Sneak peek at Forge V10.6: Model building focus and much more

While the development team is busy giving the finishing touches to Forge V10.6, let’s have a quick look at what is new in this release.

Improved predictions through new models

Forge users told us that the development of QSAR models with strong predictive ability was still a pain point for their projects. Not surprisingly, this is what made us focus on model building in this release.

Forge V10.6 comes with a full panel of well-known and robust Machine Learning (ML) methods (Support Vector Machines, Relevance Vector Machines, Random Forests, kNN classification) which complement those available in previous versions (Field QSAR and kNN regression).

These ML methods can be used to build both regression and classification models, and this is reflected in a QSAR Model widget completely re-designed to provide relevant visualizations and statistics for both model types (Figure 1). While each regression and classification model can be built individually, there is an option in Forge to automatically run all the ML models and pick the best one for you.


Figure 1. Left: Observed vs. Predicted Activity graph for a SVM regression model. Right: Confusion matrix and statistics for a SVM classification model.

Generating qualitative models on small datasets

Activity Atlas is a qualitative method for summarizing the SAR for a series into visual 3D maps that can be used to inform new molecule design. Forge V10.6 includes a new Activity Cliff Summary method which generates more detailed SAR maps by slightly downsizing the importance of the strongest activity cliffs.

You may want to use the new flavor of the method for understanding the SAR of small to medium size data sets, as this will provide a finer level of detail. For larger data sets (e.g., for quickly understanding patent SAR information), the original algorithm will help you focus on the prevalent SAR signals.

More responsive GUI for larger projects

Working with large projects (more than 1,000 molecules with multiple alignments and QSAR models) will be much more efficient in Forge V10.6. You will see improvements in the performance of common operations such as application of filters, calculation and interaction with custom plots, exporting data. The calculation of the large similarity matrices in Activity Miner and Activity Atlas will also be faster, more robust and use less memory.

Furthermore, there is now an option to set-up Forge to use all the available local CPUs, if appropriate, as we have relaxed the 16-CPUs limitation in the previous release of the software.


Figure 2. Forge running on multiple local CPUs.

Improved interface to Blaze for virtual screening

The improved Blaze results window now shows an enrichment plot and statistics for each Blaze refinement level.


Figure 3. Improved interface to Blaze in Forge.

Stay tuned for more

Subscribe to our newsletter to receive the product release announcement, or contact us to learn more about Forge.

Flare Viewer: Free access to Flare for structure-based design

We are pleased to announce the introduction of Flare Viewer, a free licensing option of Flare, our structure-based design application. With Flare Viewer you can easily visualize and analyze your protein-ligand complexes, use our proprietary electrostatics to design new ligands, and communicate your ideas with high quality graphics and pictures.

Focus on ligands

Read in protein-ligand complexes by opening a file in a local or remote disk location, downloading multiple entries from the Protein Data Bank, or by drag-and-drop from your desktop if you are a Windows user. Ligands can be moved into the dedicated ligands table by drag-and-drop, with each ligand keeping the association with the protein it belongs to. Here they can be easily organized into custom groups, to keep your project tidy.

The dedicated ligand table and interactive menu gives easy access to all ligand actions: for example, sorting on any column, control visibility, tagging and filtering on structure, tags and numerical and text columns. A physico-chemical profile is calculated for every ligand and summarized in a fully customizable radial plot and multi-parametric score to help you design and select the ligands with the best fit to your ideal project profile.


Figure 1: The ligand-centric organization of Flare gives easy access to all ligand actions.

Explore ligand-protein interactions

Flare calculates and displays a variety of ligand-protein interactions. These include H-bonds, steric clashes, aromatic-aromatic, cation-pi interactions and more, also including water-mediated and intra-molecular interactions as an option.

Each ligand can be displayed with its associated protein in grid mode making comparisons between ligands or proteins straightforward.


Figure 2: Each ligand can be displayed with its associated protein, making it easy to compare the interactions of different ligands.

Iterative molecular design meets ligand electrostatics

Understanding ligand electrostatics is key in the design of improved ligands. In Flare, electrostatic interaction potentials calculated with the Cresset XED force field can be visualized as ligand fields or by mapping the electrostatic potential onto the ligand’s molecular surface.


Figure 3: Ligand electrostatics can be shown as ligand fields (left) and by mapping the electrostatic potential of the ligand on its surface (right). Color coding: cyan = negative electrostatics; red = positive electrostatics.

Designing new ligands in Flare gives you immediate feedback on electrostatic changes in the context of the protein active site. In the molecule editor, the ligand or a selected part of the ligand can be minimized ensuring bonds, angles and torsions have low energy values.


Figure 4: The Molecule Editor.

Compare multiple proteins

Multiple protein structures can be imported in the same project and displayed in the same frame of reference using the sequence alignment and superimposition functions in Flare. You can choose the protein to superimpose to, whether all proteins are to move and if all residues or selected residues are superimposed. The protein structure can be optimized by flipping flexible residues or changing tautomeric and charge states for relevant residues.

Once opened, the proteins will sit in a dedicated table where all their components (chains, ligands, crystallographic waters and cofactors) are clearly visible, enabling a rapid inspection of specific chains or residues.

Protein surfaces can be displayed and colored by solid color, atom, secondary structure and hydrophobicity, and saved in a dedicated protein surface window.


Figure 5: Comparing multiple protein-ligand complexes is made easy by working in grid mode, showing ribbons and applying different protein surfaces styles.
Important scenes can be captured and annotated in the Storyboard to be recalled when needed.  Images can be easily copied and exported, with many options to configure the image or file size.

A dedicated extended atom picking widget enables complex queries and gives you full control on what is selected and displayed in the 3D window.

Protein viewer with an intuitive GUI

The ribbon menu structure of Flare makes it easy to identify the commands and controls you are looking for, as all actions are always visible and organized in a logical structure.


Figure 6: All actions are always visible in the Flare ribbon menu.

Upgrade to the Flare Python API

Upgrading Flare Viewer to include the Flare Python API will enable you to create your own workflows, automate common tasks, add custom controls and context menus, access Python modules such as the RDKit cheminformatics toolkit, NumPy, SciPy, and Matplotlib. We also provide a collection of featured python extensions that extend the existing Flare functionality.

Discover Flare Viewer

See the features of Flare Viewer, and download your free 1 year license.

Bespoke free licensing options for academic users are also available; see the announcement.

Flare for Academics

We believe that the lively academic environment is an amazing source of new scientific ideas, algorithms and computational methods. Flare for Academics is a free* licensing option of Flare, our structure-based design software, which has specifically been designed for academic users.

Flare for Academics is a user-friendly environment where academic users can easily develop and test their ideas and methods, or plug-in the most interesting open-source algorithms. It extends on the functionality of Flare Viewer to provide an excellent platform for drug discovery, with a focus on ligand design and electrostatics.

Discover the power of the Python API

The Flare Python API gives academic researchers the opportunity to make their science more accessible through integration into a user-friendly environment.

An environment to build upon and create great science

You will benefit from a robust, commercial standard SBDD environment that enables focus on science by utilizing methods such as protein preparation, protein minimization and multi-core docking. Access is also given to the RDKit cheminformatics toolkit, NumPy, SciPy, and Matplotlib, which are all integral to Flare. Beyond these, virtually any other Python module can be pip-installed making Flare infinitely extendable. An ever-growing collection of featured python extensions that enhance the existing Flare functionality are also provided, these include: plotting, protein mutation, and custom workflows (see also the new Jupyter Notebook integration).


Figure 1. The ‘Extensions’ tab in Flare 2.0.

Low-level access to the graphical user interface and internal processes

The Flare Python API not only provides an environment to develop your own algorithms but also a way to deploy them across a wider user base. The API provides access to all elements of the Flare interface through addition of user-defined controls and context menus.

For example, you may add custom controls into an existing Flare ribbon, or create a new Flare ribbon for Python scripts you frequently use. Custom-created controls in Flare can be created as small or large buttons, spin boxes, custom sliders, or complex dialogues with signals and call-back functions (Figure 2).


Figure 2. Some types of custom controls which can be added to a Flare ribbon.

Automate and distribute Flare calculations

Whenever you need to carry out a completely automated task, for example the overnight preparation of a panel of proteins followed by docking of several ligand series, the most convenient option is to write a Python script that runs outside the Flare GUI. It can then be distributed on a cluster via a queueing system for maximum performance. The pyflare binary is a Python interpreter giving you access to Flare functions using either custom developed or Cresset released scripts.

Upgrade Flare with the Jupyter QtConsole

The native GUI of Flare embeds the Python Console and Python Interpreter widgets. The Python Console is the simplest option to run one-line commands. With the Python Interpreter you can handle slightly more complex scripts: for example, you can load a script, interactively edit it inside Flare and then save your modifications. Both the Python Console and the Python Interpreter have a multi-tab interface that makes it possible to work on multiple Python snippets at the same time.

Python enthusiasts can easily upgrade Flare with the Jupyter QtConsole for access to all the Jupyter features, e.g.: TAB completion, auto-indentation, syntax highlighting, context help, inline graphics, and more. Using this widget, you can type Python commands, examine molecules and draw plots, all in the same window.

Upgrade Flare with the Jupyter Notebook

The Flare Python Notebook is an instance of the Jupyter Notebook embedded into the Flare GUI. It has direct access to the Flare GUI objects and methods, offers an even richer interface and enables editing and running individual code cells.


Figure 3. The Python Qt-Console (left) and Python Notebook (right) in Flare.

Not just a viewer

Flare for Academics is not just a viewer, but a complete, user friendly platform for iterative molecule design in drug discovery.

Multiple protein structures can be easily imported in the Flare project and displayed in the same frame of reference using the sequence alignment and superimposition functions.

Flare’s protein preparation will enable you to optimize your protein-ligand structures by adding hydrogen atoms, optimizing hydrogen bonds, removing atomic clashes and assigning optimal protonation states. Further optimization of the protein active site can be achieved by protein minimization based on the XED force field, and by manually flipping flexible residues or changing tautomeric and charge states for relevant residues.


Figure 4. Flare for Academics is user friendly platform for iterative molecule design in drug discovery.

 
Smart visualization of protein-ligand complexes in grid mode facilitates the comparison between ligand or proteins. The display of a variety of non-bonded ligand-protein interactions makes it easy to understand the different binding modes for your ligands.

The ligand-centric structure of Flare includes a dedicated ligand table and interactive menu giving easy access to all ligand actions, such as sorting on any column, control visibility, tagging and filtering on structure, tags and numerical and text columns, grouping of ligands in custom-created roles. In the ligand table, each molecule is associated to calculated physico-chemical properties, a radial plot and a multi-parametric score to help you design and select the ligands that best match the ideal project profile. Ligand electrostatic interaction potentials calculated with the XED force field can be visualized in the 3D window and in the molecule editor, and used to inform ligand design.

Multi-core docking experiments can be run to predict the 3D structure of flexible ligands in the active site of your protein. Docking in Flare uses Lead Finder™ to provide excellent pose prediction and detailed feedback on new molecule designs.

Discover Flare for Academics

See the features of Flare for Academics, and apply for your 1 year license.

* In most countries; contact us to see if you are eligible for a free license.

Python extension enabling Jupyter Notebook integration in Flare released

In a recent post I wrote about Integrating Jupyter Notebook into Flare as a new Python extension dedicated to Python developers and enthusiasts. The Python extension that makes this possible is now released (Figure 1).


Figure 1. The button which enables the Python Notebook extension.
While using it to carry out my daily Python coding tasks, I have identified a number of features that the protoype extension was missing and were worth implementing. So, there are a few more highlights that I’d like to share with you.

As discussed in my previous post, the feature that personally I enjoy most is the fact that the Flare Python Notebook has direct access to the Flare main_window() object, and hence allows you to work on the project currently loaded in the main viewport, i.e., interact with ligands and proteins, visualize molecular and field surfaces, etc. As this involves running the Python code in the main GUI thread, only a single Python Notebook may have access to the GUI at any given time.

However, I thought it would be useful to be able to run other concurrent, separate pyflare processes within the same Python Notebook while the main GUI process is busy doing a computation, e.g., preparing a protein (Figure 2):


Figure 2. Download a PDB complex in the GUI, then run Protein Preparation.
The Python Notebook remains responsive while the Protein Preparation task is run by a FieldEngine process in the background. This means I can open a second Python Notebook tab and, for example, visualize the 2D ligand structure. Since the new notebook tab runs as a separate pyflare process, it does not have access to the Flare main_window() object, as shown by the absence of the Flare icon and by the tooltip (Figure 3):


Figure 3. Open another tab and carry out some other task in a separate process.
Once the calculation has finished, you can switch back to the main tab and keep on working there.

To provide better integration with the Flare GUI, I have moved the familiar ‘Kernel’ notebook menu controls to the bottom of the window (Figure 4):


Figure 4. Restart/Stop commands can be accessed from bottom left buttons.
Also, the Load/Save commands were moved from the File menu to buttons, in order to provide more control on the location the notebooks can be saved to or retrieved from (Figure 5):


Figure 5. Load/Save notebooks through a standard file dialog.
The Python Notebook extension is now ready for download from Developers extension on our GitLab page. I’d be really keen on hearing thoughts and ideas from other Python enthusiasts out there, so please do not hesitate to get in touch if you would like more information, have feedback or have suggestions for new features in the next version of the Python Notebook.

Which macrocycle should I try first? Picking the best linkers with Flare™ and Spark™

At Cresset, we enjoy seeing our products work in synergy. By combining the most recent scientific methods and workflows we deliver solutions to address molecule design challenges. In this post, we use the new Electrostatic Complementarity™ (EC) maps and scores in Flare to help the post-processing of a Spark macrocyclization experiment.

Using Electrostatic Complementarity in Flare to post-process the Spark results

In the case study Using Spark to design macrocycle BRD4 inhibitors, we used Spark, our bioisostere replacement and scaffold hopping tool, to design macrocyclization strategies for non-macrocyclic, pyridone BRD4 inhibitors and evaluate results against experimental data reported by Wang et al [1]. The results showed that Spark successfully reproduced the experimental data.

In a real drug discovery project where no retrospective data is available, it would be useful to have criteria based on the existing knowledge of the system under study helping a further post-processing of the Spark results. Here we show how to use Spark in synergy with the EC maps and scores in Flare, our structure-based design tool, to pick the most promising candidates for synthesis.

Electrostatic interactions are essential for molecular recognition and are also key contributors to the binding free energy ΔG of protein-ligand complexes. Assessing the electrostatic match between ligands and binding pockets provides important insights into why ligands bind and what can be changed to improve binding.

The 100 top scoring results from the BRD4 Spark experiment were opened in Flare using the ‘Send to Flare’ functionality in Spark, which also transfers the related starter molecule (compound 1 in Figure 1) and excluded volume protein (5UEY). The protein was prepared in Flare, removing the water molecules that do not make clear interactions with both the ligand and protein. EC scores and maps were then calculated for compound 1 and the experimentally validated macrocycle 2 reported by Wang et al. towards the same 5UEY protein, as shown in Figure 1. As expected, the EC maps for both compounds show good complementarity to the protein and a very similar EC R score of 0.52/0.53 (Pearson’s r correlation coefficient). Spark linkers showing similar (or better) maps/score should provide interesting ideas for synthesis.


Figure 1: EC maps and scores for compound 1 and macrocycle 2, calculated towards protein 5UEY. Color coding: green = good complementarity; red = electrostatic clash.

Picking the winners

Figure 2 shows a couple of the most interesting linkers in terms of EC score.


Figure 2: EC maps and scores (top panel) for two ‘matching’ Spark linkers, calculated towards protein 5UEY. Color coding: green = good complementarity; red = electrostatic clash. The bottom panel shows electrostatic potential maps for the same Spark results. Color coding: cyan = negative electrostatic; red = positive electrostatic.

In the first example (Figure 2 – left), the π-system in the double bond linker complements the positive electrostatic field at the NH proton of His437 better than compound 1 or a fully saturated linker of similar length, as in macrocycle 2.

Another interesting example of good electrostatic match is the mercaptoethanol linker (Figure 2 – right). The negative electrostatic field of the thioether group is also in close proximity to the polarized NH of His437.

For both compounds, the increase in EC towards the protein is due to the introduction of a more negative ligand electrostatic in the region near His437, as shown by the electrostatic potential maps for both linkers (Figure 2 – bottom).

Discarding the losers

In contrast, an analysis of the EC maps for two of the linkers with the lowest EC scores (Figure 3) immediately highlights the reasons why these should be down-prioritized.


Figure 3. EC maps and scores (top panel) for two ‘clashing’ Spark linkers, calculated towards protein 5UEY. Color coding: green = good complementarity; red = electrostatic clash. The bottom panel shows electrostatic potential maps for the same Spark results. Color coding: cyan = negative electrostatic; red = positive electrostatic.
These linkers expose an area of negative interaction potential towards the carbonyl of Asn443, resulting in a strong electrostatic clash.

Conclusion

Are you surprised that a few linkers with low EC ended up among the top 100 scoring Spark results? Don’t forget that Spark works on ligand similarity. In macrocyclization (and fragment linking) experiments we are stretching the method to explore regions in space where ‘no ligand has gone before’.

In such cases, adding protein information is clearly highly beneficial to help post-processing. EC maps in Flare are an intuitive visual method for rationalizing the choice of the best ideas to progress, while EC scores provide a rapid way of scoring and filtering the 500 Spark results in just a few minutes.

To try Spark or Flare on your projects, request your free evaluation.

  1. Wang, L.; McDaniel, K. F.; Kati, W. M. Fragment-Based, Structure-Enabled Discovery of Novel Pyridones and Pyridone Macrocycles as Potent Bromodomain and Extra-Terminal Domain (BET) Family Bromodomain Inhibitors. J. Med. Chem. 2017, 60 (9), 3828–3850.

Sneak peek: PickR to select electrostatically diverse monomers for libraries

In a presentation at the Cresset User Group Meeting in 2016, Nik Stiefl and Finton Sirockin from Novartis discussed the selection of building blocks for DNA encoded libraries using electrostatic and shape diversity as the key descriptor. This work was powered by a custom binary and scripts written by Cresset. Over the last couple of years this approach has been applied to a wider range of library designs and has gained a reputation as a method of choice for many library designs.

PickR will be a new tool that formalizes the approaches that we developed in collaboration with Novartis. It is a command line binary that provides a diverse pick of reagents to be incorporated into a library. Unlike other approaches, PickR uses the electrostatic and shape properties of molecules to generate the descriptor matrix that is used as the basis for the diversity pick.

Generating diversity using a 3D property is not straight forward. It is necessary to explore conformations of R-groups and rotate them about the proposed connection point in order to fully understand the distribution of properties. Much of this was well described in Nik and Finton’s presentation (slides 12-15). I will leave a formal discussion to the final release announcement.

Application of PickR to amino acid side chains

I applied a dataset of approximately 1,000 amino acids that can be purchased from eMolecules. The raw reagents were processed to convert the side chains into R-groups with the C-alpha atom being converted to Iodine (all other Iodine containing reagents were excluded as were those containing Br and those with side chains >150Da). Using PickR, I generated a 3D similarity matrix for the side chains, aligning on the C-alpha to C-beta bond. 100 clusters were requested initially.

 


3D representations of all 100 clusters generated from amino acid side chains, aligned to each other using the I-C bond of the fragments.

Looking at the results, there are some very nice relationships. For example, in Cluster 2, together with tyrosine, are other phenolic side chains but also an indazole that contains the donor-acceptor motif.

 


2D representations of all the side chains in the same cluster as tyrosine (highlighted).

Along with the indole of trytophan are other substituted indoles, pyropyridines and benzofuran. In with the isobutyl side chain of leucine are a number of cyclic analogues which I expect would cause issues with many 2D similarity methods. Interestingly, indoline is placed together with the equivalent of homo-phenylalanine.

 


2D representations of the leucine related cluster.

 


2D representations of the homo-phenylalanine related side chains

 

The cluster containing the phenylalanine side chain highlights the major difference of PickR over other methods – R-groups are clustered on 3D electrostatic properties. Hence, together with the phenylalanine side chain you have thiophenes and pyroles but few other aromatics – pyridine and pyrimidines go to their own clusters because using electrostatics they are quite different to a plain phenyl ring.

 



3D and 2D pictures of all side chains in the phenylalanine cluster

 

Request project file and find out more

Contact me to receive the full results in a Forge project file.

Contact your account manager if you are interested in learning more about PickR.

Integrating Jupyter Notebook into Flare

In a recent blog post I have shown the integration of the Jupyter QtConsole in Flare.

The Jupyter QtConsole nicely fits in with the rest of the Flare GUI and provides a comfortable Python shell environment with most of the nifty Jupyter features such as history, TAB completion, syntax highlighting, embedding of images, etc.

Since I published that post, I have started thinking that it would have been great to embed a Jupyter Notebook, as it offers an even richer interface: most importantly, it enables editing and running individual code cells, thus constituting the ideal environment for Python enthusiasts.
There were a few technical hoops that I had to jump through to get this to work, but I finally managed.

So here I am proudly presenting the Flare Python Notebook, i.e. an instance of the Jupyter Notebook embedded into the Flare GUI which has direct access to the Flare GUI objects and methods just as the Python QtConsole (Figure 1).


Figure 1. A screenshot showing the new Python Notebook embedded in Flare.
To demonstrate the new fuctionality, last week at the 7th RDKit UGM in Cambridge I gave a lightning talk showcasing a sample Python Notebook which downloads a set of AChE inhibitors from ChEMBL, loads them into Flare, generates 3D coordinates and field points, and finally docks them to a crystallographic AChE protein downloaded from the Protein Data Bank.
The RDKit is used to compute 2D similarities and maximum common substructure across ligands and to generate 2D molecule layouts. RDKit molecules are fully interoperable with Cresset molecules within Flare, so Cresset 3D technologies and RDKit methods can be synergycally combined in one Python workflow.
Matplotlib, NumPy and SciPy are used to generate a scatterplot with a regression line and compute some statistics.

Now, on to the Jupyter Notebook:

The Flare Python Notebook unleashes the full potential of embedding a highly interactive Python environment within the Flare GUI.
RDKit cheminformatics methods and Cresset 3D technologies can be used side-by-side and their results visualized in real time while writing Python code, making the development cycle much more efficient and expedite.
The Jupyter Notebook has made scripting a simple everyday task for cheminformaticians, bioinformaticians and data scientists. I am confident that the Flare Python Notebook will do the same for CADD scientists and computational chemists.

And if you are not a Python guru, I can help you out; actually, I can’t wait to write my next script!

We will be releasing the Jupyter Notebook integration to our GitLab repository of Python extensions as soon as it is finalised. To find out more about Flare or to talk about how the Python integration can help you in your research or to request a Python script to achieve a particular task within Flare, then please contact us.

Using Python in Flare to find common contacts

In a recent blog post Pat Walters nicely used the structures of Viagra and Cialis when bound to PDE5 to argue that scaffold hopping between these two drugs was not a task that could be performed easily. He used Python to demonstrate that each drug interacted with siginificantly different parts of the protein and that they only shared interactions with 4 residues. Inspired by this, I sought (with the help of Paolo Tosco) to implement Pat’s code in Flare.

Paolo has been working on the implementation of a Jupyter notebook within Python (see his post here) and this provides the ideal environment to implement and discuss code to explore the common and specific interactions of Cialis and Viagra with PDE5. The notebook contents are shown in the iframe below.

As you see the output (last line) is the same as originally reported. However, with the addition of the Flare interface we are able to create a nice visual representation of the results, rendering the common and ligand specific residues differently. The script takes around a 30 seconds to run:

If you would like to learn more about Flare and using Python to customize, script or automate common actions or you would like to try the code out for yourself then please contact us. The current range of Python extensions for Flare are avaiable from our GitLab repository.

Investigating the SAR of XIAP ligands with Electrostatic Complementarity maps and scores

Abstract

Electrostatic Complementarity™ maps implemented in Flare™,1 Cresset’s structure-based design application, were used to investigate the protein-ligand electrostatic interactions and the Structure-Activity Relationship (SAR) of a small set of inhibitors of the X-linked IAP (XIAP)-caspase protein-protein interaction. A good correlation was also obtained between XIAP-BIR3 affinity and the Electrostatic Complementarity scores for the same data set.

Introduction

Inhibitor of apoptosis proteins (IAPs) are key regulators of antiapoptotic and pro-survival signaling pathways.2-4 Their deregulation occurs in various cancers and is associated with tumor growth, resistance to treatment and poor prognosis. This makes them an attractive target for anticancer drug discovery.5-7  The best characterized IAP, X-linked IAP (XIAP), exerts its antiapoptotic activity by binding and inactivation of caspases 3, 7, and 9 via its BIR domains. Disruption of the protein-protein interaction (PPI) between XIAP-BIR domains and caspases via small molecules is a promising strategy to inhibit XIAP. However, drugging PPIs can be particularly challenging due to their unusual binding interfaces, which are unlike classical binding sites generally flat and large.8

A recent paper from Astex9 reports that the XIAP-BIR3 activity of the small dataset of antagonists in Table 1 is increased by the introduction of electron-withdrawing substituents on the indoline ring, and shows a nice correlation between the XIAP-BIR3 pIC50 and Hammett’s σp.

In this case study, we used the Electrostatic Complementarity maps available in Flare to investigate the protein-ligand electrostatic interactions and the SAR of the molecules in Table 1. Electrostatic Complementarity scores calculated with Flare were used to quantitatively model XIAP-BIR3 pIC50.

Table 1. XIAP-BIR3 affinity of C-6 substituted indolines.8

Method

Protein preparation

The 5C7A ligand-protein complex was downloaded from the Protein Data Bank into Flare and prepared using the Build Model10 tool from BioMolTech,11 to add hydrogen atoms, optimize hydrogen bonds, remove atomic clashes and assign optimal protonation states to the protein structure. Any truncated protein chain was capped as part of protein preparation. The binding site was visually inspected to check for correct protonation states of ligands and amino acid side chains and re-optimize water orientations of suboptimal water hydrogen bonding networks. We chose to keep only water molecules in and close to the binding site that have at least 2 hydrogen bonding contacts to the protein or at least 1 hydrogen bond to ligand and protein for electrostatic complementarity calculations. As many of the modeled binding modes (e.g., compounds 9, 11, 15, 16) clash with the flexible side chain of Lys297 (Figure 1), the side chain atoms were minimized with the XED force field12 for each ligand. The resulting receptors were used to compute the electrostatic complementarity of the respective compounds.


Figure 1. The PDB: 5C7A ligand-protein complex.

Data set construction

The compounds in Table 1 were drawn using the molecule editor in Flare, starting from the crystal structure of the ligand in PDB:5C7A (compound 7 in Table 1). The 11 compounds were then aligned in Forge13 to the 5C7A ligand, using a Maximum Common Substructure alignment to minimize the conformational noise in the common indoline-piperazine scaffold.

Electrostatic Complementarity surfaces and scores

Electrostatic Complementarity maps and scoring functions are an extension of Flare’s Protein Interaction Potentials based on Cresset’s polarizable XED force field. In contrast to classical force fields that rely on atom-centered charges, XED enables description of anisotropic charge distribution around atoms which is usually only possible with ab initio approaches. Polarization effects and description of atomic charge anisotropy are especially useful for computing electrostatic properties of aromatic or unsaturated hydrocarbons, sp2 hybridized oxygen atoms, sp or sp2 hybridized nitrogen atoms, and aromatic halogens (sigma hole of Cl, Br, and I).14-16

To calculate the Electrostatic Complementarity map for a ligand towards a protein of interest, the solvent-accessible surface is first placed over the ligand. A calculation of electrostatic potentials due to the ligand and the protein is then carried out at each vertex on the surface.

These potentials are then scaled, added together, and normalized to yield the Electrostatic Complementarity score. Perfect electrostatic complementarity means that at each vertex point the ligand electrostatic potential value is paired with a protein electrostatic potential value of the same magnitude with reverse sign. Regions of the ligand surface where there is electrostatic complementarity with the protein are colored green, while the regions where there is a electrostatic clash are colored red. A more detailed description of the electrostatic potential and complementarity methodology will be presented elsewhere.17

The Electrostatic Complementarity scores quantify the ligand-protein electrostatic complementarity with three different metrics suitable for diverse protein-ligand scenarios.

The first computed score (‘Complementarity’) is the normalized surface integral of the complementarity score over the surface of the ligand (effectively the average value of that score over the surface of the ligand).

The other two scores (‘Complementarity r’ and ‘Complementarity rho’) are the Pearson’s correlation coefficient and the Spearman rank correlation coefficient, respectively, which are computed on the raw ligand and protein electrostatic potentials sampled on the surface vertices.

All three measures range from 1 (perfect complementarity) to -1 (perfect clash) but have different characteristics. The Complementarity score includes some compensation for desolvation effects, and so may be more robust when these are significant. The Pearson and Spearman correlation coefficients can provide a better indication of ligand activity in some cases, but are more susceptible to noise (r more than rho). The Spearman’s rho number is more robust against background electric fields, which may be useful if the computed protein electric potential is being biased by a large net charge on the protein.

The calculation is fast and predictive: scoring a hundred ligands normally takes less than a couple of minutes on an average laptop and gives important insights into protein-ligand electrostatics, which typically correlate with compound activity.

Mapping the electrostatics of the XIAP active site

The Electrostatic Complementarity map of compound 7 in the XIAP active site (PDB: 5C7A, Figure 2 – left) shows a strong electrostatic clash (red) in the region above the indoline ring. This is caused by an area of negative electrostatic potential in the protein’s active site, generated by the backbone carbonyl of Gly306 and the phenolic oxygen of Tyr324 (Figure 2 – middle), clashing with the negative electrostatic field associated with the indoline ring (Figure 2 – right). A less pronounced electrostatic clash can be seen between the positive electrostatic field of the protonated side chain of Lys297 (Figure 2 – middle) and the positive electrostatic field of the sigma hydrogens of the indoline ring (Figure 2 – right).

According to this map (and in agreement with the reported correlation8), electron-withdrawing substituents which make the indoline ring less electron-rich are expected to increase XIAP binding. Substituents associated with a more negative (or less positive) electrostatic field, favoring the interaction with the protonated side chain of Lys297, should also be beneficial.


Figure 2. Left: Electrostatic Complementarity map for the PDB:5C7A ligand (green: good complementarity; red: electrostatic clash). Middle: protein electrostatic potential map for PDB:5C7A (red: positive; cyan: negative). Right: ligand fields for the ligand in PDB:5C7A (red: positive; cyan: negative.

Electrostatic Complementarity and XIAP SAR

Figure 3 shows the Electrostatic Complementarity maps for the compounds in Table 1, shown in order of increasing XIAP-BIR3 activity from left to right.

A clear trend can be observed as we move from the electron-donating substituents (-NH2, -OMe), to the electron-withdrawing substituents -F, -Cl, -SO2Me. These make the indoline ring less electron-rich, reducing the clash with the negative electrostatic of the XIAP active site.


Figure 3. Electrostatic complementarity maps for some of the ligands in Table 1 (green: good complementarity; red: electrostatic clash).

 The substituents for the three most potent compounds are also associated with a negative ligand field of their own (Figure 4), favoring the interaction with the protonated side chain of Lys297, according to our initial hypothesis.


Figure 4. Negative ligand fields (cyan) for compounds 17, 15 and 16.

 These qualitative observations are confirmed by the nice correlation (r2 = 0.671) between XIAP-BIR3 pIC50 and the ‘Complementarity rho’ score shown in Figure 5.


Figure 5. Plot of XIAP-BIR3 pIC50 versus Complementarity rho.

Electrostatic Complementarity scores and MW

We monitored the correlation between MW and XIAP-BIR3 affinity/Complementarity rho to verify whether the Electrostatic Complementarity scores provide information which goes beyond the use of simple physico-chemical descriptors for drug design.

The correlation between MW and XIAP-BIR3 pIC50 (r2 = 0.613, Figure 6 – left), would possibly point towards a space filling effect as the simplest explanation of the changes in XIAP affinity in this data set.

However, the low correlation between Complementarity rho and MW (Figure 6 – right) confirms that the Electrostatic Complementarity scores are size independent.

Using the Electrostatic Complementarity scores for quantitative SAR modeling, therefore, generates trends completely independent from size effects.

Furthermore, Electrostatic Complementarity maps provide visual insight into ligand-protein binding and SAR which cannot be derived from traditional, simple physico-chemical descriptors such as MW and Hammett’s σp, thus providing invaluable information for drug design.


Figure 6. Left: Plot of XIAP-BIR3 pIC50 versus MW. Right: Plot of Complementarity rho versus MW.

Conclusions

Application of Electrostatic Complementarity to a reported XIAP-BIR3 data set showed that our method can detect and quantify electrostatic differences in XIAP ligands that cause changes in bioactivity. Electrostatic Complementarity scores and maps in Flare V2, based on Cresset’s polarizable XED force field, provide rapid activity prediction with visual feedback on new molecule designs. They provide useful information for understanding ligand binding and SAR and can be used for rapidly ranking of new molecule designs.

References and Links

  1. https://www.cresset-group.com/flare
  2. Salvesen, G. S. et al., Rev. Mol. Cell Biol. 2002, 3 (6), 401-10
  3. Gyrd-Hansen et al., Nat. Cancer 2010, 10 (8), 561-74
  4. Silke, J. et al., Cold Spring Harbor Perspect. Biol. 2013, 5 (2), a008730
  5. I et al., Clin. Cancer Res. 2004, 10 (11), 3737-3744
  6. Mizutani, Y. et al., Int. J. Oncol. 2007, 30 (4), 919-925
  7. Fulda S. et al., Nat. Rev. Drug Discovery 2012, 11 (2), 109 -124
  8. Arkin, M. R. et al., Chem. Biol. 2014, 21 (9), 1102-1114
  9. Chessari, G. et al., J. Med. 2015, 58 (16), 6574-6588
  10. V. Stroganov et al., Proteins 2011, 79 (9), 2693-2710
  11. https://www.biomoltech.com/
  12. https://www.cresset-group.com/science/field-technology/calculating-field-patterns/
  13. https://www.cresset-group.com/products/forge/
  14. Vinter, J. G., Comput. Aided Mol. Des. 1994, 8 (6), 653–668
  15. Vinter, J. G., Comput. Aided Mol. Des. 1996, 10 (5), 417–426
  16. Chessari, G. et al., Chem. Eur. J. 2002, 8 (13), 2860–2867
  17. Bauer, M. R. & Mackey, M. D. et al., manuscript in preparation

Rapid and accessible in silico macrocycle design

Abstract

Macrocyclization of pharmaceutical compounds plays an increasing role in drug discovery. Macrocycles can provide several advantages such as favorable drug-like properties, increased selectivity and improved binding affinity.

The assessment of potential macrocyclizations is a challenging computational problem. Linkers should be synthetically feasible, not too flexible, have a sensible conformation and be pharmacophorically compatible with the active site. Few computational strategies are available for this task, so most design is done ad hoc.

Here we present a modification of our existing bioisostere searching tool, Spark, to this problem. Traditional bioisostere searches specify a fragment to replace in the starting ligand, and look for similar fragments in a database. Instead, Spark assesses bioisosteric replacements in product space, which allows more complex experiments. A modification of the Spark scoring method to include similarity to other ligands known to bind in the region of the linker, as well as constraints from the protein active site and known pharmacophoric requirements, allows sensible ranking of potential linkers.

Here we present case studies of designing macrocyclization strategies for BRD4 and for Orexin 2. The Spark algorithms enable a rapid assessment of the ideal linker length and suggested chemistry for each cyclization option.

See the ‘Rapid and accessible in silico macrocycle design’ presentation Rob Scoffin presented in the COMP Division at the 256th ACS National Meeting.