Structure-based drug design (SBDD) is a paradigm in drug discovery focused on developing and interpreting 3D models of protein–ligand binding. Today, SBDD has become "an integral part of most industrial drug discovery programs" (Anderson). New computational and experimental technologies can serve to accelerate and improve the accuracy of SBDD approaches, making it likely that SBDD will remain a mainstay of drug discovery for decades to come.
Since structure-based drug design is based around atomistic models of the protein and ligands under study, acquiring accurate atomistic models is a crucial step in SBDD. Both computational and experimental techniques can be used here, and both have their advantages and disadvantages.
Traditionally, X-ray crystallography was almost always employed for SBDD: the target protein could be crystallized by itself to provide an unbound structure, or with a representative ligand to yield the structure of the protein–ligand complex. Crystallography remains a workhorse method, but is often challenging, particularly for certain classes of proteins (like membrane proteins). Identification of appropriate crystallographic conditions is highly empirical and can be extremely time-consuming, and any imperfection can lead to lower resolution and inaccurate structures. Accurate SBDD requires high-resolution structures, since minute differences in side-chain conformation can be crucial to any analysis of binding interactions.
Recent years have seen alternative methods emerge as alternatives to conventional X-ray crystallography, like NMR-based methods and cryoEM. CryoEM in particular addresses many of the challenges with crystallography while allowing for structures of complexes which are not compatible with crystallography. Although access to cryoEM facilities remains limited today for many researchers, it is likely that this method will see increasing use in the coming decades.
Advances in machine learning enable accurate protein structures to be predicted from sequence data, allowing scientists to predict 3D structures of their target purely in silico. Docking algorithms, either based on conventional scoring functions (like AutoDock Vina) or diffusion models (like DiffDock) can be used to generate ligand-bound poses, although the accuracy of docked structures is typically lower than that obtained by crystallographic methods.
Recently, a new generation of protein ML models allow for protein–ligand co-folding, or simultaneous prediction of protein structure and protein–ligand binding mode: this includes models like AlphaFold3, HelixFold3, and Chai. Although the accuracy of these models may be lower than conventional crystallographic methods, the ability to quickly generate structures for SBDD promises to accelerate SBDD, particularly in cases where experimental approaches prove intractable.
Once a suitable model of ligand binding has been obtained, drug designers can study the balance of intermolecular interactions and rationally design new drugs that should lead to higher binding affinity.
Visualization of the protein–ligand interaction allows scientists to see which parts of the ligand are interacting with which residues in the protein. Typically, a number of discrete intermolecular interactions like cation–π interactions, hydrogen bonds, or halogen bonds can be identified. Once this has been complete, scientists can perform rational stereoelectronic modulation to enhance the strength of these interactions and increase overall binding: for instance, adding electron-withdrawing groups to a phenol makes it a better hydrogen-bond donor, while adding a methoxy group to a pyridine makes it a better hydrogen-bond acceptor. The effect of these modifications can be assessed using in silico simulation before the compounds are synthesized experimentally.
Many drugs bind in a conformation that is not their lowest-energy conformation, necessitating the molecule to pay an energetic price to bind and lowering the overall binding affinity. Strategic conformational modifications, such as macrocyclization or biaryl substitution, can be employed to shift the global minimum towards the bound conformation and increase the strength of the binding interaction. In particular, torsional effects are often observed to be important sources of strain, and designing molecules with improved torsional profiles often leads to enhanced protein affinity.
Hydrophobic effects are often a major driver of protein–ligand binding, and ligands which optimally fill a hydrophobic pocket in a protein are typically very effective binders. Accurate structural models can be used to design ligands that better extend to fill all available space without causing destabilizing steric clashes with the protein; even ejecting one additional water molecule can lead to dramatically enhanced binding affinity.
Since structure-based drug design is so closely tied to an accurate understanding of the protein–ligand potential-energy surface, fast and accurate computational methods can be very powerful in SBDD. Rowan's molecular simulation platform makes it simple and straightforward to run accurate state-of-the-art computational workflows for SBDD, allowing you to get critical results faster and prioritize the experimental work that will actually make a difference.
Rowan's conformational search workflows use modern ML techniques in combination with physics-based methods to quickly and accurately generate conformational ensembles for any molecule, making it simple to assess how far the bound conformer is from the overall minimum, and survey the effect of further conformational restriction on ligand strain. Here's the output of a conformational search in Rowan—you can view conformers one at a time, or superimpose the entire ensemble.
As discussed above, torsional energy preferences can be key in SBDD. To analyze the PES around any given torsion, Rowan contains an optimized workflow for quickly generating torsional energy profiles. We use modern machine-learned interatomic potentials to compute accurate torsional barriers many times faster than conventional quantum-based simulations. In programs where speed of execution is crucial, Rowan helps your calculations finish in minutes—not days.
When you run calculations on Rowan, it's simple to share the results with your team. Our modern browser-based interface allows our users to easily gain crucial structural intuition about the binding poses of their ligand and generate publication-quality results right away. Plus, you can share calculations with other scientists right in your browser—no specialized viewing software required!
