binding pockets
Benchmark of pocket-finding Algorithms (CavBench)
Validation and benchmarking (CavBench)
accuracy of the pocket site predictions to be validated and calibrated with specific
examples of complexes found in the BindingDB database
compare methods in terms of efficiency, speedup and accuracy with state of the
art methods
Acknowledgements