LapDDPM: Spectral Perturbation Diffusion for Robust Single-Cell Manifold Generation
cs.LG, cs.AI, q-bio.BM, q-bio.CB, q-bio.GN
Submitted: 2025-06-16
Updated: 2026-09-25
Comments: LapDDPM is a novel conditional graph diffusion model for scRNA-seq generation. Leveraging spectral adversarial perturbations, it ensures robustness and yields high-fidelity, biologically plausible, and cell-type-specific samples for complex data. Proceedings of Machine Learning Research 333:1 17, 2026 Conference on Health, Inference, and Learning (CHIL) 2026, Seattle, WA
Journal ref: Proceedings of Machine Learning Research 333: 1 17, 2026
License: http://creativecommons.org/licenses/by/4.0/
The gist: Generating high-fidelity and biologically plausible synthetic single-cell RNA sequencing (scRNA-seq) data is a critical challenge in computational biology, driven by the need to model
Terminology
Abstract
Generating high-fidelity and biologically plausible synthetic single-cell RNA sequencing (scRNA-seq) data is a critical challenge in computational biology, driven by the need to model high-dimensional, sparse, and non-linear cellular manifolds. Existing generative models often fail to capture the complex topology of cellular differentiation or lack robustness against technical noise and structural variability. We introduce LapDDPM, a novel conditional Graph Diffusion Probabilistic Model designed for robust manifold learning and high-fidelity generation. LapDDPM integrates graph-based inductive biases with score-based generative modeling, enhanced by a novel spectral adversarial perturbation mechanism. By systematically perturbing graph edge weights along principal spectral modes during training, our method acts as a Distributionally Robust Optimization (DRO) framework, enforcing invariance to structural noise. We further extend LapDDPM to spatial transcriptomics and multi-modal data, treating generation as a robust inverse problem on cellular graphs. Extensive experiments on diverse datasets, including PBMC3K, Dentate Gyrus, HLCA, Visium, and 10x Multiome, demonstrate that LapDDPM significantly outperforms state-of-the-art baselines in distribution matching, manifold preservation, and downstream utility, generating biologically coherent cell states.
Sources
- NICE: Non-linear Independent Components Estimation
- Explaining and Harnessing Adversarial Examples
- Multi-Modal and Multi-Attribute Generation of Single Cells with CFGen
- Robust Optimization as Data Augmentation for Large-scale Graphs
- Existence results for Schr\"odinger $p(x)$-Laplace equations involving critical growth in $\mathbb{R}^N$
- Learning Dynamics from Multicellular Graphs with Deep Neural Networks
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks