You are here

Rui Costa defends master dissertation entitled "Scalar algorithms for molecular docking in heterogeneous platforms"

Rui Sérgio Magalhães da Costa has requested the defence of his dissertation, entitled "Scalar algorithms for molecular docking in heterogeneous platforms". The defence will be held in room A2 of the Departamento de Informática, on December 15, 2011, 03:00pm. The public is invited.

The examining committee is the following:

  • João M. Fernandes (UMinho) - president
  • João M. P. Cardoso (FEUP) - opponent
  • Alberto Proença (UMinho) - supervisor
  • Nuno Micaêlo (UMinho) - supervisor


The high throughput screening of new candidate drugs uses computational intensive molecular docking simulations. State-of-the art implementations for multicore-CPU systems still have performance, precision and accuracy limitations, which require an increase in the efficiency and scalability of both molecular docking algorithms and their coding. Current heterogenous platforms that merge multicore CPU with CUDA enabled GPU devices are an affordable accelerating technology that may overcome the performance limitations.

The dissertation work aims to efficiently port to an heterogeneous platform a popular open source software package for molecular docking, Autodock Vina, keeping the functionality of the original algorithms whenever feasible. The new version, ScalaVina, predicts the non-covalent chemical interaction of a small molecule (ligand) against the binding site of a receptor macromolecule (receptor), evaluating the fitness of the ligand inside the binding pocket using a scoring function, and searching the best structural fit between ligand and receptor with the lowest local minima binding energy.

The original Vina supports multithreaded parallelism at a high level, by launching multiple starting points with sequential activities, which include the global optimizer heuristic algorithm iterated local search, ILS, the energy minimization function using the BFGS algorithm, the multivariable scoring function and other support functions. The development of the ScalaVina version to take advantage of a CUDA enabled GPU required the parallelization of these functions under the SIMD paradigm, while attempting to minimize data transfers to/from the accelerating board and to improve data access patterns. Each starting point was mapped into one streaming multiprocessor at the GPU device, exploring the SIMD multithreaded parallelism to compute the values required for the docking operations above mentioned. This approach mimics the original Vina when it allocates a hardware block (SM) in the GPU for each starting point; however, these functions are executed in parallel in the GPU.

Obtained results show that the GPU output of ScalaVina is in qualitative and quantitative agreement with experimental data and closely matches the output of Vina for the same set of receptor:ligand pairs. Preliminary performance results also show that there is room for improvements with careful tuning.


home contacts RSS Feed last update: 16-Apr-2020 share facebook
Drupal theme by D7 ver.1.1