![[2512.18454] Out-of-Distribution Detection in Molecular Complexes via Diffusion Models for Irregular Graphs](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2512.18454] Out-of-Distribution Detection in Molecular Complexes via Diffusion Models for Irregular Graphs
Summary
This paper presents a novel framework for out-of-distribution (OOD) detection in molecular complexes using diffusion models tailored for irregular graphs, enhancing predictive reliability in machine learning applications.
Why It Matters
The ability to accurately detect OOD data is crucial for deploying machine learning models in real-world scenarios, particularly in fields like bioinformatics where molecular data can be complex and varied. This research addresses a significant gap in current methodologies, offering a robust solution that could improve model reliability and safety.
Key Takeaways
- Introduces a probabilistic OOD detection framework for irregular 3D graphs.
- Utilizes a unified continuous diffusion model for categorical and continuous data.
- Demonstrates strong correlation between OOD likelihoods and prediction errors in protein-ligand complexes.
- Offers a label-free quantification workflow for geometric deep learning.
- Enhances detection sensitivity through multi-scale trajectory statistics.
Computer Science > Machine Learning arXiv:2512.18454 (cs) [Submitted on 20 Dec 2025 (v1), last revised 18 Feb 2026 (this version, v2)] Title:Out-of-Distribution Detection in Molecular Complexes via Diffusion Models for Irregular Graphs Authors:David Graber, Victor Armegioiu, Rebecca Buller, Siddhartha Mishra View a PDF of the paper titled Out-of-Distribution Detection in Molecular Complexes via Diffusion Models for Irregular Graphs, by David Graber and 3 other authors View PDF HTML (experimental) Abstract:Predictive machine learning models generally excel on in-distribution data, but their performance degrades on out-of-distribution (OOD) inputs. Reliable deployment therefore requires robust OOD detection, yet this is particularly challenging for irregular 3D graphs that combine continuous geometry with categorical identities and are unordered by construction. Here, we present a probabilistic OOD detection framework for complex 3D graph data built on a diffusion model that learns a density of the training distribution in a fully unsupervised manner. A key ingredient we introduce is a unified continuous diffusion over both 3D coordinates and discrete features: categorical identities are embedded in a continuous space and trained with cross-entropy, while the corresponding diffusion score is obtained analytically via posterior-mean interpolation from predicted class probabilities. This yields a single self-consistent probability-flow ODE (PF-ODE) that produces per-sample log...