Extending Polymer Analysis and Support [ GSoC 2025 Project Perspective ]

[TRY-ER]

Hello coders and researchers,

I've been diving into MDAnalysis for molecular dynamics and simulations, and I'm impressed by the extensive support it offers for various tasks. Since my work focuses on polymers, I've encountered challenges when trying to simulate their dynamics and explored MDAnalysis for potential solutions. Currently, it provides some support for polymer analysis through the PersistenceLength class in MDAnalysis.analysis.polymer, though a few methods trigger deprecation warnings.

Converters like RDKitParser in MDAnalysis.converters can process certain PSMILES to generate polymer topologies, but they fall short when it comes to depicting polymers as weighted graph representations. Additionally, transformations in MDAnalysis.transformations don’t always align well with the repetitive nature of polymers. After exploring the codebase extensively, these are the insights I've gathered so far.

I propose developing additional analysis tools for polymers, enhancing converters to interpret polymer topologies from different formats, refining transformation configurations, and improving exception handling. I also aim to contribute to visualization setups for polymer chains.

I would greatly appreciate your feedback on how I should proceed or any suggestions for better approaches in this area.

Looking forward to your thoughts!

[orbeckst]

Hello @TRY-ER , welcome in the MDA community. Are you aiming to develop polymer-related code for GSOC 2025 or are you asking in general?

EDIT: From your subject it looks like the former, so I assume you're asking if what you're proposing is in scope for a self-proposed GSoC project. At first glance I'd say yes, that could be interesting. Starting with an analysis of what is needed, where current MDA falls short, and concrete objectives for how to enhance the code base would be good. A common question for enhancements is "Could this be written as an MDAKit?" so you should be thinking about if your changes need to be in the core library (https://github.com/MDAnalysis/mdanalysis/) or if they could be their own MDAKit.

[TRY-ER]

Yes, I am aiming to develop polymer-related code for GSoC 2025. I was also asking for other suggestions in case I have missed some details about the available toolkit for polymers!

[TRY-ER]

Hey @orbeckst ,

It should be an MDAKit. I am looking into the details in the registry. In case, we have some base components for macro-molecular simulations, we can consider them for inclusion in the core library. I am going into details of available tooling. I will provide an analysis of what is needed and a concrete objective in a couple of days.

[TRY-ER]

Hey @orbeckst ,

Should I be sharing the details of my proposed contribution here in the discussion? Is there a more convenient way to share it, like DM you or using the private channel in Discord?

[orbeckst]

We won't be using DMs or private discussions so here is a good place for a discussion. I expect that you're writing a short summary and ask questions, not that you're posting a full proposal. This would be similar to raising an issue for a new feature, which is also done in public.

When you write your pre-proposal you can also include a link to this discussion.

If you're worried about plagiarism: If we see proposals that are similar or look as if they are AI generated from a prompt we look very closely. Linking to the public discussion where you posted some parts will then help us to establish priority.

---

It should be clear, but just to be explicit: Anyone caught plagiarizing someone else's work will be rejected and also reported to the GSoC organizers.

[orbeckst]

See Submitting the Pre-Proposal for more information on the first step.

[TRY-ER]

Understood. I will share a summary of the contribution perspective in this discussion.

[TRY-ER]

• Where MDA falls short:

  • The loading of polymer to the MDA Universe is done either from an external topology and trajectory file (non-exclusive to polymer) or directly loaded using MDAnalysisData (as PEG is loaded from an external URL using the fetch_PEG_1chain function)
  • There is a lack of a standardized inclusion process for polymers to load and process from a standard polymer representation like PSMILES or weighted directed graph (WDG) strings. (instead of the non-exclusive topology and trajectory files used right now)
  • There should be a process of directly including the polymer representations, such as PSMILES, to either generate a topology and trajectory or load the mapped files from a repository.
  • Edited: The number of repetitions of a unit monomer (degree of polymerization) should be handled properly to depict the polymer chain and repetition pattern of the unit monomer (could be of various types: random, block, alternate patterns).
  • Edited: While utilizing some representation methods like PSMILES, it is possible that the same polymer has different representations that are still chemically valid. (For Example: Poly(ethylene oxide) can be depicted as [[*]CCO[*], [*]COC[*], [*]OCC[*]]; there could be length-wise variations as well. Poly(Polyethylene) can be depicted as [[*]C[*], [*]CC[*], [*]CCC[*]]). There is a requirement to convert it to the shortest and unique representation to refer to the corresponding topology and trajectory file or generation. ( For example: Poly(Polyethylene) will be represented with [*]C[*] ) (This process of uniquely identifying a valid representation is called canonicalization)
  • Added: Even if we have the required form of polymer representation with the number of repetitions, we lack a composer to combine these units in a given pattern to form the chain.
  • The current tooling is missing the above functionalities to handle polymer-specific molecular dynamics simulations.

• Objectives to resolve the above shortcomings:

  • A wrapper kit could be developed for polymers to include polymer representations like PSMILES and weighted directed graphs to parse the chemical data along with the number of monomer repetitions (degree of polymerization).
  • Additional Objective: The unique formation of the polymer representation will be mapped with existing topology and trajectory files in the inhouse MDAnalsysisData repository.
  • Edited: We can establish a workflow for its generation by determining the canonicalization version of the input representation mapped with the number of repetitions to give a virtual molecular object of the polymer chain.
  • Edited: Computationally generating the topology and trajectory details will require to load the chemical and co-ordinate details of the resultant virtual chain utilizing rdkit and other available tools.
  • Edited: The resultant topology and trajectory will be passed along with support for other scientific metadata to form the simulation with elevated accuracy.
  • The fundamental objective is to allow standardized polymer representations like PSMILES and WDG to be implemented into MDAnalysis, such that it will be directly compatible with the other workflows implemented for the generation and analysis of polymers.

I came up with the above points by exploration and sample utilization of the tooling from MDA. In case there are other better ways to handle polymers, I am eager to know, learn, and implement them.

[TRY-ER]

Hey @micaela-matta

Great to know that you're working on this domain.

Just for exploration purposes ! Can you share some of the packages that are built on top of MDA (for polymer specific applications). It would really help me understand existing features and identify what better can be done.

[micaela-matta]

The fundamental objective is to allow standardized polymer representations like PSMILES and WDG to be implemented into MDAnalysis, such that it will be directly compatible with the other workflows implemented for the generation and analysis of polymers.

MDAnalysis is developed to deal with MD trajectories; it is agnostic to how the topologies are generated, and does not target complex system building.
Can you explain why having a PSMILES/WDG representation would be useful to analyse MD trajectories, ideally with an example of how your code would work? What other packages are you alluding to, that MDAnalysis could interface with?

[TRY-ER]

To be transparent, I have been working with Deepchem from GSoC, 2024 on polymer generative pipeline. The generations must be more than just chemically valid in order to be utilised for certain applications. The generations will go through a molecular dynamics evaluation for certain validation. The current setup does not directly consider the standard polymer representations ( PSMILES, WDG) so the streamlined operations could not be completed. Hence I thought of integrating it to handle MD directly from the polymer representations to validate the generated candidates through MDA.

[TRY-ER]

I am out for a conference, will share the details of the mock use case around mid of next week.

[TRY-ER]

The fundamental objective is to allow standardized polymer representations like PSMILES and WDG to be implemented into MDAnalysis, such that it will be directly compatible with the other workflows implemented for the generation and analysis of polymers.

MDAnalysis is developed to deal with MD trajectories; it is agnostic to how the topologies are generated, and does not target complex system building. Can you explain why having a PSMILES/WDG representation would be useful to analyse MD trajectories, ideally with an example of how your code would work? What other packages are you alluding to, that MDAnalysis could interface with?

Hey @micaela-matta,

Apologies for the late reply.

You are absolutely right that MDAnalysis is agnostic to how the topologies are generated. That’s exactly why I am focusing more on an MDAKit. PSMILES/WDG representation holds the composition details of the polymer that can be used to compose the topologies of a given candidate and use an external computation engine for trajectory file generation. But still, the process is compute-intensive for each newly generated PSMILES / WDG candidate. The very nature of these representations is that they can have different textual forms that refer to the same topology. Hence, irrespective of the topologies generated, a wrapper is necessary to generate or map the polymer representations with the corresponding universe/ topology & trajectory based on the canonicalized form and existing dataset (topology & trajectory), which can make the process efficient both in terms of time and computing (considering we use the same configuration for trajectory generation). This setup might not drastically improve the analysis of the MD trajectory for the polymers, but it will provide an initial interface to handle polymer molecular dynamics in a standardized and efficient manner. Beyond the scope of GSoC-2025, I believe there is potential to implement polymer-specific molecular simulations more easily by building upon this foundational setup.

I will be sharing a mock-up of the code out of this thread.

[TRY-ER]

Hi,

Apologies for the late response! @orbeckst @micaela-matta

The following are mock-up codes of the implementation. The code highlights the example for the user/developer end.

# implementation of the polymer topology and trajectory generation and mapping from its representations
from MDAnalysis as mda
from mdapoly import PolymerHandler
from mdapoly.mapper import PolymerTopologyMapper
from mdapoly.config import DEFAULT_CONFIG
from mdapoly.generate import get_tpr_trr

mapper = PolymerToplologyMapper(path=’path-to-the-existing-polymer-topology-and-trajectory) # in case the topology and trajectory are stored in disk

# mapper = PolymerToplologyMapper(object=UniverseCollectionObject) # in case the topology and trajectory are mapped in memory

psmiles_handler = PolymerHandler(input_type = ‘psmiles’, mapper=mapper, generator=get_tpr_trr)

sample_psmiles = [“[*]CC[*]”]

topology_blobs, trajectory_blobs = psmiles_handler.handle(source=sample_psmiles, **DEFAULT_CONFIG)

# the above line will perform following steps
# 1. Check for the validity of the entries
# 2. Form the unique representations of the entries
# 3. Pass to the mapper to find the equivalent topology and trajectory
# 	i. In case of no match, generate using the self.generator and add the generation to the mapper
# 	ii. In case there is a match, return the stored topology and trajectory
# 4. Return the topology and trajectory for the candidates


u = mda.Universe(topology_blob[0], trajectory_blob[0])

# required validation operations can be performed for the candidate from here

# Interfacing with DeepChem
from MDAnalysis as mda
from mdapoly import PolymerHandler
from mdapoly.mapper import PolymerTopologyMapper
from mdapoly.generate import get_tpr_trr
from deepchem.generator.pipelines import PolymerGeneratorPipeline, PostMDValidator

mapper = PolymerToplologyMapper(path=’path-to-the-existing-polymer-topology-and-trajectory) # in case the topology and trajectory are stored in disk

# mapper = PolymerToplologyMapper(object=UniverseCollectionObject) # in case the topology and trajectory are mapped in memory

psmiles_handler = PolymerHandler(input_type = ‘psmiles’, mapper=mapper, generator=get_tpr_trr)

post_validator = PostMDValidator(handler=psmiles_handler)

generator_pipeline = PolymerGeneratorPipeline(output_type =’psmiles’, post_validator=post_validator)

candidates, validation_outputs = generator_pipeline.generate(num_gen=10, requirements_config={})

# the above line will perform following steps
# 1. Use the generator to generate candidates iteratively
# 2. Use an internal discriminator to come up with some filtered candidates
# 3. Pass the candidates to the MDAKit for post-generation validation
# 4. Use the requirement_config with MDA Universe to check if candidates conform as expected
# 5. The pipeline returns the candidates along with the matched simulation outputs

[orbeckst]

There's a lot going on in your example. Can you explain where MDAnalysis comes in, keeping in mind that fundamentally MDAnalysis deals with point particles that have time-dependent coordinates (and perhaps velocities, forces) as well as static per-particle properties?

[TRY-ER]

I aim to utilize MDAnalysis to generate the corresponding universe for a given polymer chain based on its standard representation, number of repetitions of monomer (degree of polymerization), and additional constraints. The fundamental idea is to pick up a standard representation of the polymer, composing the point particles that depict the repetitions in the polymer chain and using additional constraints to develop the trajectory of the system. This will allow the user to address MD analysis by simply using a polymer representation with experimental constraints.

I don't propose any internal changes for MDAnalysis, but an interface for handling polymer chains from its standard representation such that it can come up with a more accurate universe depicting the repetition pattern of the polymer chain.

[orbeckst]

To me it sounds that the main issue is to generate an atomistic-like topology + coordinates from PSMILES (a bit like the RDKit converter can generate coordinates from SMILES?)... but I am not a polymer person.

I suggest you submit your pre-proposal that makes clear how your code wants to interact with MDA specifically and what, if successful, the resulting code will allow users to do that they couldn't do before. Please write in a way that is also understandable to non-polymer physicists/chemists (but don't shy away from including specific details). Ultimately, we want to understand if what you're proposing is of sufficiently immediate broad interest to the MDA project and community. (Our application evaluation criteria are also relevant for pre-proposals.)

[TRY-ER]

That sounds good. I will submit a pre-proposal soon.

[TRY-ER]

Hey @orbeckst,

I just submitted the pre-proposal using the Google form. I was expecting a submission in the form of a document. Anyway, I shared it in text form. Considering your advice to focus more on the functionality of MDAnalysis, I added details about form factor analysis of polymer in scattering systems. I also added another version of mean-distance displacement, which considers the rigid parts of the polymer as beads and operates considering coordinates at its centroid. I added details of functionalities for analysis of random force fields on the polymer chain under solvent drag and defined external trajectory.

Along with these features for trajectory analysis, I focused on an MDAKit that handles the polymer representation to set up a core ChainGroup to access these analysis features easily. Is there a way I could share the pre-proposal I prepared as a document, so it would be easy to illustrate and understand? In terms of concepts, there were equations and references that I could not include in the text form through the Google Forms response.

[TRY-ER]

I have updated the response with a linear representation of the equations. I hope it helps.

[TRY-ER]

Hey @orbeckst ,
I hope you're doing well. I just raised the PR for test coverage (GSoC initial contribution)-> 5000
Kindly give it a look and suggest the changes required.

[TRY-ER]

The changes in the PR increase the code coverage for tests in 2D visualization by 38%.

[TRY-ER]

Hi @orbeckst ,

Thank you for your guidance and support. Although I have not been selected for GSoC, I would still like to continue contributing. I will work on the polymer kit for the corresponding application on my own. At a later stage, if you evaluate it as a good contribution, we can include it in the registry.

Congratulations to everyone selected for GSoC!

[orbeckst]

It would be great to see your kit submitted to the MDAKit registry!