3D QSAR Modeling Floes Release Notes

v1.2.0 August 2025

General Notice

• This package is built using OpenEye-orionplatform==6.5.1, OpenEye-toolkits==2025.1.0, and OpenEye-Snowball==0.30.0.

Floe Updates

• 3D QSAR Model: Builder

  • The floe now allows users to select a set of 3D models to build.

  • Two new 3D models, ROCS-GPR-NO-2D and EON-GPR-NO-2D, have been added. Unlike ROCS-GPR and EON-GPR, these two GPR models only use 3D Tanimoto as descriptors.

  • The floe now accepts kcal/mol and kJ/mol as potency units.

  • The default cross-validation method has been switched from “leave one out” to “random” to accommodate for a larger training set.

  • The option Include 2D in COMBO has been added to allow the user to decide whether to incorporate 2D-GPR into final model predictions.

  • A new parameter section, Cube Memory Parameters, has been added to allow users to increase the memory of some cubes for relatively large training sets to avoid cube memory errors.

• 3D QSAR Model: Interpretation

  • Grids corresponding to color atom probes have been added to the output dataset, along with contour values used for generating high and low surfaces.

  • The domain grid has been deleted from the output dataset.

  • The Generate ROCS Query parameter has been removed. ROCS query output will be generated by default along with ROCS-kPLS model interpretation.

  • A new parameter, Selected Models, has been added, allowing for the selection of kPLS models to interpret.

  • A new parameter, Percentage for Surface Contour Values, has been added to give the user more flexibility to choose contour values for high and low surfaces.

  • A new parameter section, Cube Memory Parameters, has been added to allow users to increase the memory of some cubes for relatively large training sets to avoid cube memory errors.

• 3D QSAR Model: Validation

  • A new parameter section, Cube Memory Parameters, has been added to allow users to increase the memory of some cubes for relatively large training sets to avoid cube memory errors.

• 3D QSAR Model: Predictor

  • A new parameter section, Cube Memory Parameters, has been added to allow users to increase the memory of some cubes for relatively large training sets to avoid cube memory errors.

v1.1.5 February 2025

General Notice

• This package depends on OpenEye-Orionplatform==6.2.0, OpenEye-Toolkits==2024.2.0, and OpenEye-Snowball==0.29.1.

Floe Updates

• 3D QSAR Model: Builder

  • A baseline 2D-GPR model is built by default and added to the output model dataset. The training set size for the 2D model is not necessarily the same as for 3D models because some molecules could be ignored in 3D model building for reasons such as failure of conformer generation, charge assignment, and so on.

• 3D QSAR Model: Predictor

  • Prediction for the COMBO model has been updated to be the weighted average of all individual models including 2D, with their corresponding prediction confidence as weights.

• 3D QSAR Model: Validation

  • Prediction for the COMBO model has been updated to be the weighted average of all individual models including 2D, with their corresponding prediction confidence as weights.

  • Histograms of pairwise similarities between validation and training sets have been added to the Floe Report.

• 3D QSAR Model: Interpretation

  • A new functionality has been added to optionally generate ROCS queries for virtual screening purposes.

  • Surface colors in the output dataset have been changed to be more consistent with atom colors in Orion.

New Floes

• 3D QSAR Model: Affinity Data Type Converter

  • The 3D QSAR Model: Affinity Data Type Converter Floe is a supplementary tool for converting affinity data from range or integer type to float type for model building.

v1.0.0 February 2024

General Notice

• This is the first release of the OpenEye 3D QSAR Models Floe package.

• This package depends on OpenEye-Orionplatform==6.0.0, OpenEye-Toolkits==2023.2.3, OpenEye-Eon==2.4.1.0, and OpenEye-Snowball==0.27.0.

New Floes

• 3D QSAR Model: Builder

  • 3D QSAR Model: Builder is a tool for building models with 3D descriptors. The floe incorporates:

    1. Pose conformer generation and charge assignment.

    2. Hyperparameter optimization for ROCS®- and EON-based kernel-PLS model building.

    3. Cross-validation.

    4. Model building.

    5. Optional external validation.

• 3D QSAR Model: Predictor

  • 3D QSAR Model: Predictor is a tool for making potency predictions based on a model against an external dataset.

• 3D QSAR Model: Validation

  • 3D QSAR Model: Validation is a tool for performing validation of a model against an external dataset.

• 3D QSAR Model: Interpretation

  • 3D QSAR Model: Interpretation is a tool for interpreting kernel PLS models.