AI Paper Atlas

Research map / Graph learning / Graph self-supervised learning

TetraGT: Tetrahedral Geometry-Driven Explicit Token Interactions with Graph Transformer for Molecular Representation Learning

Jinjia Feng, Zhewei Wei, Taifeng Wang, Zongyang Qiu

ICLR 2026 · Graph learning › Graph self-supervised learning · approach: Molecular graph pretraining

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 changedWhereMeasuredEffect
add tetrahedral interaction module (ablation)TetraGT 12-layer · PCQM4Mv2 valid (12-layer ablation)MAE (meV): 73.6 -> 71.0Improved
add directed cycle angle loss (ablation)TetraGT 12-layer · PCQM4Mv2 valid (12-layer ablation)MAE (meV): 71.0 -> 70.6Improved
add hierarchical virtual node (ratio 1:1) (ablation)TetraGT 12-layer · PCQM4Mv2 valid (12-layer ablation)MAE (meV): 70.6 -> 70.2Improved
raise distance:angle loss ratio from 1:1 to 4:1 (ablation)TetraGT 12-layer · PCQM4Mv2 valid (12-layer ablation)MAE (meV): 70.2 -> 68.8Improved
raise distance:angle loss ratio to 8:1 (ablation)TetraGT 12-layer · PCQM4Mv2 valid (12-layer ablation)MAE (meV): 68.8 -> 70.1Hurt
tetrahedral attention instead of full attention for angles (ablation)TetraGT conformer predictor · PCQM4Mv2 validation-3Ddistance cross-entropy: 1.179 -> 1.125Improved
tetrahedral attention instead of axial attention (ablation)TetraGT conformer predictor · PCQM4Mv2 validation-3Dangle cross-entropy: 1.310 -> 1.231Improved
tetrahedral attention instead of full attention (training time per epoch) (ablation)TetraGT conformer predictor · PCQM4Mv2 validation-3Dtime/epoch: 1.43 -> 1.12Improved
increase depth from 12 to 24 layers (ablation)TetraGT · PCQM4Mv2 validMAE (meV): 68.1 -> 67.1Improved
add RDKit conformers as inputTetraGT 24-layer · PCQM4Mv2 validMAE (meV): 67.1 -> 65.9Improved
replace TGT with TetraGT conformer predictorconformer prediction · PCQM4Mv2 validation-3Dtorsion angle MAE (rad): 0.563 -> 0.306Improved
replace TGT with TetraGTchirality prediction · PCQM4M train chirality evalchirality accuracy (%): 32.5 -> 76.7Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy