REMUS – Alignment-Driven Virtual Screening with Adaptive ScoringJoel Graef1, Florian Flachsenberg1, Johannes Kirchmair1, Matthias Rarey1 |
|
| 1Center of Bioinformatics, University of Hamburg | |
| In the absence of a target protein structure, virtual screening based on 3D molecular alignment is usually the method of choice [1]. The active ligand is aligned against every compound from a large collection and the result assessed with respect to common pharmacophore properties and shape. We present REMUS (Rigid flExible MolecUlar Superimposer), a new tool for the efficient calculation of ligand 3D similarity. A special feature of REMUS is a range of combinable Gaussian functions, some of which are coordinate centered rather than atom-centered. This enables a range of new possibilities, for example the representation of ring systems as single coordinate points, describing the ring normal vectors or the target points of hydrogen bond interactions. REMUS applies a step-limited variant of the BFGS optimization algorithm. The starting positions are generated by randomly changing the translation and rotation of one of the molecules in a bounding box. Molecular flexibility is addressed by a two-phase approach. The alignment procedure is started with a conformation ensemble calculated with Conformator [2]. Because these conformers are not always leading to the best alignment, a further optimization step is required. For this purpose, the torsion angles will be adapted by using a torsion angle scoring function. This function calculates continuous torsion angle potentials based on torsion angle peaks recorded in a public torsion angle library [3]. It uses the von Mises function [4] as the kernel for curve approximation and estimates the curve width by connecting the second peak tolerance and the peak score from the torsion library with the measure of the von Mises function. A clash term scoring policy was implemented to avoid intramolecular clashes. Alongside these features, several scoring policies were developed to analyze which steric and chemical properties are best suited to calculate molecular alignments. This process resulted in four policies: The AADR policy mainly uses the chemical elements of all atoms of both ligands. The Hydro policy uses the molecular shape, hydrophobicity and hydrogen acceptors/donors. The Gauss policy uses only atom-centered properties like the Van-der-Waals radii, the partial charge state, whether the atom is polar, in a ring, hydrophobic or which functional group it belongs to. The last policy, PenRew, does not only reward the alignment of the same properties but also penalizes certain alignments of dissimilar properties. All scoring policy terms were optimized on a dataset by changing the weight of each property in a predefined range. The policies were evaluated on two datasets in terms of alignment accuracy. Regarding the screening performance, REMUS performs on par with other comparable methods. To enable users with little or no technical knowledge, REMUS is equipped with a graphical interface. Besides choosing and testing scoring policies, conformers and alignments within an interactive 3D visualization, Remus features a manual superimposition mode. Atom pairs between two ligands can be picked followed by an RMSD minimization using the Kabsch algorithm [5, 6] and real time visualization.[1] Ashutosh Kumar and Kam Y. J. Zhang. Advances in the Development of Shape Similarity Methods and Their Application in Drug Discovery. Frontiers in Chemistry, 6:315, 2018. [2] Nils-Ole Friedrich, Florian Flachsenberg, Agnes Meyder, Kai Sommer, Johannes Kirchmair, Matthias Rarey. Conformator: A Novel Method for the Generation of Conformer Ensembles. Journal of Chemical Information and Modeling, accepted for publication. [3] Wolfgang Guba, Agnes Meyder, Matthias Rarey, and Jérôme Hert. Torsion Library Reloaded: A New Version of Expert-Derived SMARTS Rules for Assessing Conformations of Small Molecules. Journal of Chemical Information and Modeling, 56(1):1–5, 2016. [4] Evans, M., Hastings, N., and Peacock, B., “von Mises Distribution”. Ch. 41 in Statistical Distributions, 3rd ed. New York, Wiley 2000. [5] W. Kabsch. A solution for the best rotation to relate two sets of vectors. Acta Crystallographica Section A, 32(5):922–923, 1976. [6] W. Kabsch. A discussion of the solution for the best rotation to relate two sets of vectors. Acta Crystallographica Section A, 34(5):827–828, 1978. |
|
