.avif)
Your Ligand Binds, F-SAPT Reveals Why
F-SAPT uncovers hidden molecular interactions when traditional binding energy calculations fall short
You’ve been running Relative Binding Free Energy (RBFE) calculations on your lead series, trying to rationalize why certain modifications drastically impact activity—but the structure-activity relationship (SAR) just does not make sense. Your medicinal chemist is debating whether to synthesize additional compounds or pivot the series entirely. What do you do next?
In this post, we will explore how to apply F-SAPT (Functional group Symmetry Adapted Perturbation Theory) to this scenario to describe complicated SAR.
You will learn:
- RBFE vs F-SAPT: Why RBFE calculation fall short, and how F-SAPT can provide insight
- Why Promethium's F-SAPT is uniquely situated to overcome computational bottlenecks
- How researchers leverage F-SAPT in the lead optimization phase of drug discovery
Why Traditional Approaches Like Relative Binding Free Energy Sometimes Fail
Before taking further action, let’s discuss RBFE. While predicting experimental activity with free energy calculations is certainly useful in practice, much like any theory, this method fails to capture every effect happening at your active site. This is particularly true in cases where important non-covalent interactions drive activity—ones that are not captured by molecular mechanics.
Take the example of the STING protein, with its large, flexible, and polar binding site.1 The protein undergoes a large structural change from the open (inactive) to closed (active) upon binding with endogenous cyclic dinucleotides (CDNs). Designing a small-molecule STING agonist is difficult because there needs to be a balance of both strong binding and drug-like properties.
Balancing affinity, permeability, and bioavailability requires overcoming the inherent polarity and flexibility of the binding pocket while ensuring functional activation.
Researchers at Silicon Therapeutics and Roivant undertook a bespoke computational study to design a new STING agonist by combining physics based simulation and AI.2 Unlike traditional small-molecule designs, the STING agonist in this study functions through a self-dimerization mechanism, stabilized largely by face-to-face π-π stacking interactions.
This complexity makes this system difficult to analyze with RBFE calculations alone—driving them to look beyond out-of-the-box RBFE approaches.
Why F-SAPT is Essential in Drug Discovery
One of the tools employed to design the STING agonist in this study was F-SAPT. The researchers "chose to implement a QM-based approach for the dimerization interactions because traditional molecular mechanics force fields do not accurately account for interactions where polarization plays a significant role, such as the π − π interaction between the two molecules in the hSTING binding site. F-SAPT precisely predicts the quantum mechanical interaction energy (including polarization) between user-defined functional groups."2
F-SAPT is a method that can complement RBFE calculations by providing insight into the intermolecular interactions that drive protein-ligand binding. It is a wholly different method from RBFE, capturing physics of the forces that drive binding—the non-covalent interactions at the quantum scale. It differs from RBFE in many important aspects, but one difference is key.
The Key Difference Between FSAPT and RBFE
Fundamentally, it answers a different but important question—not how strongly your protein and ligand bind as RBFE does, but rather, what forces drive the binding from a quantum-mechanical point of view. This provides direct insight into the why of the SAR.
Over the last several decades, F-SAPT has evolved into a powerful tool in drug discovery due to advancements in electronic structure theory and computational hardware. Promethium’s F-SAPT capability can be used hand-in-hand with RBFE calculations to rationally design new ligands by leveraging insights from quantum chemistry.
Promethium's F-SAPT Implementation for Faster Ligand Design
Automation and GPU Acceleration Save Time for the Computational Chemist
Despite its power, adoption of F-SAPT has been slow due to the difficulty in setting up calculations, steep computational times, and limitations to small system sizes. Promethium’s protein-ligand cutout tool automatically fragments the protein in seconds, saving valuable time for the drug designer.
Setting up F-SAPT calculations requires fragmentation of the protein environment under study, often carried out manually by the user. Promethium has a protein-ligand cutout tool specialized to automatically fragment the protein into side chains and peptide bonds, reducing hours of setup time to seconds. A region surrounding the ligand is chosen by applying a distance cutoff, defined by the user, to save computational time.
The user can then fragment their ligand into chemically meaningful groups – ones that can be systematically modified to uncover how they drive key interactions between the protein and ligand. Ligand modifications can be evaluated using quantitative (data-driven) insights before moving to synthesis or assays.
The user can define chemically meaningful fragments of the ligand in under study. Systematic modifications can be made across a congeneric series to interrogate the effect of the specific functional groups on quantities like polarization.
Leverage GPU Acceleration Over Open-source Implementation
This end-to-end protocol removes the setup and computational barriers to using F-SAPT routinely in practice. This is in stark contrast to open-source implementations, which require users to write custom scripts to fragment the protein, in addition to compiling and maintaining optimized code on their own CPU hardware instances.
If you prefer writing scripts to a graphical user interface, Promethium offers a RESTful API with a software development kit in Python. With Promethium’s API, you can submit batch workflows using Python, making it simple to modify a congeneric series of ligands to evaluate with F-SAPT. Promethium scales elastically on AWS, removing bottlenecks of GPU availability on self-hosted compute.
By leveraging the power of GPU acceleration, Promethium makes it possible to use F-SAPT routinely in the Design-Make-Test-Analyze (DMTA) cycle. For example, our STING agonist system (PDB: 4LOH) takes ~3.5 hours to complete an F-SAPT calculation on a single A100 GPU. Calculations that took days on CPU architecture can be reduced to hours.
Benchmarks of Promethium's SAPT implementation compared with open source platform Psi43. In addition to offering a significant speed-up, Promethium can scale to system sizes of around 2000 atoms. Method/basis: SAPT-D3/jun-cc-pVDZ, Run on 1 A100 GPU.
To illustrate how these methods work in practice, let’s look at how F-SAPT was applied to STING agonist design, solving challenges that traditional RBFE could not.
Case Study: F-SAPT Solves a STING Agonist Binding Challenge
F-SAPT Provides Deeper Molecular Insights to Identify SNX281 as a Potent STING Agonist
When standard RBFE failed to distinguish between modifications, F-SAPT provided additional electronic structure-based insights, leading to better predictions of binding activity.
Throughout the study, 54,829 molecules were treated with a modified protocol of RBFE, while 1,293 of these systems were treated with F-SAPT. Modifications at several positions of the core ligand were refined using the F-SAPT method, allowing for a more precise prediction of ligand-ligand stabilization within the binding pocket.
This study successfully leveraged F-SAPT and other tools as part of a comprehensive workflow, identifying SNX281 as a potent compound through computational techniques and mouse models, ultimately leading to its advancement to clinical trials.
The outcome of this approach was striking: “Over the course of the project, we explored millions of virtual molecules while synthesizing and testing only 208 molecules in the lab.”2
Using F-SAPT in Your Structure-based Drug Discovery Toolbox
How common are cases like these—where predicted binding affinity is not well correlated with experimental activity? It’s difficult to say. For cases where binding originates from subtle electronic effects, RBFE calculations alone will not paint a clear picture. The molecular mechanics force field treats electronic effects in an averaged way, missing key contributions from interactions such as π-π stacking or intermolecular hydrogen bonds.
There is no one-size-fits-all computational protocol when it comes to optimizing your ligand, especially in a rapidly evolving structure-based drug discovery (SBDD) landscape. Interrogating intermolecular interactions using F-SAPT adds another important technique to employ as a computational chemist in drug design, unlocking more insight throughout the process.
The efficient optimization and validation of the STING agonist demonstrates the power of F-SAPT to enhance rational design in a way that RBFE alone could not.
“Theory guides, experiment decides.” — George Pimentel
So if RBFE generates puzzling SAR or you suspect strong electronic effects, queue up an F-SAPT run. You can integrate those results to refine the next set of analogs or confirm predicted potency ranking.
Ready to try?
Want to see it in action?
Looking for more uses of F-SAPT? See how researchers are using F-SAPT to:
- Uncover hidden intermediate range contacts that led to unintuitive SAR for a Factor Xa inhibitor:
- Understand how weak interactions stabilize transition states in organocatalysis:
- Build a machine learning model for interaction energies:
References: