Do we need equivariant models for molecule generation?
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908448691585024 |
|---|---|
| author | Nowara, Ewa M. Rackers, Joshua Suriana, Patricia Kessel, Pan Shen, Max Watkins, Andrew Martin Maser, Michael |
| author_facet | Nowara, Ewa M. Rackers, Joshua Suriana, Patricia Kessel, Pan Shen, Max Watkins, Andrew Martin Maser, Michael |
| contents | Deep generative models are increasingly used for molecular discovery, with most recent approaches relying on equivariant graph neural networks (GNNs) under the assumption that explicit equivariance is essential for generating high-quality 3D molecules. However, these models are complex, difficult to train, and scale poorly.
We investigate whether non-equivariant convolutional neural networks (CNNs) trained with rotation augmentations can learn equivariance and match the performance of equivariant models. We derive a loss decomposition that separates prediction error from equivariance error, and evaluate how model size, dataset size, and training duration affect performance across denoising, molecule generation, and property prediction. To our knowledge, this is the first study to analyze learned equivariance in generative tasks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_09753 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Do we need equivariant models for molecule generation? Nowara, Ewa M. Rackers, Joshua Suriana, Patricia Kessel, Pan Shen, Max Watkins, Andrew Martin Maser, Michael Machine Learning Quantitative Methods Deep generative models are increasingly used for molecular discovery, with most recent approaches relying on equivariant graph neural networks (GNNs) under the assumption that explicit equivariance is essential for generating high-quality 3D molecules. However, these models are complex, difficult to train, and scale poorly. We investigate whether non-equivariant convolutional neural networks (CNNs) trained with rotation augmentations can learn equivariance and match the performance of equivariant models. We derive a loss decomposition that separates prediction error from equivariance error, and evaluate how model size, dataset size, and training duration affect performance across denoising, molecule generation, and property prediction. To our knowledge, this is the first study to analyze learned equivariance in generative tasks. |
| title | Do we need equivariant models for molecule generation? |
| topic | Machine Learning Quantitative Methods |
| url | https://arxiv.org/abs/2507.09753 |