Dr. Alex M. Clark
In a previous article we described how Collaborative Drug Discovery is using a recently unearthed feature to encode monomers in specification-compliant V3000 molfiles. We covered methods for creating linear & cyclic peptides, nucleotides, mixed hybrids of chemistry and biochemistry, antibodies and antibody drug conjugates. In this article we will cover some of the improvements and new developments including robust custom monomer management functionality and the ability to bulk-create siRNA entities.
A large proportion of the macromolecule functionality is based on the ability to define custom monomers: non-natural amino acids, unnatural nucleotides (bases/sugars/phosphates), post-translational modifications, and arbitrary chemical fragments. While our chosen datastructure (V3000 self-contained sequence representation) embeds the chemical structures and connection patterns for each of the monomers, they still have to come from somewhere when we create these molecular entities.
CDD Vault offers two ways to create chemically aware biologics: either using Ketcher to draw them one at a time with essentially no limits on what you can put together; or using our special import composers to batch-convert precursor sequence tokens to a rich information-dense representation. Our monomer management tools have the following properties:
A monomer is a chemical structure with additional properties:
It is very common for monomers to be made available from vendors and these are often supplied as structures that depict the delivered chemical substance. This will almost always be just a regular molecule with no markup to indicate the leaving groups, and they quite often have protecting groups which will be cleaved off before the core monomer unit is incorporated into a macromolecule. Because it is useful to record both the thing-in-the-bottle and the marked up monomer unit, we decided to extend the CDD Vault molecule datastructure so that it can capture the monomer definition as well as the precursor structure.
Turning a molecule into a monomer can be very simple. In the example below (Figure A) the precursor structure for a fluorinated derivative of phenylalanine has been registered, and a single click initiates the conversion into a monomer:
Figure A
There is quite a bit of activity going on under the hood, which can become quite sophisticated. In this case the algorithms were easily able to identify that it is a peptide unit with a hydrogen (R1) leaving group at the N-terminus and a hydroxyl (R2) leaving group at the C-terminus.
Building one monomer at a time is fine when you just have a few, or when you're creating your own original molecules. For importing a whole catalog the best strategy is to use the monomer import composer, as shown in Figure B:
Figure B
The figure shows columns where the incoming molecular structure is shown alongside a graphical representation of the fully marked up monomer. These are examples of amino acids that you can buy from catalogs, and they are typically delivered with an accompanying SDfile (or SMILES) representation. The chemical structures indicate the thing that is in the container, and often these need extra pre-processing steps before they can be marked up as the monomer repeat unit that is needed for building up macromolecules. Commercially available amino acids often arrive with protecting groups or as salts. The bulk monomer importer has options to automatically disconnect these and convert the structure to its "raw" form prior to continuing with the markup. The importer can also orient the structures so that the amine is on the left and acid on the right, which is important for perception, and especially useful when the incoming structures have not been carefully drawn out according to orientation conventions.
Last but not least the natural analog will be guessed by default, unless the input content is introduced with a column indicating what it should be. Knowing the natural analog is particularly useful for sequence searching and large scale bioinformatics analyses.
Things can get tricky with nucleotides, especially when an incoming unit is represented as a block of 3 units (base, sugar, phosphate) in its raw chemical form. For a custom nucleotide that is closely related to the natural DNA/RNA building blocks it is usually easy to tell the difference between the base, sugar and phosphate. There are however a lot of synthetic analogs where this is far from obvious: sometimes the scaffold that takes the place of the sugar is quite different, and the phosphate and base can take many different forms. To make matters worse, commercially available nucleotide fragments can have the phosphate attached to either the 5' or 3' position on the sugar, which can be hard to ascertain if the sugar isn't the canonical ribose structure. For this reason we put in a considerable amount of effort deriving recognition patterns and fine tuning our algorithms on a large training set of vendor-supplied nucleotides. As shown in Figure C, most of the time the automatic characterization works rather well:
Figure C
There are however edge cases where explicit fine-grained control over monomer definition is needed, and this can be achieved using Ketcher. Also, when you are already drawing out a macromolecule with Ketcher and you need to define a new monomer, you can do that within the editor itself, as shown below in Figure D:
Figure D
In this case a nucleotide is presented with all three monomers in one structure. While the base is naturally occurring adenine, the sugar is based on a 6-membered ring rather than the natural 5. In the sugar position that is analogous to the 5' side, a triphosphate is present. In case the automatic recognition algorithms don't get everything right, the Ketcher functionality can be used to ensure that everything is defined exactly the way it should be.
The interface provides very explicit control over the leaving group definitions, the classification of each block, and metadata about the monomers. This feature exists in standalone deployments of Ketcher but there is a difference: when you save the monomer you have a choice. You can just keep using the monomer transiently in the current editing session, which means that the monomer definition will be saved within the structure you are editing, but otherwise it will evaporate once the Ketcher window is closed. The other option is to register it in CDD Vault which means that it will be persistently available to all subsequent editing sessions, and you have access to all of the molecule management functionality for editing, sharing, deleting, etc.
Peptides can be imported in bulk by composing text sequences into either linear or cyclic peptides containing both natural and synthetic amino acids. This process has been complemented by functionality to make it more powerful: chemical fragments can be referenced inline, and connections between both terminal backbone connections and sidechains can be incorporated.
Figure E
In example (a) the linear peptide sequence is acetylated at the N-terminus and amidated at the C-terminus, both of which are common peptide modifications, in addition to a functionalized sidechain. Example (b) shows a peptide that is heavily modified with bulky fragments attached to the side chains. The composition syntax for both of these cases uses inline SMILES strings. The layout for the chemical fragments and the hybrid macromolecule is all done automatically, which means that this import process is a viable alternative to drawing out each structure individually, and can be done at scale. Example (c) shows a cyclic peptide that has been bridged through two cysteine groups, creating a bicyclic ring.
Since the initial release of the macromolecule composer feature provided the ability to create single- or double-stranded sequences of RNA/DNA from incoming strings, we added a second nucleotide category which allows double-stranded oligonucleotides to be created with a much higher degree of control upon the composition. The molecular entity class is referred to as siRNA (aka silencing RNA or small interfering RNA) but it refers to the general category of related biomolecules.
These double-stranded macromolecules frequently differ from natural sense/antisense strands in several important ways:
Figure F
These structures are quite time consuming to draw, and most publications use cartoon imagery to convey their composition. This is a reasonable approach for human-to-human communication but more analyses can be performed upon visual diagrams that are underwritten with the bioinformatics and cheminformatics definition. The macromolecule format used by CDD Vault is ideal for this purpose, so we just need a way to create the content.
One way is to draw them manually using Ketcher, but if you want to create them in bulk, or just prefer to script it using text input, we have a new kind of composer for this purpose. The siRNA composer needs two input strands, one each for the sense and antisense chains which are made up of nucleotides, and optionally chemical fragments at the end. In addition to the definition of these strands, the importer needs to know which bases are paired up together.
Figure G
The strand composition and base pairings are all that is needed: the rest of the layout is automatic. The two strands are lined up so that hydrogen-bonded bases are stacked vertically. Hairpins and scrunchies are arranged in a partially circular layout, while un-paired bases are shifted away from each other for visual effect.
The result is a data object that satisfies the three primary requirements: it can be rendered in publication-ready form, it captures the biochemistry, and the whole thing can be reduced to individual atoms and bonds without any guesswork.
Our import composer tools are designed to take a prototype line notation (a sequence of tokens) and combine that with a monomer library and institutional knowledge about what the tokens are supposed to represent, and then combine this knowledge together to create a self-contained representation that looks great and stores everything in a single file that can be shared anywhere without information loss.
There are other line notations for macromolecules which encode more metadata. Two of these are HELM (Hierarchical Editing Language for Macromolecules) and IDT (from Integrated DNA Technologies). Both of these line notations differ from the V3000 SCSR template functionality in that they are tied to global dictionaries for monomer definitions. And like all line notations for molecular content, human-friendly display is predicated on having an effective layout algorithm.
The Ketcher team has put considerable effort into creating importers for both of these formats, and we have included pass-through functionality in CDD Vault, so you can import HELM and IDT strings and upconvert them into monomer-containing molfiles. Like any other ways of creating chemically aware biologics they can be edited after the fact, rendered as-is, and shared with anyone else who has software that implements the V3000 functionality.
Figure H
One caveat is that the Ketcher team has enumerated a long list of common monomer codes for use by these line notations. If your incoming content is operating within the boundaries of common synthetic peptides/nucleotides and fragments then it will likely just work.
The composition tools that we created for assembling peptides, nucleotides, siRNA and antibodies were originally designed for bulk import, but there are scenarios where the ability to create a single molecular entity from text, custom monomers and other options is useful. This allows you to have all the specific molecular information in an interactive context:
Figure I
The above example (Figure I) shows an attempt to represent a schematic representation of Patisiran (1) using the chemically aware biologics pattern. The dialog shown in (2) has the base code identities entered for the two strands, both in 5' to 3' order. The diagram shown underneath is updated interactively.
In (3) the structure has been registered as a CDD Vault molecule, but there is an oversight: the actual structure of the therapeutic entity includes non-standard functionalizations. Some of the sugars are modified and some of the thymine sugars are deoxygenated. This can be fixed by re-opening the single instance composer, shown in (4). The underlying algorithm will re-derive the sequence inputs from the structure and fill in the fields so you can pick up where you left off. In this case, the definitions of some of the sequence tokens are updated, and the corrected structure can be saved.
The chosen format for representing macromolecules (V3000 molfiles) has a long list of positive characteristics, one of them being that the chemical definitions of each monomer template are embedded within the molfile itself, making it completely portable because it does not depend on a global dictionary.
The benefit of not having a global dictionary is that the definition of the macromolecule doesn't change when the dictionary changes.
The down side of not having a global dictionary is that the definition of the macromolecule doesn't change when the dictionary changes.
Sometimes we realize that we have created a whole series of macromolecules, but there's a mistake: a common one is that a monomer was drawn wrong, which means the underlying chemical definition of each of the macromolecules is wrong. Fixing one mistake can be quite a tedious endeavor.
We provide some assistance with the Monomer Care feature, which is an optional report column that generates a schematic view of each macromolecule, highlighting problems:
Figure J
These problems are often fixable, e.g. the encoded monomer from the structure is different from the one in your library, and can be substituted out. There is a button for applying the correction: it operates one at a time rather than in bulk, but it is still substantially more convenient than fixing up the underlying structure by other means.
A fundamental benefit of encoding chemically aware biologics using V3000 templates is that the bioinformatics layer is encoded along with the chemistry and visual aesthetics. For purposes of analyzing the bioinformatics layer using conventional sequence-based tools the information first has to be extracted out from the monomer graph. Getting the information out is usually straightforward from a software engineering point of view, but it is important to keep in mind that the variety of shapes and connectivity that is allowed is endless (e.g. multiple chains, rings, junctions, hybrid types).
A lot of the shapes can't be represented by a sequence of tokens - but we can at least try. For example, we can represent a cyclic peptide by starting at the monomer designated with id 1 and pulling out the chain; we can process an siRNA entity by extracting both the chains and presenting them with their linear base identities, separated by dot ("."). Monomers that are neither peptides nor nucleotides are usually not assigned a sequence ID, so these are left out.
Because most kinds of sequence analysis tools assume natural amino acids and nucleotide bases it makes sense to extract two versions of the sequence: one with just the natural analogs, and another using the monomer codes for the monomers. It is also useful to categorize the sequence and make that available to external tools (e.g. whether it refers to a peptide or nucleotide; whether it is linear or cyclic; or that the 4 chains are actually an antibody).
When a macromolecule is registered in CDD Vault, we extract out this information, or at least the closest approximation in the instance of more exotic structures.
Figure K
This information is cached and stored within the molecule record and treated analogously to molecules that are registered using just sequence information: this enables searching, property calculation, and exporting of raw sequence data.
The chemically aware biologics project has focused initially on content creation, since the V3000 template representation is effectively novel for biologics: despite having existed for a long time there was neither a pre-existing corpus of data nor a vibrant ecosystem of tools. This is the opposite of the small molecule use case, and so we have had to build and sponsor tools to get scientists on board with this highly effective method of capturing biomolecule data. Now that we have effective methods for creating general purpose and domain-specific biologics and managing custom monomers, we are starting to turn our attention toward tooling for analyzing this content.
Extraction of sequence information is an important first step but this is mainly to ensure feature parity with simpler methods of describing peptides and nucleotides (i.e. a single line with a list of tokens). There are a number of projects that we have on the drawing board, including but not limited to:
In addition to the functionality that we are continuing to build for CDD Vault we are committed to continuing our support of the Ketcher project as it moves forward with more and better functionality for macromolecules. CDD Vault is rarely more than a few weeks behind the latest and greatest release of Ketcher, and our continued feedback influences the priorities of the Ketcher team. Similarly CDD Vault contributes code and keeps up-to-date with the latest advances from the RDKit community.
Model peptides, oligos, ADCs, and more in an environment built for collaborative research.
See how CDD Vault can help your team move faster, book a demo.