ALFA: Automatic Ligand Flexibility Assignment Javier Klett, † A ́ lvaro Corte ́ s-Cabrera, †,‡ Rube ́ n Gil-Redondo, †,§ Federico Gago, ‡ and Antonio Morreale* ,†,∥ † Unidad de Bioinforma ́ tica, Centro de Biología Molecular Severo Ochoa (CSIC-UAM), Campus de Cantoblanco UAM, E-28049 Madrid, Spain ‡ A ́ rea de Farmacología, Departamento de Ciencias Biome ́ dicas, Unidad Asociada de I+D+I al IQM-CSIC, Universidad de Alcala ́ , Alcala ́ de Henares, E-28871 Madrid, Spain § SmartLigs Bioinforma ́ tica S.L., Fundació n Parque Científico de Madrid, c/Faraday, 7. Campus de Cantoblanco UAM, E-28049 Madrid, Spain * S Supporting Information ABSTRACT: ALFA is a fast computational tool for the conformational analysis of small molecules that uses a custom-made iterative algorithm to provide a set of representative conformers in an attempt to reproduce the diversity of states in which small molecules can exist, either isolated in solution or bound to a target. The results shown in this work prove that ALFA is fast enough to be integrated into massive high-throughput virtual screening protocols with the aim of incorporating ligand flexibility and also that ALFA reproduces crystallographic X-ray structures of bound ligands with great accuracy. Furthermore, the application includes a graphical user interface that allows its use through the popular molecular graphics program PyMOL to make it accessible to nonexpert users. ALFA is distributed free of charge upon request from the authors. ■ INTRODUCTION Molecules are known to exist in a range of environmentally dependent conformations. The selection of a diverse set of representative conformers for a small molecule is a challenging task in modern theoretical approaches to drug discovery. Di fferent tools have been developed to overcome this problem 1−6 and they perform, in general, reasonably well according to the discussion by Ebejer et al. 6 insofar as they successfully provide ligand conformations that are closely similar to those found in experimentally determined target-bound complexes (almost exclusively from X-ray crystallography) for a high percentage of the molecules comprised in different validation sets. 7,8 Validating the performance of these tools can be done from a dual perspective: on one hand, by assessing the quality of the conformers in terms of how plausible they are from a structural and/or energetic point of view; on the other hand, by estimating whether or not the set of generated conformers is diverse enough to appropriately populate the phase space of the molecule under study. As the diversity of the set increases, the probability of finding structures more similar to the experimental ones also increases. Different strategies have been developed to achieve both goals, and they usually rely on energy and/or shape criteria. Energy evaluation has been used in many different ways as in ConFab, 3 which is based on a force-field energy function, or in more sophisticated methods 9 that rely on first-principles calculations. Similarity can be measured using shape descrip- tors 10 or other metrics such as the root-mean-squared deviation (RMSD), Tanimoto combo scores, 10 etc. Having a similarity measure ready at hand, the next step is to classify the solutions and reduce the set by selecting only those conformers that add diversity to the ensemble. This is done by means of clustering algorithms such as k-means, hierarchical clustering, etc. For these methods to work, the all vs all distance matrices need to be precalculated and stored. This turns out to be a problem when dealing with very flexible molecules because the number of possible conformations grows exponentially with the number of rotatable bonds and so do the computational requirements. There are di fferent approaches to select conformations on the basis of results from clustering algorithms. Some of them apply clustering as a mere refining tool or to classify the solutions while they are being produced 3 (i.e., on the fly); whereas, others rely, for example, on different energy criteria or Monte Carlo methods. 10 It is well-known that when force field-based potential energies are used as a criterion to select conformers, a certain bias exists toward those conformations favored by the parameters implemented in the force field and certain areas of the phase space can be left unexplored. This limitation is not so crucial if one considers only molecules in solution but may be of importance when the molecule is bound to its macromolecular target, as the binding site environment is thought to be ultimately responsible for the real pose and may select a conformation that does not exactly correspond to a minimum on the potential energy surface. Received: July 30, 2013 Published: January 6, 2014 Article pubs.acs.org/jcim © 2014 American Chemical Society 314 dx.doi.org/10.1021/ci400453n | J. Chem. Inf. Model. 2014, 54, 314−323