We focus on the following areas:
- Small-molecule conformer generation, including complex macrocycles
- Molecular docking and protein binding site comparison and analysis
- Three-dimensional molecular similarity and multiple ligand alignment
- Prediction of affinity and pose for biologically interesting molecules
Our approach to predictive modeling is different because we respect physics, chemistry, and biology while also taking machine learning and parameter optimization seriously.
JCIM: Improvements in Docking
April 9, 2020
The latest improvements in Surflex-Dock, especially with respect to virtual screening, has been published in a special issue of JCIM. Ensemble docking is shown to produce much better results than single-structure docking on the DUD-E+ benchmark. Ligand-based methods employing eSim are shown to be competitive as well. Hybrid approaches are much better than single-mode approaches.
JCAMD: The eSim Similarity Method
October 24, 2019
We have published, with Stephen Johnson, a new 3D molecular similarity method (called eSim) that directly incorporates Coulombic field comparison with surface-shape and hydrogen-bonding comparison. It is both faster and more accurate that commonly used alternatives, both for virtual screening and pose prediction. The paper is entitled “Electrostatic-field and surface-shape similarity for virtual screening and pose prediction.”
JCAMD: ForceGen on Complex Macrocycles
May 25, 2019
We have published a new macrocycle-focused paper with Edward C. Sherer, Mikhail Y. Reibarkh, Qi Gao, Xiao Wang, and Yizhou Liu. The paper is entitled “Complex macrocycle exploration: Parallel, heuristic, and constraint-based conformer generation using ForceGen.” Extensive benchmarking results show seconds to minutes conformer elaboration for typical drug-like macrocycles and accurate ensemble generation for large peptidic macrocycles through the use of NMR-based constraints.
JCAMD: QuanSA Affinity Prediction Method
June 18, 2018
Our paper describing the new QuanSA approach (Quantitative Surface-Field Analysis) has been published. The method offers a novel mutliple-instance machine-learning method for binding affinity and pose prediction. It is applicable in cases where ligand structure and activity data are available either with or without a crystallographic structure of the protein target. The method is competitive with the FEP approach when structures are available, and the methods have uncorrelated errors, resulting in improved predictions using the approaches in combination.