Required skills: All of our existing models are built for PyTorch, so experience with Python and PyTorch/Tensorflow workflows is preferred. The ability to work with datasets, such as segmenting video and generating graph visualizations is essential. An ability to build web interfaces, UI design, basic knowledge of biology, open-source practices, and applied mathematical tools will also be useful.
Time commitment: Full time (350 hours)
About: The DevoWorm group has developed an open-source Graph neural network (GNN) framework for embryogenetic data called DevoGraph. Developmental GNNs (D-GNNs) allow us to characterize a growing network that undergoes shape transformations along with increases in size. This is ultimately important for understanding formation of the connectome and the origins of embodied behavior.
Aims: For this year’s project, the successful applicant will work on extending our two outcomes from last year:
The first direction involves working with Neural Developmental Programs to build growing neural networks. This provides a means to model the function of embryogenetic networks, developing connectomes, and other growth processes
The second direction involves working with hyper graph representations, enabling multiscale modeling from a network perspective.
We aim to tie our D-GNN work into the group’s ongoing theoretical and computational work. As such, this project will require the ability to work with mathematical models and associated algorithms. Knowledge of graph and/or network theory is helpful, but not required.
What can I do before GSoC?
You can ask one of the mentors to direct you to the data source and you can start working on it. Please feel free to join the Openworm Slack or attend our meetings to raise questions/discussions regarding your approach to the problem.
GSOC 2026 project #14 : I have explored the demo site (analysim.tech) and the codebase. To better understand the project goals, could you please share some current UI/UX pain points or specific areas where the user experience needs the most improvement? Also, are there any recommended ‘good first issues’ for me to get started with the codebase?"
My name is Ashu Rajput and I’m interested in GSoC 2026 Project #6 – OpenWorm DevoWorm: DevoGraph.
I have experience working with Python, PyTorch, and machine learning. Recently I have been exploring graph neural networks and studying the DevoGraph repository along with related ideas around developmental GNNs.
From my understanding, the project focuses on modeling growing biological networks, where nodes and connections evolve over time, and extending this framework using approaches such as neural developmental programs and hypergraph representations for multiscale modeling.
I’m currently trying to better understand how these developmental processes could be represented computationally (for example through dynamic graph construction or generative graph models) and how they might integrate with the existing DevoGraph framework.
I’m looking forward to learning more about the ongoing work and contributing to discussions around possible modeling approaches.
I’ve been reading through the NDP-HNN code. In ‘hyperedges.py’, hyperedges are constructed using KNN on raw x,y,z coordinates and lineage relationships, the model’s learned embeddings play no role in deciding which cells form a hyperedge. So the model receives hyperedges as fixed input rather than learning to generate them.
Is the direction for this year to close that gap, making hyperedge formation a learned, dynamic process driven by the model’s own representations rather than hardcoded geometry? And if so, would the NDP shared program be the right place to add that decision, or would it live somewhere else in the architecture?
Project #6 caught my attention because it lines up closely with what I’ve been working on. I’m currently interning at Chile, modelling cancer cell migration using physics-informed neural surrogates and an SDE–PDE coupled framework, so the idea of representing biological growth as an evolving graph feels like a very natural extension of that.
I have experience with PyTorch and I’m fairly comfortable with time-series biological datasets and mathematical modelling. Would love to know more about the data sources being used this year and where things currently stand with Neural Developmental Programs and hypergraph extensions. Joining the OpenWorm Slack as well.
I’m Tenghai Long, currently a OMSCS student at Georgia Tech (AI/ML track), taking Deep Learning and Deterministic Optimization this semester. Before this, I did a PhD in Systems Neuroscience working on primate neural recording for spatial cognition and working memory. After that I worked as a Research Engineer developing deep learning algorithms for EEG-based clinical monitoring.
I’ve been interested in the OpenWorm project for a while, and DevoGraph feels like a natural entry point for me given my background in both neuroscience and deep learning. I’m comfortable with PyTorch and have a solid grounding in neural circuit development and network modeling, so I expect to ramp up quickly.
I’d love to get pointed to the data sources to start exploring, and would also like to join the OpenWorm Slack and attend meetings if possible.
I’m Deepesh, a pre-final year BTech student currently building an Ontogenetic Intelligence System a computational model of developmental intelligence grounded in predictive coding, active inference, and complementary learning systems.
The current D-GNN framework models topological changes in a growing network, but the growth rules themselves appear to be externally specified rather than learned from developmental dynamics. In biological connectome formation, growth is driven by activity-dependent processes axons extend toward regions of correlated activity, synapses strengthen through Hebbian-like mechanisms.
The question I would like to raise: is the NDP direction this year exploring activity-dependent graph growth where the network’s own firing patterns determine which edges form rather than purely geometry-driven construction? And if so, how are you thinking about representing the temporal dynamics of that process within the hypergraph framework?
My OIS work gives me direct grounding in hierarchical generative models and temporal prediction over developing systems. I would like to understand where the current data sources stand and will join the OpenWorm Slack this week.