Lead Optimization Overview Tutorial

Lead optimization is an iterative process. Lead optimization tools in this module can be combined to accomplish various tasks depending on specific project goals and available data. This tutorial illustrates one example that combines free-energy nonequilibrium switching (FE-NES) calculations and a 3D-QSAR model to guide optimization of a congeneric series, helping prioritize compounds with improved binding affinity or compounds that maintain affinity while improving other properties.

Putting Lead-Optimization Tools Together

  1. Choose an analog-generation method. There are several approaches that allow you to select a method that fits your design hypothesis and available inputs. For example, you can use the BROOD - 3D Fragment Replacement Floe to explore bioisosteric replacements of selected fragments in an existing lead, or the Focused Library - Core Input Floe to enumerate substituents at defined sites on a core using a reaction and reagent database. FE-NES comparisons require congeneric ligands that share a common core and binding mode. The analog-generation, posing, and post-posing triage steps described here are intended to preserve that relationship.

  2. Filter by simple properties. Next, you can apply project-specific cutoffs to known or readily calculated molecular properties to remove compounds outside the desired ranges. This provides a quick, threshold-based triage before more computationally intensive prediction methods. Early application of these filters helps focus downstream, structure-based evaluation and FE-NES calculations on compounds that satisfy basic project criteria. See the dataset filtering workflows for more information.

  3. Prepare molecular variants for structure-based evaluation. Before running the FE-NES Floe, compounds need a plausible receptor-bound pose that preserves the binding mode of the shared core. Review poses to confirm that the shared core retains a consistent orientation in the binding pocket. Depending on the ideation method, molecules may be generated as 2D structures, 3D conformers, or receptor-bound poses, and additional conformer generation and/or ligand posing may be required. For example, the POSIT - Ligand Guided Small Molecule Posing Floe can generate poses guided by a reference ligand. POSIT requires a protein input that retains the reference ligand so that analogs can be positioned in a related binding mode. Interaction analysis. can support pose assessment and help triage compounds based on their agreement with the binding hypothesis.

  4. Estimate relative binding affinity with FE-NES. With the ligands prepared and posed in the receptor, the FE-NES Floe estimates relative binding free-energy differences. For most systems, Rapid FE-NES supports thousands of affinity predictions per day, enabling same-day evaluation of large analog series. These affinity estimates can be used to prioritize compounds for synthesis, test binding hypotheses, explore SAR, discriminate among alternative binding modes, or train predictive models.

  5. Expand prioritization with 3D-QSAR active learning. Beyond direct prioritization, FE-NES predictions can also provide the training data needed to build predictive 3D-QSAR models. A trained model can be used directly to predict and prioritize compounds in larger sets. Alternatively, it can guide an active-learning campaign over a large, pregenerated ligand library. You can generate, filter, and pose the library, then run FE-NES on an initial subset. Those results can be used to train a 3D-QSAR model, which can then be applied to the remaining posed ligands to prioritize the next batch for FE-NES. Next, add the FE-NES results from each batch to the accumulated training data, update the model, and repeat. This approach uses FE-NES to sample the broader design space without running it on every molecule. Selection of the next batch can emphasize exploitation (choosing the top-ranked compounds that have not yet been run through FE-NES); exploration (choosing compounds for which the model is least confident in its predictions); or other active-learning strategies. In practice, this iterative approach can enable exploration of libraries containing thousands of compounds while running FE-NES on only a small fraction of them. For example, we have applied this approach to a library of approximately 5,000 compounds using five cycles of 100 Rapid FE-NES predictions.

Explore Other Approaches

This approach is one example of how FE-NES and 3D-QSAR can be combined in an iterative lead-optimization campaign to evaluate and prioritize analogs. Other lead-optimization approaches may combine different tools to accomplish other project goals. See the Lead Optimization Module for additional tools and workflows. For assistance choosing or combining approaches, please contact OpenEye Support.