OEToolkits 2024.2

Release Highlights 2024.2.0

OMEGA: Thompson Sampling for Torsion Driving

A Bayesian approach to explore conformer space with torsion driving has been added as an alternative to the exhaustive searching of the space in OMEGA. This uses the framework of Thompson sampling to quickly direct the conformer search towards the low energy structures, reducing the time required to find the lowest energy conformer. As such, Thompson sampling is particularly suitable for generating small to moderately sized ensembles, such as used in the FastROCS or ROCS mode. Thompson sampling provides significant speedup for such ensemble generation, particularly for molecules with a large number of rotors.

Conformer ensembles for FastROCS usage are generated with OMEGA using both the standard exhaustive sampling and the newly implemented Thompson sampling, for ~5500 Iridium HT protein bound small molecule crystal structures. Performance between the two sampling methods is compared in terms of RMSD between the generated ensemble and the crystal pose, and in the time required to generate the ensembles. The results of the comparison, as shown in Figure 1 below, show that the accuracy of the generated ensembles, measured in RMSD, remains unaffected when using exhaustive versus Thompson sampling, whereas there is significant reduction in time usage to generate those ensembles when using Thompson sampling, especially as the size of the molecules grows.

../_images/omega_thompson_mmds.png

Figure 1. Comparison of OMEGA performance between exhaustive and Thompson sampling.

Ligand-based virtual screening performance, using both the exhaustive and Thompson sampling generated ensembles, was also compared and displayed in Figure 2. The comparisons were performed using directory of useful decoys (DUD-Z) datasets containing 38 targets. The results ensure that there is no meaningful change in accuracy when using Thompson sampling for ensemble generation for usage in ligand-based virtual screening.

../_images/omega_thompson_vs.png

Figure 2. Comparison of performance of ligand-based virtual screening success rate with conformer ensembles generated using exhaustive versus Thompson sampling.

BROOD: Bioisosteric Fragment Replacement in Macrocyclic Peptides

A new pose generation algorithm has been introduced in Bioisostere TK, as well as in BROOD, based on flexible overlay of shape, color, and force fields, as used in pose generation with ShapeFit. The new algorithm generates high quality poses of molecules that are obtained from fragment replacement in Bioisostere TK and BROOD. The robust pose generation algorithm enables usage of Bioisostere TK and BROOD for molecules with complex geometry, including macrocyclic peptides. Figure 3 shows an example of a residue fragment replacement in a macrocyclic peptide.

../_images/brood_macro_peptide.png

Figure 3. Bioisosteric replacement of a residue fragment in a macrocyclic peptide using BROOD.

Eon TK: New Toolkit for Overlay Optimization with Shape and Charge Density

The 2024.2 release introduces Eon TK, a new toolkit for molecular similarity based on shape and electrostatics. This provides toolkit-level access to the existing EON application functionality. Following EON, the new toolkit has both molecular charge density and electrostatic potential as options to use as descriptors of electrostatic properties. Figure 4 shows an example of EON overlay with charge density visualization.

../_images/eon_electrostatic_overlay.png

Figure 4. EON shape charge density overlay between two molecules.

Eon TK provides tools for molecular overlay optimization with shape and charge density, similarity estimation based on shape and charge density or electrostatic potential, and corresponding hit list building.

Supported Platforms

Package

Versions

Linux

Windows

macOS (x64 and arm64)

Python

3.9 - 3.12

RHEL8/9, Ubuntu20/22/22-ARM/24

Win10/11

12, 13, 14

C++

RHEL8/9, Ubuntu20/22/22-ARM/24

Win10/11 (VS2022)

12, 13, 14

Java

8, 11, 20

RHEL8/9, Ubuntu20/22/22-ARM/24

Win10/11

12, 13, 14

C#

Win10/11 (VS2022)

General Notices

  • Support for Java 8 and 11 is now available on all platforms, including macOS arm64.

  • Support for Ubuntu 24.04 has been added for C++, Python, and Java toolkits. Support for applications and VIDA will be added in the next release.

  • Support for macOS 15 is not available in this release but will be added in the next release. This release will be the last to support macOS 12.

  • Support for Python 3.13 is not available in this release but will be added in the next release.

  • Support for Ubuntu 20.04 has been continued in this release, which will be the last to provide support for Ubuntu 20.04.