Locus Technology Validation

Locus Technology Validation
The first step in the validation of the Locus Technology was to determine whether the algorithms function accurately, reliably and reproducibly. The following are results from some of our validation experiments.

Confirmation of Known Protein Binding Sites
A basic validation of the Locus Technology is the ability to identify of known binding sites on proteins and to predict binding modes for known ligands. In the examples below, protein binding sites are defined by superimposing binding modes for multiple ligands or for multiple fragments, obtained experimentally via co-crystal structures or via Locus’ calculated fragment-protein interactions, respectively. In most cases, richer information about viable binding interactions is obtained using Locus Technology compared to experiment.

Calculated ligand binding modes are obtained by decomposing ligands into fragments, calculating the binding interactions with lowest free energy for each fragment, and reassembling the fragments to give back the structure of the parent molecule. Good agreement between experimental and calculated modes is expected for fragment-protein interactions with low entropies of interaction.

Elastase
Human neutrophil elastase is implicated in pulmonary inflammatory diseases such as emphysema and asthma. Locus Technology correctly profiled the elastase binding site after only 3 weeks of analysis, compared to 10 years of synthetic chemistry and crystallography required to identify the site experimentally.


Dihydrofolate reductase
Dihydrofolate reductase is a target for cancer chemotherapy. The binding Df methotrexate, a well-known anti-cancer agent, to dihydrofolate reductase is an intensively studied test case in structural biology. Locus Technology identified key fragments of methotrexate in the correct binding positions on the protein and rebuilt the whole molecule in the experimentally observed binding mode. It is also known that the tail region of methotrexate does not bind to the surface of the protein, and does not add any potency to the drug. Locus algorithms correctly predicted that the tail region of the molecule does not contribute to binding affinity:


HIV-1 Protease
HIV-1 protease is an important protein target for AIDS therapy. Locus algorithms predicted the binding site occupied by Crixivan, a commercially successful protease inhibitor, and correctly predicted the binding modes of both high affinity and lower affinity fragments this ligand.


Calculation of Fragment Binding Free Energies
The distinguishing capability of the Locus Technology is the ability to rapidly and accurately calculate relative binding free energies for a series of fragments. The simplest test of this capability is to compare calculated vs. experimental binding free energies for a congeneric series of molecules. Since only a single fragment position is varied in a congeneric series, this essentially tests the ability to evaluate binding energies for individual fragments. We provide data below for binding affinities of known molecules in two different types of protein binding sites, a relatively inflexible protease active site and a structurally flexible kinase allosteric site.

A more demanding test is to vary fragments at multiple positions across a molecule simultaneously, which is more sensitive to the effects of protein flexibility than is variation at a single position. Thus, while our calculations are based on a snapshot of a protein structure frozen in time, proteins are flexible and the structure of protein-ligand complexes can differ from that of native proteins. Although our technology cannot accommodate major reorganizations of protein structure, an example below shows that fragment-based molecule building is relatively insensitive to protein flexibility.

In evaluating the accuracy of our free energy predictions, one can use the accuracy of free energy perturbation calculations as a comparator. Although free energy perturbation calculations are too slow to be practical for de novo inhibitor design, they are the most accurate and rigorous alternative to Locus Technology. The accuracy of free energy perturbation calculations is on the order of 1 kcal/mol, which translates to 0.7 pIC50 units for binding data. In the examples below, Locus predicted free energies are relative and not absolute. A difference of 6 Locus energy units corresponds to approximately 1 log difference in IC50 values.

Protease Inhibition
The data below is for a series of protease inhibitors. The experimental results are still proprietary so that we cannot provide background on the protease. Nonetheless, we consider this protease active site to be an example of relatively inflexible binding sites, so that the prediction of fragment binding energies can be evaluated independent of protein flexibility with this data.


p38 Map Kinase
p38 Map kinase is a key protein in the signaling cascade for inflammatory factors such as IL-1 and is implicated in inflammatory diseases such as rheumatoid arthritis. This kinase has two substrate binding sites and a regulatory allosteric site. Allosteric sites such as that for p38 represent, by the nature of their biological function, highly flexible regions of the protein. Predictions of activity at this site, therefore, test the validity of fragment binding free energy calculations in the face of protein flexibility. Activity and crystallographic data used in these evaluations are from published structure activity data on compounds developed by Boehringer-Ingelheim. (Regan, J.; Breitfelder, S. et. al. J. Med. Chem. 2002, 45, 2994-3008.)




Free energy based predictions
The locus technology is based on a novel method to compute accurate free energies of binding of small molecules to proteins. These are then connected using proprietary de novo design software that computes the free energy of the overall molecule. The method has been validated by directing the de novo software to reproduce known ligands and their predicted binding. The rankings compare well to the experimental bindings, and the results are relatively insensitive to modest changes in side chain positions.

Comparison to docking methods
Current docking methods provide a score for different poses of a ligand in the binding site. While these are useful tools, the Locus method computes accurate free energies, which are more closely related to the measured binding.


Summary
The novel Locus technology, which includes free energy computation and de novo design, is able to accurately predict binding sites as well as map out the energetic nature of the binding site. The system provides a rich diversity of tightly binding molecules, and ranks their binding affinity. This provides a powerful guide for Locus and its partners to develop novel therapeutics.