Computational Technology

Locus' discovery research programs are driven by fragment-based molecule design technology, which, in contrast to other fragment-based approaches, is purely computational. The in silico approach is not limited by physical restraints such as solubility and minimum binding potency so that the only limit on fragments is the imagination. The result is a very strong potential for novelty, which in turn, opens the door to efficient development of new lead optimization hypotheses. The accuracy of the computational methods permits confidently taking fragment-derived molecules into a lead optimization program, and the speed of the platform makes this practical.

Locus Core Technologies consist of three core building blocks:

• An in-silico fragment-based approach to discovery and lead optimization
• Rapid calculation of fragment binding free energy
• Methods of exploring protein mobility

An in-silico fragment-based approach to discovery and lead optimization

Locus has built an internal core library of several thousand fragments with a much larger expanded set available. Use of these fragments has been carefully engineered into Locus' in silico platform and provides a rapid building block approach to novel chemical space and intellectual property.

The core library was selected from the literature of biologically active molecules and carefully culled to ensure a breadth of physical properties, maximum synthetic feasibility, and minimum potential for toxicity. Additional fragments can be added to the library ad hoc based on the specific geometries of target binding sites and chemistry considerations.

Fragment binding simulations naturally identify all viable and sometimes novel binding pockets, without the need of known ligands. Figure 1, for example, shows the results of a Locus fragment analysis applied to a binding site that had been characterized as highly hydrophobic based on known SAR. Fragment analysis identified a hydrophilic fragment with high affinity for the binding site and, in addition, two binding poses (magenta and yellow) that differed fundamentally from the previously observed crystallographic pose (green).

Rapid calculation of fragment binding free energy in solution

The accuracy of Locus' binding affinity calculations makes in silico screening a viable and preferred alternative to experimental fragment crystallization. For that reason, Locus experimental crystallography efforts emphasize whole molecule validation of computational technology predictions rather than on finding fragment leads.

Locus has pioneered the use of Grand Canonical Ensemble thermodynamics to rapidly and accurately compute the free energy of binding of ligands to proteins.¹ While the initial algorithms were a conceptual advance, they were performed for the gas phase and required days of cpu time per fragment. By rigorously combining concepts of thermodynamics and principles of software engineering, these algorithms have been optimized to compute fragment binding affinities in a matter of cpu-hours. For the first time, it is feasible to simulate fragment binding in aqueous solution at the resolution of individual water molecules.

Simulation of fragment binding in the presence of water clearly identifies and quantifies all of the expected effects of solvation and is essential for reliable prediction of binding free energies (Figure 2):

• Affinity for water can exceed the affinity for organic fragments, eliminating some binding pockets from productive ligand design. (red circle).
• Ligands can be designed to make favorable interactions with tightly bound water (yellow circle).
• Maximum ligand binding efficiency is achieved in poorly-solvated regions, and these regions dominate most of the volume of the binding pocket (green circle).

Methods of exploring protein mobility

While protein crystal structures are at the heart of today's structure-based drug discovery efforts, protein motion is one of the most difficult challenges that limit success. For this reason, Locus has developed proprietary methods to explore large ranges of coupled motions over very long time scales.

The methods apply principles of mechanical engineering to simulate the motions of molecular bonds using internal coordinates. The use of internal coordinates greatly reduces the system complexity and provides more meaningful answers than standard dynamics and normal-modes calculations.² These methods allow Locus to identify novel pockets that open as a protein flexes and to avoid highly flexible regions where ligand binding affinity will be negated by protein entropy.

By integrating binding energy calculations with the ability to rapidly explore multiple protein states, we can use SAR data to drive refinements of protein crystal structures. We do this by simulating protein dynamics in the presence of the largest and most potent ligands in order to understand how ligands induce changes in the protein structure. This allows us to computationally refine crystallographic structures iteratively as a project progresses.

In the most advanced cases, this process of iterative structure enhancement has been applied to refine structural homology models to structure-based design quality. Homology models by themselves are seldom accurate enough for reliable ligand design, but if at least one ligand is known, we can optimize the homology model to a usable state by molding it to the shape of the ligand. Results of such a refinement are shown in Figure 3. In this case the structure of the allosteric site of Lck kinase was constructed by homology to the Abl-gleevec co-crystal structure. The Lck allosteric site was then refined by molecular dynamics in the presence of a known allosteric ligand. Figure 3 shows the overlap between the computed structure and an experimental co-crystal structure subsequently reported in the literature. The minimal starting point for the application of the Locus method, therefore, is a protein structure at homology model quality plus a known experimental ligand.

¹Grand Canonical Monte Carlo Simulation of Ligand-Protein Binding. Clark M, Guarnieri F, Shkurko I, and Wiseman J. (2006) J Chem Inf Model, 46:231-242.
² Closing of the Flaps of HIV-1 Protease Induced by Substrate Binding: A Model of a Flap Closing Mechanism in Retroviral Aspartic Proteases. Toth M. and Borics A. (2006) Biochemistry, 45: 6606-6614.