Research map / Probabilistic and scientific modeling / Molecules and proteins
ATOM: A Pretrained Neural Operator for Multitask Molecular Dynamics
Luke Thompson, Davy Guan, Slade Matthews, Dai Shi, Junbin Gao, Andi Han
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add equivariant lifting layer instead of standard linear layers (ablation) | ATOM · MD17 Aspirin | S2T MSE: 29.40 -> 6.92 | Improved |
| use quasi-equivariance (lifting only) instead of fully equivariant blocks (ablation) | ATOM · MD17 Aspirin | MSE: 10.25 -> 6.77 | Improved |
| use heterogeneous cross-attention instead of standard self-attention (ablation) | ATOM · MD17 Aspirin | S2S MSE: 7.39 -> 6.92 | Improved |
| add T-RoPE under stochastic Delta T (ablation) | ATOM · single-task with stochastic Delta T | MSE: 5.262 -> 4.197 | Improved |
| add T-RoPE under fixed Delta T (ablation) | ATOM · single-task fixed Delta T | MSE | No clear effect |
| add label noise regularization (ablation) | ATOM · MD17 Aspirin | S2T MSE: 8.13 -> 6.92 | Improved |
| add random-walk positional encoding for multitask pretraining (ablation) | ATOM · TG80 Cluster 1 | MSE | Improved |
| use fully connected point-cloud attention instead of GATv2 on bond graph (ablation) | ATOM · MD22 | MSE | Improved |
| multitask pretraining on TG80 versus EGNO, out-of-domain | ATOM vs EGNO · TG80 OOD S2T | S2T MSE: 39.74% average improvement | Improved |
| multitask pretraining on TG80 versus EGNO, in-distribution | ATOM vs EGNO · TG80 ID S2T | S2T MSE: 83.96% average improvement | Improved |
| train single model on all data versus single-task ATOM (ablation) | ATOM · TG80 Formic Acid S2S | MSE: 26.40 -> 22.39 | Improved |
| train single model on all data versus single-task ATOM (ablation) | ATOM · TG80 Formic Acid S2T | MSE: 18.10 -> 18.72 | Hurt |
Closest papers by meaning
- UniSim: A Unified Simulator for Time-Coarsened Dynamics of Biomolecules (ICML 2025 · similarity 0.90)
- All-atom Diffusion Transformers: Unified generative modelling of molecules and materials (ICML 2025 · similarity 0.89)
- Beyond Atoms: Enhancing Molecular Pretrained Representations with 3D Space Modeling (ICML 2025 · similarity 0.88)
- FlashMD: long-stride, universal prediction of molecular dynamics (NeurIPS 2025 · similarity 0.88)
- Learning Over Molecular Conformer Ensembles: Datasets and Benchmarks (ICLR 2024 · similarity 0.88)
- WGFormer: An SE(3)-Transformer Driven by Wasserstein Gradient Flows for Molecular Ground-State Conformation Prediction (ICML 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
