Research map / Probabilistic and scientific modeling / Molecules and proteins
Enhancing Molecular Property Predictions by Learning from Bond Modelling and Interactions
Yunqing LIU, Yi Zhou, Wenqi Fan
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 the bond-centric graph to the atom-centric graph (ablation) | atom-centric model with torsion encoding · PCQM4Mv2 validation | MAE (meV): 75.5 -> 65.4 | Improved |
| use the atom-centric graph only instead of the bond-centric graph only (ablation) | single-graph model · PCQM4Mv2 validation | MAE (meV): 89.9 -> 77.2 | Improved |
| add torsion angle position encoding (ablation) | atom-centric graph · PCQM4Mv2 validation | MAE (meV): 77.2 -> 75.5 | Improved |
| add covalent radii prediction and structure-aware mask (ablation) | atom-centric graph · PCQM4Mv2 validation | MAE (meV): 77.2 -> 76.4 | Improved |
| add bond-atom attention (ablation) | dual graph with covalent radii, torsion, mask and bond-bond attention · PCQM4Mv2 validation | MAE (meV): 64.8 -> 60.3 | Improved |
| add structure-aware mask (ablation) | full DeMol · PCQM4Mv2 validation | MAE (meV): 61.7 -> 60.3 | Improved |
| add structure-aware mask (ablation) | dual graph without bond-atom attention · PCQM4Mv2 validation | MAE (meV): 65.4 -> 64.8 | Improved |
| add covalent radii prediction regularization (ablation) | full DeMol · PCQM4Mv2 validation | MAE (meV): 61.1 -> 60.3 | Improved |
| replace TGT-At with DeMol | PCQM4Mv2 best baseline · PCQM4Mv2 validation | MAE (eV): 0.0671 -> 0.0603 | Improved |
| replace Unimol+ with DeMol | OC20 IS2RE best baseline · OC20 IS2RE validation (average) | energy MAE (eV): 0.4088 -> 0.3879 | Improved |
| replace PaiNN with DeMol | QM9 U0, U, H · QM9 | MAE: U0 5.85 -> 6.24; U 5.83 -> 6.1; H 5.98 -> 6.35 | Hurt |
| replace TorchMD-Net with DeMol | QM9 dipole mu and R2 · QM9 | mu and R^2 MAE: mu 0.011 -> 0.024; R2 0.033 -> 0.101 | Hurt |
Closest papers by meaning
- I2Mole: Interaction-aware Invariant Molecular Learning For Generalizable Property Prediction (ICLR 2026 · similarity 0.89)
- Learning Over Molecular Conformer Ensembles: Datasets and Benchmarks (ICLR 2024 · similarity 0.88)
- TRIDENT: Tri-Modal Molecular Representation Learning with Taxonomic Annotations and Local Correspondence (NeurIPS 2025 · similarity 0.86)
- Multimodal Molecular Pretraining via Modality Blending (ICLR 2024 · similarity 0.86)
- MOL-Mamba: Enhancing Molecular Representation with Structural & Electronic Insights (AAAI 2025 · similarity 0.86)
- Benchmarking Compositional Generalisation for Machine Learning Interatomic Potentials (NeurIPS 2026 · similarity 0.85)
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.
