Release Highlights 2022.2

OEToolkits 2022.2.2

The OEToolkits 2022.2.2 is a bug-fix of the OEToolkits 2022.2.1 release.

OEToolkits 2022.2.1

MCS based fix during OMEGA Conformer Generation

The ability to constrain a fragment of the molecules during conformer generation with torsion driving in OMEGA and the OMEGA toolkit has been extended to include fixing based on maximum common subgraph (MCS) match. In this mode of fixing, the MCS between the template fixmol and the generated molecule is determined, and subsequently that common subgraph portion of the generated molecule is fixed to the proved conformer of the fixmol.

The MCS based fixing during conformer generation can be turned on by setting OEConfFixOptions::SetFixMCS to true for use in the OMEGA toolkit. Additional API points have also been added to allow control during MCS search. A new flag -fixMCS is available in the OMEGA application to facilitate this.

Conformers for 3VV8_8838 Conformers for 4BFD_4239

OMEGA generated conformers for two different molecules, with MCS based fix against a single bound ligand pose. Figure shows that with MCS based fix, two different fragments are fixed during conformer generation of the two different molecules.

New ShapeFit Algorithm for Pose Prediction

The algorithm for pose prediction with the OEPositMethod::SHAPEFIT method has been modified. The new algorithm simultaneously optimizes the shape and chemical similarity between the fit molecule and bound ligand, and intra-molecular force-field energies of the fit molecule. The new simultaneous optimization algorithm replaces the previous adiabatic optimization algorithm and is more robust. The new algorithm also offers flexibility to use different forcefields and uses the OELigandFFType::SAGE force field by default. Another advantage of the new algorithm is that it is capable of producing multiple distinct poses when desired. The new algorithm is automatically reflected in both the OEPosit toolkit and the POSIT application. ShapeFit is also now available in the OEDocking TK as an independent API OEShapeFit.

Cross-docking experiments based on the 22 diverse kinase types presented in [Tuccinardi-2010] shows that ShapeFit accuracy is unaffected between the previous and the new algorithm (when accuracy is measured as RMSD <=2.0). A balanced cross-docking set of 20,000 points, where points are equally distributed over the various similarity regions and kinase types, was used. However, we see clear indication that generating multiple poses helps increase the pose prediction accuracy. A separate self-docking study based on the highly trustworthy iridium dataset ([Warren-2012]) with 280 ligand-protein structures shows that the new algorithm predicts poses that are closer to the x-ray crystallographic poses.

../_images/RMSD_Cross_Docking.png ../_images/RMSD_Self_Docking_Inset.png

Left: Comparison of ShapeFit accuracy in a cross-docking experiment between the existing and the new algorithm. The x-axis shows the tanimotocombo similarity between the ligand and the bound ligand present in protein complex. The Y-axis shows the fraction of accurately predicted poses based on being within 2 angstroms of the experimental pose. Right: Comparison of ShapeFit accuracy in a self-docking experiment, between the existing and the new algorithm. Axes show the RMSD of generated poses with respect to the experimental pose.

Protein-ligand Optimization with ff14SB forcefield and PB solvent model

Protein-ligand Optimization with ff14SB-OpenFF force field and Poisson-Boltzmann (PB) solvent model is now available both in SZYBKI and the SZYBKI toolkit.

In the SZYBKI TK, the functionality is available through the OEFixedProteinLigandOptimizer and the OEFlexProteinLigandOptimizer APIs, and can be enabled by setting OESolventModel::PB as the value in OEProteinLigandOptOptions::SetSolventModel. In SZYBKI both the OptLigandInDU and OptimizeDU applications now accept pb as value for -solventModel.

Optimization of 244 highly trustworthy x-ray structures of protein-ligand complexes from the iridium dataset ([Warren-2012]) with and without PB solvation and with ff14SB-Sage forcefield shows better accuracy for the optimized ligand poses when PB solvation is turned on. Accuracy is measured as the RMSD between the crystallographic ligand pose and the optimized pose. On an average PB solvation optimized structures have an RMSD of 0.31 angstroms, compared to a value of 0.55 angstroms when explicit solvation is not turned on. A paired T-test suggests that the average improvement of 0.24 angstroms is statistically significant with a p value of 0.00001. Turing on the solvation effects also removes any significant deviations (RMSD > 2 angstroms) of pose, as was seen for a few cases when optimization was performed in vacuum.

../_images/pb_opt_rmsd.png

Comparison of RMSD for Flexible protein-ligand optimization of x-ray structures with and without PB solvation and with ff14SB-Sage forcefield.

New BROOD Fragment Databases

Two new BROOD fragment databases, brood-database-ChEMBL31 and brood-database-ChEMBL31_lite, built from the latest version of ChEMBL, have been generated and are made available. The brood-database-ChEMBL31 contains all possible fragments up to 3 attachment points, whereas the brood-database-ChEMBL31_lite is curated to prioritize fragments with medicinal relevance.

Limited validation, with 29 different queries across 14 molecules, shows that the number of hits of druglike molecules obtained from the new databases are within 1% of those generated from the existing database brood-database-chembl-3.0.0, when a maximum of 1000 hits are requested. Comparison of belief score, a measure of potential activity of the generated hits, also shows that the hits generated from the new databases are at least as good as those generated from the existing database.

../_images/filtered_molecules.png ../_images/Belief_score_lineplot.png

Left: Comparison of the number of hits of druglike molecules using the brood-database-chembl-3.0.0 (ChEMBL20), brood-database-ChEMBL31 (ChEMBL31) and brood-database-ChEMBL31_lite (ChEMBL31_lite) fragment databases. Right: Comparison of the belief score distribution of all the hits generated using the ChEMBL20, ChEMBL31 and ChEMBL31_lite fragment databases.

Supported Platforms

Package

Versions

Linux

Windows

macOS

Python

3.7, 3.8, 3.9, 3.10

RHEL7/8, Ubuntu20/20-ARM/22

Win10/11

11, 12

C++

RHEL7/8, Ubuntu20/20-ARM/22

Win10/11 (VS2017/19/22)

11, 12

Java

1.8, 11

RHEL7/8, Ubuntu20/20-ARM/22

Win10/11

11, 12

C#

Win10/11 (VS2017/19/22)

General Notices

  • Support for macOS 12 has been added.

  • This is the last release to support macOS 11. Support for macOS 13 will be added in the next release.

  • Support for Ubuntu22 has been added. Support for Ubuntu18 has been dropped.

  • This is the last release to support RHEL7. Support for RHEL9 will be added in the next release.

  • This is the last release to support Visual Studio 2017 and Visual Studio 2019.

The detailed release notes consist of these highlights plus the individual release notes and the release history.