Generating ternary complex structures to enable rational design of targeted protein degraders

Tutorial Created with Software Release: 2025-4
Topics: Small Molecule Drug Discovery
Products Used: ConfGen, Desmond, MacroModel, PIPER, Prime

Tutorial files

28 MB
This tutorial is written for use with a 3-button mouse with a scroll wheel.
Words found in the Glossary of Terms are shown like this: Workspacethe 3D display area in the center of the main window, where molecular structures are displayed

 

Tip: You can hover over a glossary term to display its definition. You can click on an image to expand it in the page.
Abstract:

In this tutorial, you will learn how to generate possible structures of the ternary complex for a target-degrader-ligase system for targeted protein degradation. You will use the Generate Degrader Ternary Complexes panel to obtain, roughly score, and refine initial poses for the ternary complex for a given degrader based on structures of the warhead-bound proteins. Then, you will use the metadynamics-based scoring in the Protein-Protein Interface Scoring panel to identify a set of three likely viable conformations for the ternary complex which you can use to enable further modeling, e.g. using FEP+ for linker design.

Tutorial Content

  1. Introduction

  1. Creating Projects and Importing Structures

  1. Prerequisites and System Preparation

  1. Generating Ternary Complexes

  1. Choosing Poses for Accurate Scoring

  1. Rigorous Scoring using Metadynamics

  1. Conclusion and References

  1. Glossary of Terms

1. Introduction

Targeted protein degradation is a promising new therapeutic modality which utilizes so-called 'degraders' to target a protein of interest for degradation by the proteasome via interaction with an E3 ubiquitin  ligase. The ligase transfers ubiquitin to the target protein, marking it for degradation. There are two main classes of protein degraders: molecular glues and bifunctional degraders. Molecular glues are similar to a traditional small molecule and typically have affinity for only one of the two proteins. On the other hand, bifunctional degraders, also called heterobifunctional degraders or PROTACs (Proteolysis Targeting Chimeras), are much larger and consist of two warheads that bind to the target protein and an E3 ligase (e.g. VHL or cereblon). Therefore, bifunctional degraders have affinity for both proteins, and both warheads are connected by a flexible linker.  In this tutorial, we will focus on generating and scoring bifunctional degrader ternary complexes, and we will use the term “degrader” to refer to this class of protein degraders.

Degrader design is quite challenging because of the unique mechanism of action and unique property space of these large and flexible molecules. At the very least, one needs to identify warheads, possible exit vectors for each protein and a linker able to bring both proteins together in a productive ternary complex that enables degradation. Knowledge of this productive complex is key for performing any kind of structure-based degrader optimization, including linker optimization. A key challenge is identifying a productive pose for the ternary complex, as a stable complex doesn't always guarantee that ubiquitination will take place, and in reality, the ternary complex might adopt more than one productive conformation.

This tutorial uses the well-described SMARCA2 – VHL system to demonstrate the recommended workflow for obtaining a reliable model of the ternary complex to enable computational linker optimization in the absence of an experimental structure. The SMARCA2 – VHL degrader used in this tutorial (PROTAC 2 in the 6HAX crystal structure)  links a SMARCA2-binding molecule with a VHL E3 ubiquitin ligase binder, facilitating the formation of a ternary complex that enables the SMARCA2 protein to be ubiquitinated and degraded, making it a key strategy for targeted cancer therapy.

Given a known active degrader and separate warhead–protein complexes, you will first generate a large set of potential structures for the ternary complex. You will use PIPER scoring and clustering to choose a subset of these structures for more rigorous scoring using metadynamics. Finally, you will analyze the results and choose the three top-ranked poses to bring forward into FEP+ linker design.

Figure 1: An overview of the workflow you’ll follow in this tutorial.

2. Creating Projects and Importing Structures

Figure 2-1. Importing the input data for this tutorial.

  1. Open Maestro and create a new project named degrader_modeling.prj for this tutorial.
  2. Download the tutorial zip file including input files and reference outputs here: https://www.schrodinger.com/sites/default/files/s3/release/current/Tutorials/zip/degrader_ternary_complex.zip
  3. After downloading the zip file, unzip the contents in your Working Directorythe location where files are saved for ease of access throughout the tutorial.
  4. Go to File > Import Structures.
  5. Find and choose degrader_ternary_complex.maegz from the tutorial files.
    • Structures are added to the Entriesa simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion and the top entry is included in the Workspacethe 3D display area in the center of the main window, where molecular structures are displayed.

3. Prerequisites and System Preparation

To generate and score ternary complexes, you will need warhead-bound structures for both the protein of interest and the ligase. If no experimental structures of the binary complexes are available, you can obtain them by using Glide or IFD-MD.

Structure files obtained from the PDB, vendors, and other sources often lack necessary information for performing modeling-related tasks mostly due to experimental limitations of structural biology techniques and insufficient or ambiguous data. Typically, these files are missing hydrogens, partial charges, side chains, and/or whole loop regions.

In this tutorial, the protein structures have already been prepared for you using the Protein Preparation Workflow. For detailed information on how to prepare protein structures, please refer to the Introduction to Structure Preparation and Visualization tutorial, and Best Practices for Protein Preparation.

4. Generating Ternary Complexes

In this section, you will generate a large initial set of plausible structures for the ternary complex, which can then be further reduced by scoring the protein-protein interface as shown in the next section.

Note that the Generate Degrader Ternary Complexes panel which you will use in this section is not applicable for molecular glues, as glue compounds are usually small molecules which do not have two warheads connected by a flexible linker.

4.1 Set up the ternary complex generation

Figure 4-1. Opening the Generate Degrader Ternary Complexes panel from the Tasks.

  1. Find and open the Generate Degrader Ternary Complexes panel in Tasks.

Figure 4-2. Loading the ligase and the target protein complexes.

First, provide the input structures of the binary protein-warhead complexes for both ligase and target.

  1. Select(1) the atoms are chosen in the Workspace. These atoms are referred to as "the selection" or "the atom selection". Workspace operations are performed on the selected atoms. (2) The entry is chosen in the Entries (and Project Table) and the row for the entry is highlighted. Project operations are performed on all selected entries the 6HAX_ligase entry and includethe entry is represented in the Workspace, the circle in the In column is blue the 6HAX_target entry.
  2. For Ligase complex, click Load and choose Selected Entry.
  3. For Target complex, click Load and choose Workspace Entry.

For the degrader structures, you can either provide complete molecules (consisting of both warheads and a linker) or a library of linkers from which the full degraders will be assembled.

If providing complete molecules, make sure that they are fully prepared using LigPrep. For the degrader in this tutorial, we will consider both the protonated and neutral forms, as both are accessible at the physiological pH.

Providing linkers only is particularly helpful when generating complexes for multiple degraders using the same warheads. You can find an example file with multiple linkers for this system in the tutorial files to help you with creating your own.

Figure 4-3. Loading the degrader structures.

The prepared degrader structures must be loaded from a file.

  1. For Degraders, choose Full molecules.
  2. Click Browse.
    1. A dialog box opens.
  3. Find and choose 6HAX_degraders_prepared.maegz from the tutorial files.
  4. Click Open.

There are multiple ways to generate the ternary complexes which are currently under development. At the moment, the only available method in the panel is Degrader Sampling. Here, the complexes are generated by first performing a search for low-strain degrader conformers. Then, the protein structures are aligned to each degrader conformer by matching the warhead positions. Poses with severe backbone clashing between the proteins are discarded. Finally, in an optional but strongly recommended step, the complexes can be refined and scored using PIPER (more on that below).

Figure 4-4: An overview of the ternary complex generation workflow. First, the conformational space of the degrader is explored (left). Then, both input proteins are aligned to their respective warheads in the degrader conformers to generate the ensemble of ternary complexes t (center). Finally, the ternary complexes are scored-in-place with PIPER (right).

The conformational sampling can be done using MacroModel, ConfGen, or a new implementation of ConfGen which uses RDKit for improved ring sampling (originally developed for Glide WS). Alternatively, you can provide pre-sampled conformers from other sources, e.g. molecular dynamics simulations or previous runs of the Degrader Ternary Complex generation. For a general introduction to conformer generation tools, see the Conformational Analysis for Small Molecules Using MacroModel and ConfGen tutorial.

We recommend running the pose generation with all three included methods in parallel and then pooling the top poses from each method for further analysis.

Figure 4-5. Specifying the sampling protocol.

First, you have to specify the sampling protocol and related options. As this is a computationally intensive job, we have provided pre-generated outputs for all three sampling methods in the tutorial files. If you want to run the job yourself, feel free to choose whichever sampling protocol you prefer (ConfGen is fastest).

  1. For Sampling protocol, choose one of MCMM, CONFGEN, or CONFGEN_RDKIT.

We recommend increasing the tolerated maximum number of clashes in the ternary complex from the default of 5 (very strict) to 10, to avoid discarding poses which might be rescued by refinement.

  1. Change Maximum clashes to 10.

Note: When applying this for your own system, you can enable Save Conformational Sampling to store all conformers generated here. If you then ever need to re-run the ternary complex generation for the same degrader(s), you can save computational time by reusing these conformers. In this case, choose Presampled as the sampling protocol.

In the Output section, you can specify how the generated poses are ranked and returned. By default, the top 50 poses are returned based on the energy property from the sampling method (i.e. an estimate of the conformational strain of the degrader).

You can instead choose to score the poses in place using PIPER, which captures the interactions in the entire complex. Additionally, you have the option to refine the protein-protein interface in the generated poses using Prime refinement for the top-ranked poses. Both of these options significantly increase the computational cost of the job, but we recommend using them as they significantly improve the quality of the top-ranked poses.

Note that none of these scoring methods are sufficiently accurate to narrow the list to a manageable number of poses (i.e. 3-5) to consider for linker optimization or other design work. You will refine the scoring in the next section.

Figure 4-6. Modifying the settings in the Output section and running the job.

We recommend significantly increasing the number of poses to return, scoring them with PIPER, and refining the top 20 poses for further work.

  1. Change Max poses to return to 1000.
  2. Enable Score with PIPER.
  3. Enable Refine complexes (interface & linker only).

Finally, you can set up the job settings. We recommend using multiple  processors per degrader, since although  it is currently not possible to parallelize the sampling per degrader, the PIPER and Prime jobs can be parallelized.

  1. Change Job name to Protac_6HAX_generate.
  2. Optional: Click Run.
    1. This job takes approximately 24 hours on two CPUs (depending on the selected sampling protocol).
    2. You can find pre-generated results in the tutorial files.
  3. Close the Generate Degrader Ternary Complexes panel.

4.2 Inspect the generated poses

After the job has run for one of the sampling methods, you will find the following output files in the job directory:

  • ProtacTask_degrader_sampler-out.maegz contains all generated ternary complexes that passed the clash threshold, ranked by the energy property from the chosen sampling method or — as we enabled scoring with PIPER — the PIPER energy (if that option is chosen).
  • ProtacTask_refine-out.maegz contains the refined structures for the top 20 poses from the previous set.
  • Various log files potentially useful for debugging.
  • If you enabled Save Conformational Sampling, you will also find all sampled degrader conformations.

To keep the size of the tutorial file package reasonable, we have only provided the full outputs for the confgen sampling as a reference for the curious. We’ve collected the refined structures for the top 20 poses by PIPER score for each of the sampling methods in one file for easy access throughout the rest of this tutorial.

Figure 4-7. Importing the pre-generated output files for all the sampling protocols.

You can now import the pre-generated output files for all three sampling protocols.

  1. Go to File > Import Structures.
  2. Find and choose Protac_6HAX_generate_all.maegz from the tutorial files.
  3. Click Open.
    1. This may take a few seconds.

We’ve split the results between the neutral and protonated forms of the ligands. Within each group, the entries are sorted by their PIPER pose energy in ascending order.

Figure 4-8. Stepping through the generated poses of the ternary complex structures.

You can now step through the results to see the diverse ternary complex structures that were generated.

  1. Confirm that the selected_PiperE_Top20_each group is selected in the Entriesa simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion.
  2. Use the left and right arrow keys to step through the poses.

Note: You can enable or disable the automatic fitting of the view when you switch between entries using the AUTO and LIGAND buttons in the main toolbar.

Figure 4-9. Choosing to show property in the Entries table.

Optional: As this is a retrospective example, you can also use the crystallographic pose to see if a crystal-like pose is recovered and how it is ranked.

  1. In the Entriesa simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion, click the three horizontal dots and choose Show Property.
    1. The Show Properties in table dialog box opens.

Figure 4-10. Adding the property to the Entries table.

  1. Click Choose.
  2. Type RMS in the search bar and choose the RMS In Place 182881 property from the search results.
    1. This is the Cɑ RMSD with respect to the crystallographic pose. It was calculated using the Superposition panel.
  3. Click OK.
  4. The chosen property is added to the Entriesa simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion table.
    1. You may need to drag the edge of the table towards the right to see the added property column.

Note: If you’re curious to see which sampling method each pose resulted from, you can also show the sampling method property we have added to the Entriesa simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion.

Figure 4-11. Comparing the generated poses with the crystallographic pose.

  1. Go to File > Import Structures.
  2. Find and choose 6HAX_prepared_full.maegz.
  3. Click Open.
  4. Right-click the In circle for the 6HAX_prepared_full.maegz entry and choose Fix in Workspace.
  5. Select(1) the atoms are chosen in the Workspace. These atoms are referred to as "the selection" or "the atom selection". Workspace operations are performed on the selected atoms. (2) The entry is chosen in the Entries (and Project Table) and the row for the entry is highlighted. Project operations are performed on all selected entries the selected_PiperE_Top20_each group Entry.
  6. Includethe entry is represented in the Workspace, the circle in the In column is blue the generated poses one at a time and compare them to the experimental pose.
    1. We would consider poses with an Cɑ RMSD < 10 Å to be crystal-like.

What can I do if I don’t get the poses I’m looking for?

In addition to enabling the PIPER-based scoring, you can –

  • increase the maximum number of poses to return.
  • use another sampling method or tweak the sampling parameters to explore more of the degrader’s conformational space, or combine conformations from multiple methods.
  • control how aggressively poses are discarded due to clashing. The default of allowing at most five clashes is a fairly strict cutoff, so you can increase this value to make this filter more permissive (i.e. by increasing Maximum clashes to 10 like in this tutorial or even further).

5. Choosing Poses for Accurate Scoring

In an ideal world, you could take all top 20 refined poses from each of the three available sampling methods in the preceding stage and score them using metadynamics. However, the metadynamics-based scoring is quite computationally costly, and so it might be helpful to add a filtering stage to reduce the number of ternary complex poses to rigorously score.

To this end, you have several options: If you have additional knowledge about the protein-protein interface (e.g. from HDX-MS or other experiments), you can use the Macromolecular Pose Filtering panel to identify poses which satisfy these criteria. Alternatively, you can use a clustering approach to identify groups of similar ternary complex poses and only bring a few representative structures from each cluster into the accurate scoring stage.

In this section, we will use both Macromolecular Pose Filtering and clustering to ensure we’re bringing forward a diverse sample of promising poses. Note that this is only one of several valid heuristics to narrow down the list of poses. You could alternatively use the clustering as described below to pick representative structures from each cluster rather than just as a diversity check.

5.1 De-duplicating entry titles using a sequential index

Figure 5-1. Choosing to add the Sequential Index to the entry titles.

As a preliminary step, you should rename the Entry titles so that there are no duplicates. You can do this by appending a sequential index to the title. This will also allow us to later look back at the top-scoring poses and see where they would have ranked based on only their PIPER energy.

  1. Open the Project Table.
  2. Select(1) the atoms are chosen in the Workspace. These atoms are referred to as "the selection" or "the atom selection". Workspace operations are performed on the selected atoms. (2) The entry is chosen in the Entries (and Project Table) and the row for the entry is highlighted. Project operations are performed on all selected entries the selected_PiperE_Top20_each group.
  3. Go to Data > Add Property > Sequential Index.
    • The Add Sequential Index Property dialog box opens.

Figure 5-2. Adding the Sequential Index column to the Project table.

  1. Enable Index each grouping level separately and Compute for selected entries only.
  2. Click Add.
    • The sequential index column is added to the table. It is based on the current sorting of the list, i.e. by PIPER energy.

Figure 5-3. The Combine Property Values dialog box.

For now, the index is a separate property. We can also append it to the Entry title to see it at a glance and give the individual poses distinct names.

  1. Go to Data > Combine Values.
    • The Combine Property Values dialog box opens.
  2. Under Properties, choose Title for the first property and Sequential Index for the second property.
  3. For Separate values with, type _.
  4. For Output result to, choose Existing property.
  5. Pick Title from the list.
  6. Choose Replace value.
  7. Enable Modify selected entries only.
  8. Click OK.
    • The entries in each of the subgroups now are numbered sequentially.
  9. Close the Project Table.

5.2 Filter protein-protein poses by property

Next, we will discard all poses which have a PIPER pose score > 0. While the PIPER scoring function is not accurate enough to rank-order poses, any pose with a positive score is highly likely to be very unfavorable.

Figure 5-4. The Macromolecular Pose Filtering panel.

Now, you can proceed with the actual filtering.

  1. Select(1) the atoms are chosen in the Workspace. These atoms are referred to as "the selection" or "the atom selection". Workspace operations are performed on the selected atoms. (2) The entry is chosen in the Entries (and Project Table) and the row for the entry is highlighted. Project operations are performed on all selected entries the degrader_protonated entry group.
  2. Find and open Macromolecular Pose Filter in the Tasks.
    • The Macromolecular Pose Filtering panel opens.
  3. For Use poses from, choose Project Table (60 selected) and click Load.

You can filter by structure-level properties, for example docking scores, or the RMSD to a reference structure. Alternatively, you can also filter on a residue level by specifying interactions which should be present or solvent-accessibility of certain residues. Finally, you can load in experimental hydrogen-deuterium exchange (HDX) scores and filter structures by how well they fit the experimental data. See the full documentation for the Macromolecular Pose Filtering panel for additional details.

Figure 5-5. Running the pose filtering job for the protonated degrader poses.

Here, you will use a simple filter to only retain poses with a PIPER pose score < 0.

  1. For Filter poses by, choose Structure properties.
  2. Click Add filter.
  3. Find and choose PIPER pose score from the list.
  4. In the text box for the PIPER pose score, type 0.
  5. For Job name, type 6HAX_complex_protonated.
  6. Click Run.
    • This job takes a few seconds..
    • A banner appears once the job is finished, with a shortcut to directly View Results.
    • You can find reference outputs in the tutorial files.

Figure 5-6. Viewing the filtering results.

The job doesn’t automatically load the resulting entries in the Workspacethe 3D display area in the center of the main window, where molecular structures are displayed. You will review the results and add the filtered entries.

  1. At the top of the Macromolecular Pose Filtering panel, for Choose task, switch to View filtering results.
  2. For load results from, click Browse and choose 6HAX_complex_protonated.
    • The results appear in a table in the lower part of the panel.

Note: The panel lists all recent pose filtering jobs run in this project. If you want to load other outputs, click Browse and navigate to the job directory.

Figure 5-7. Adding the filtering results to the Project Table.

Now you can add the filtered entries to the Workspacethe 3D display area in the center of the main window, where molecular structures are displayed.

  1. Shift+Click to select all the rows in the table.
  2. Click Add to Project Table.
    • A new Entry group named MMPF - 6HAX_complex_protonated is added to the Entriesa simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion.

Figure 5-8. Running the pose filtering job for neutral degrader poses.

You can now repeat the filtering steps with the degrader_neutral group.

  1. Select the degrader_neutral Entry group.
  2. Switch the Macromolecular Pose Filtering panel back to Pose filtering mode.
  3. For Use poses from, choose Project Table (60 selected) and click Load.
  4. Rename the job to 6HAX_complex_neutral and click Run.
  5. Once the job finishes, repeat steps 24 – 27 to add the filtered Entries to the project.
  6. Close the Macromolecular Pose Filtering panel.

After filtering out poses with very unfavorable PIPER scores, we are still left with around 40 poses — a bit too many to score with metadynamics. As these poses were all refined with Prime, you can also rank them via the Prime energy rather than the PIPER pose energy or PIPER pose score for further analysis. This is still a very approximate filter, but will allow you to carry forward poses with good interactions at the complex interface thanks to the refinement.

Figure 5-9. Sorting the selected entries by Prime Energy.

We will use the Prime Energy to choose the top 10 poses for both the protonated and neutral ligands.

  1. Open the Project Table.
  2. Search the Property Tree for Prime energy.
  3. Right-click one of the Prime Energy options and choose Show Only.
  4. Select(1) the atoms are chosen in the Workspace. These atoms are referred to as "the selection" or "the atom selection". Workspace operations are performed on the selected atoms. (2) The entry is chosen in the Entries (and Project Table) and the row for the entry is highlighted. Project operations are performed on all selected entries the selected_PiperE_Top20_each group in the table.
  5. Right-click the Prime Energy column title and choose Sort selected (Ascending).
    1. The entries within each of the groups are sorted by the Prime Energy.

Figure 5-10. Selecting and duplicating the poses  into a new group for metadynamics.

  1. Select(1) the atoms are chosen in the Workspace. These atoms are referred to as "the selection" or "the atom selection". Workspace operations are performed on the selected atoms. (2) The entry is chosen in the Entries (and Project Table) and the row for the entry is highlighted. Project operations are performed on all selected entries the top 10 poses from both the MMPF – 6HAX_complex_protonated and MMPF – 6HAX_complex_neutral groups in the table.
  2. Right-click one of the selected table rows and choose Duplicate > Into new group.
    1. The Duplicate into New Group dialog box appears.
  3. For New group title, type selected_for_MtD.
  4. For Location of new group, choose At top level : End of table.
  5. Click Duplicate.
    1. A new group titled selected_for_MtD with 20 poses is added at the end of the table.
  6. Close the Project Table.

This final set of 20 poses is what we will bring forward into the more accurate metadynamics-based scoring.

5.3 Optional: Use clustering to choose a diverse set of poses

Finally, we will show you how to perform interaction fingerprint analysis-based clustering on the generated poses. You can use this to confirm that the Prime energy based selection from the previous section has chosen a reasonably diverse set of poses. Alternatively, you can use this clustering as a standalone method to choose poses to bring forward.

The clustering is done in two steps. First, you need to generate interaction fingerprints which quantify the interaction patterns between the binding partners you specify. Then, the clustering stage calculates the distances between the fingerprints and groups similar structures together as clusters.

Figure 5-11. Generating the Interaction Fingerprints.

We will only cluster the poses that passed the PIPER score based filter we applied in the previous section.

  1. Ctrl+Click (Cmd+Click) to select(1) the atoms are chosen in the Workspace. These atoms are referred to as "the selection" or "the atom selection". Workspace operations are performed on the selected atoms. (2) The entry is chosen in the Entries (and Project Table) and the row for the entry is highlighted. Project operations are performed on all selected entries both the MMPF – 6HAX_complex_protonated and MMPF – 6HAX_complex_neutral Entry groups.
  2. Find and open Interaction Fingerprints in the Tasks.
  3. Choose Protein-protein complexes to generate fingerprints.

You now need to specify the binding partners. The degrader needs to be listed with the receptor chains, as the “ligand” for this analysis may only be a protein chain.

  1. For Analyze interactions of receptor chains, uncheck the box for chain B.
    1. Interactions of the receptor chains A, X bound to protein ligand chains B will be analyzed.
  2. Click Generate Fingerprints.
    1. This takes a few seconds..
    2. The Similarity and Cluster tabs of the panel are enabled once the analysis is complete.

With the newly generated fingerprints, you can now perform the cluster analysis and extract the cluster assignments. The backend automatically identifies the best number of clusters for your data set.

Figure 5-12. The Cluster tab in the Interaction Fingerprints panel.

With the newly generated fingerprints, you can now perform the cluster analysis and extract the cluster assignments.

  1. Switch to the Cluster tab.
  2. Click Calculate Clustering.
    1. This takes a few seconds..
    2. The Clustering Results section  updates.
    3. If you’re curious, you can look at the Clustering Statistics or Distance Matrix.
  3. In the Apply Clustering section, choose Duplicate entries to a new group for each cluster.
  4. Click Apply Clustering.
    1. New entry groups are added for each cluster, containing the structures assigned to it.

Note: See the panel documentation for more details on the different options in this panel.

  1. Close the Interaction Fingerprints panel.

Figure 5-13. Checking the entries assigned to the clusters.

Optional: You can now check whether the list of poses we chose in the previous section is well-distributed across the different clusters. Rather than comparing the lists manually, you can also use the Project Table.

First, reorder the Entries so that the selected_for_MtD group is at the top, followed by the results from the clustering.

  1. Open the Project Table.
  2. Select all Cluster groups, right-click them and choose Move to > Top of Table
  3. Repeat this with the selected_for_MtD group.

Now, we’ll use the fact that the poses share the same Entry titles to select those poses from the Cluster groups that share a title with an Entry from the selected_for_MtD group.

  1. Go to Select > First Entry per Title from > All Entries
    1. For each unique Entry title, the first Entry it is used for is selected.
  2. Go to Select > Invert All
    1. The new selection contains all Entries that are not the first occurrence with that title.
    2. Within the Cluster groups, the currently selected entries correspond to the poses chosen for metadynamics.

The poses we chose to bring forward to the metadynamics scoring stage cover four of the 11 identified clusters, with the largest two clusters being represented more frequently. This distribution is a good balance between choosing diverse structures while still focusing on the most promising areas of conformational space.

6. Rigorous Scoring using Metadynamics

The scoring done by the Generate Degrader Ternary Complexes panel is optimized for speed, and is very rough as a result. The Prime energy we used to choose the top ten poses for each of our ligand states is also a very approximate measure of pose stability as it is calculated from a static structure using an implicit solvent model.

Our validation has shown that crystal-like poses are likely to be found in the top 20 poses when ranking based on a combination of PIPER and Prime energies as we used in the previous section. Narrowing that list down to the 2-3 most likely poses is necessary for making it feasible to use these structures to enable design, but requires more accurate and computationally intensive scoring, which you will now perform.

6.1 Set up the Protein-Protein Interface Scoring

Before scoring ternary complex poses with MD-based methods, you should at least refine them with Prime, either manually or by enabling the option in the Generate Degrader Ternary Complexes panel. Re-running the Protein Preparation Workflow for the generated poses should not be necessary unless you suspect that there might be relevant H-bond pattern changes induced at the protein-protein-degrader interfaces.

Figure 6-1. Setting up the Workspace and loading the input structures for metadynamics.

Optional: If you’re just jumping into this tutorial here, find and import selected_for_MtD.maegz from the tutorial files.

  1. Right-click the In circle for 6HAX – prepared entry and choose Clear Fix.
  2. Shift+Click to select(1) the atoms are chosen in the Workspace. These atoms are referred to as "the selection" or "the atom selection". Workspace operations are performed on the selected atoms. (2) The entry is chosen in the Entries (and Project Table) and the row for the entry is highlighted. Project operations are performed on all selected entries all entries for the protonated degrader (with ..._prot_… in their title) from the selected_for_MtD group.
  3. Includethe entry is represented in the Workspace, the circle in the In column is blue the top entry in the selected_for_MtD group.
  4. Find and open the Protein-Protein Interface Scoring panel from the Tasks.

First, you will load the input structure(s) and specify the two protein chains whose interface should be evaluated.

  1. In the panel, for Get from, choose Project Table (10 selected) and click Load.
  2. For Chain Set 1, check the box for  A.
  3. For Chain Set 2, check the box for  B.

Note: Chain X is assigned to the degrader.

While you will only perform the set up of the scoring for the protonated degrader, the tutorial files contain the results for both the neutral and protonated forms of the linker.

Figure 6-2. Using the customized Force Field parameters.

We strongly recommend making sure that all torsions in your ligand are properly parametrized. You can use the Force Field Builder to automatically scan your ligand(s) for missing torsions and parametrize these using either DFT or a machine learning force field. For this degrader, we have provided the parameters in the tutorial files.

  1. For Force field, click the three dots next to Use customized version to specify the custom parameter location for this project.
  2. Click Browse.
  3. Find and choose the degrader_6HAX_oplsdir directory from the tutorial files.
  4. Back in the Protein-Protein Interface Scoring panel, enable the Use customized version option.

Note: See the Introduction to All-atom MD with Desmond tutorial for more details on choosing the right size and shape for the simulation box.

At the moment, the only scoring method available in the panel is by using metadynamics.

The main idea for this approach is to create a time-dependent bias as a function of a collective variable (CV) to sample the movement of the target and the ligase in the complex. From the biased simulation, we can output physical observables that might correlate with the interface stability, such as the potential of mean force (PMF) to separate the interface. However, for such complex systems,  obtaining converged results for the PMF would require a prohibitive amount of computing resources. 

To obtain meaningful scores without running the simulation to full convergence, we were inspired by a previously published study by Schrödinger (Clark et al, 2016) to rank order the relative stability of candidate ligand binding poses by the average ligand RMSD in reference to the starting pose from metadynamics simulations. For scoring protein-protein interfaces, we implemented a similar method that performs metadynamics simulations of each target-ligase interface with the distance between key residues on each side of the interface as the CV. The score is then calculated from a weighted average of the values for this distance observed in a simulation. Finally, the score for a given pose is normalized so different candidate ternary structure scores can be directly compared, and averaged over multiple independent simulations.

A manuscript containing the details of the scoring is in preparation.

Figure 6-3. Selecting and loading the target and ligase residues to be included in CV.

You can now define the residues from each binding site which will be included in the CV. Choose 2-5 residues from each pocket which roughly surround the warhead and, if possible, form stable interactions with it.

  1. Turn on the Interactions Toggle in the bottom right corner to make it easier to identify key residues.
  2. Press the R key to enter residue selection mode.
  3. On the target side (chain A), select PHE 1409, TYR 1421, and ASN 1464 residues in the Workspacethe 3D display area in the center of the main window, where molecular structures are displayed.

Note: You can also use the search icon in the Hierarchy to easily locate and select the residues in the Workspacethe 3D display area in the center of the main window, where molecular structures are displayed.

  1. In the panel, for Residue group 1, click Load Selected.
  2. On the ligase side (chain B), select PHE 76, SER 111, and HIE 115 in the Workspacethe 3D display area in the center of the main window, where molecular structures are displayed.
  3. In the panel, for Residue group 2, click Load Selected.

The default metadynamics parameters for this job are set to perform a slower, softer perturbation compared to the defaults in the Metadynamics panel, as our validation showed that this approach was more reliably able to recover crystal-like poses in the top-scoring poses. There should be no need to adjust these parameters to your system unless you experience issues.

The default combination of using six replica simulations of 20ns length per pose to score was chosen as a compromise between computational cost and scoring accuracy. Reducing the number of replicas to three is reasonable, but does reduce the reliability of the method to recover a crystal-like pose in the top three poses by score.

Recall that the goal for this simulation is not to obtain a fully converged PMF for each pose, which would require more replica and longer simulation times. Instead, we just probe how sensitive the protein-protein interface for each pose is to perturbation, with the underlying assumption that unstable PPIs will start “falling apart” more easily than stable ones.

Figure 6-4. Running the metadynamics job.

Finally, you can specify the job name and settings. As this job uses Desmond, you will need to run it on a Linux host with a GPU.

  1. Change Job name to MtMD_prot.
  2. Click the Job settings cog and choose the desired hosts for running the job.
  3. Optional: Click Run.
    • This is a computationally intensive job, running for approximately 20-24 hours per replica.
    • We do not recommend that you run this tutorial job yourself – use the pre-generated results from the tutorial files.

Note: As each replica will run for approximately one day, to improve throughput, we recommend using one GPU per replica if licenses and compute resources permit.

  1. Close the Protein-Protein Interface Scoring with Desmond panel.

6.2 Analyze the results

The key output files for the protein-protein interface scoring job are:

  • <jobname>_scored_poses.mae contains the input structures of the scored poses, and each pose’s score is accessible via the ppi score Entry property.
  • <jobname>_summary.csv contains the ranking of the input poses based on their scores (lower is better) averaged over the individual replica.

When looking into potential problems, it might be helpful to look into the following additional outputs:

  • <jobname>_consolidated_pmf.png is a plot of all the PMFs reconstructed from the simulations of the poses. This might be helpful to identify cases where the PMF has sudden jumps which might indicate a problem with the CV.
  • <jobname>_2-out.tgz contains the individual trajectories for each replica for each pose for detailed troubleshooting.

As before, we have only provided the full outputs for one of the jobs (in this case, for the protonated degrader). For the neutral form of the degrader, we have only provided the scored poses.

Figure 6-5. Importing the scored poses for both protonated and neutral degraders.

First, import the scored poses for both forms of the degrader.

  1. Go to File > Import Structures.
  2. Find and choose MtD_prot_scored_poses.mae and MtD_neut_scored_poses.mae from the tutorial files.
  3. Click Open.
    • Two new groups are added to the Entriesa simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion.

Figure 6-6. Adding the ppi score to the Entries table and sorting it in ascending order.

First, let’s sort the entries according to their scores.

  1. Click the three dots in the Entriesa simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion table and choose Show Property.
    • The Show Properties in Table dialog box opens.
  2. Click Choose.
  3. Type ppi in the search bar and click the result labeled ppi score.
  4. Click OK.
    • The ppi score is added to the Entriesa simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion table.
  5. Right-click the ppi score column header in the Entriesa simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion and choose Sort selected (Ascending).
    • The poses with the lowest scores are the best ones.

Note: You may need to drag the right edge of the Entries pane to make the ppi score column visible.

Looking through the results, you can see that the neutral form of the degrader generally returns slightly better scores. As an aside, performing the scoring on the crystallographic poses for the neutral and protonated degrader shows the same trend, with scores of 0.268 and 0.382, respectively.

Also, the top three highest-scoring poses for the protonated degrader would have ranked quite low based on the PIPER pose energy (reminder: the number in the entry title denotes the rank for this pose based on PIPER energy). Ranking by the Prime energy only also would not have given us the same poses in the top 3-5.

Figure 6-7. Stepping through the poses and comparing them with the crystallographic pose.

You can now compare the top-ranked poses to the crystallographic pose.

  1. Double-click the In circle of the 6HAX_prepared entry to fix the structure in the Workspacethe 3D display area in the center of the main window, where molecular structures are displayed again.
  2. Includethe entry is represented in the Workspace, the circle in the In column is blue the scored poses one by one and compare them to the experimental pose.
  3. Optional: Add the RMS in Place 182881 column to the Project table again to view the Cɑ RMSD for each pose compared to the crystal structure.

As you can see, in this retrospective example, we were able to recover crystal-like structures for the ternary complex for both the neutral and the protonated form of the degrader in the top three scoring poses.

When applying this approach to your own system, you will probably not have an experimental pose to compare against. In this case, you can use the top three poses to design modified degraders which are only compatible with one of the protein-protein interfaces. By obtaining experimental data on whether the modified degraders are active, you can then narrow down the list of possible poses for the ternary complex.

However, any modifications you introduce for this purpose should be small, as any change that affects the exit vector or linker attachment points, or significantly alters the length of the linker is likely to result in a completely different ternary complex geometry.

Also, keep in mind that in this tutorial, we focus on selecting the most stable ternary complexes for further degrader optimization. This doesn’t necessarily mean that these complexes will be productive resulting in degradation. In cases in which degradation is kinetically-driven, instead of thermodynamically-driven, the ternary complex is unstable and transient. In this regime, the use of structure-based methods becomes very challenging.

7. Conclusion and References

In this tutorial, you generated poses for a target-degrader-ligase complex based on the warhead-bound structures of SMARCA2 and VHL and a given linker. Using a rough scoring based on the conformer strain of the degrader, you refined the top-ranked poses and used them as inputs for a more rigorous, metadynamics based scoring. Finally, you examined the top-scoring poses and chose a small subset to use for further linker optimization with FEP+.

For further reading:
  • Prediction of Protein–Ligand Binding Poses via a Combination of Induced Fit Docking and Metadynamics Simulations (Clark et al, 2016; https://doi.org/10.1021/acs.jctc.6b00201)
    • This publication introduces the key concepts motivating the metadynamics-based scoring.

8. Glossary of Terms

Entries - a simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion

Included - the entry is represented in the Workspace, the circle in the In column is blue

Project Table - displays the contents of a project and is also an interface for performing operations on selected entries, viewing properties, and organizing structures and data

Recent actions - This is a list of your recent actions, which you can use to reopen a panel, displayed below the Browse row. (Right-click to delete.)

Scratch Project - a temporary project in which work is not saved, closing a scratch project removes all current work and begins a new scratch project

Selected - (1) the atoms are chosen in the Workspace. These atoms are referred to as "the selection" or "the atom selection". Workspace operations are performed on the selected atoms. (2) The entry is chosen in the Entries (and Project Table) and the row for the entry is highlighted. Project operations are performed on all selected entries

Working Directory - the location where files are saved

Workspace - the 3D display area in the center of the main window, where molecular structures are displayed