AutoDock Versions

Current Release: AutoDock 3.0

AutoDock 3.0 has a new genetic algorithm (GA) search engine, amongst several other new features, notably an empirical free energy function for estimating binding free energies and inhibition constants, K_i. It consists of both C and C++ source code, and several binaries for common architectures. The GA can be further enhanced by combination with a local search (LS) engine. We call this hybrid search Lamarckian GA (LGA), since offspring are allowed to inherit 'environmental adaptations'. We have recently shown (see References) that the adaptive LGA method can handle problems with higher dimensionality more efficiently than simulated annealing, and is more efficient and more reliable than pure GA alone.

Much Faster Docking
Recent improvements in the code and schedule parameters mean that we can now perform 10 LGA dockings in about 15 minutes, (using a Silicon Graphics 250MHz Octane). Before, they used to take around 7 hours (using a Silicon Graphics 250MHz Reality Engine). The success rate is comparable: 9 out of 10 dockings match the crystal structure, in a test with XK-263, a cyclic urea HIV-1 protease inhibitor docking to the HIV-1 protease in 1HVR. This opens up the door to the possibility of screening libraries.
Adaptive Search Technques
The new optimization methods come out of an on-going collaboration with Professor Rik Belew from the Department of Computer Science at UCSD. The hybrid GA is based on the doctoral work of William E. Hart, on ``Adaptive Global Optimization with Local Search''.
Incorporation of desolvation free energy
This has been achieved using atomic solvation parameters using the method of Stouten et al., and is now built into AutoGrid 3.0.
Empirical free energy function
We have carefully investigated a variety of possible terms for predicting free energy of binding, using S-PLUS, a statistical analysis package. The model was callibrated on a set of diverse, structurally known ligand-protein complexes. The best model, out of a total of 900, was able to predict binding free energies to within 2 kcal/mol. This is sufficient to distinguish ligands with Ki's that are millimolar, micromolar or nanomolar: particularly important in drug design. A graph showing the correlation between observed and predicted free energies of binding is shown below. The Brookhaven PDB accession codes, in the form of three letters followed by the number, are shown also.

Previous Versions

AutoDock 2.4 was released until 5 March, 1999. It consists of a suite of three C programs:

AutoTors, which facilitates the input of ligand coordinates and definition of rotatable bonds,
AutoGrid, which calculates a three-dimensional grid of interaction energy based on macromolecular coordinates, and
AutoDock, which performs the docking simulation.

These programs are accompanied by several shell scripts, awk and C programs, to help create input files, to launch AutoDock simulations in parallel on a cluster of machines, and to help in the conformational cluster analysis of the docked structures.

The movies described here were created from work done with AutoDock 2.4. This work has been published in the Journal of Computer-Aided Molecular Design, see the References page. You will also find several publications from some of the groups that have obtained AutoDock, showing diverse applications of AutoDock.

AutoDock 2.2 was released until 21 October, 1996. It is still available if you need it, but version 2.4 has more features and is more robust. We recommend obtaining 2.4.

Coming Soon...

AutoDock 4.0 - with side chain flexibility...

AutoDock

Versions