Graph-level molecular classification on BACE (Accuracy)
85.8AccuracyESA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ESA2024.02 | 85.8 | — | |
| DropGIN2024.02 | 82.4 | — | |
| GIN2024.02 | 82 | — | |
| GATv22024.02 | 82 | — | |
| PNA2024.02 | 81.7 | — | |
| GCN2024.02 | 81.3 | — | |
| GAT2024.02 | 81.3 | — | |
| GPS2024.02 | 80.4 | — | |
| TokenGT2024.02 | 78.6 | — | |
| Graphormer2024.02 | 75.8 | — | |
| GLIP-MistralCategory=Ours, LLM Backbone=Mistral-7B, Supervision Strategy=Semi-Supervised2026.06 | 62.63 | — | |
| TEA-GLM-MistralCategory=Graph-LLM Method, LLM Backbone=Mistral-7B, Supervision Strategy=Semi-Supervised2026.06 | 60.96 | — | |
| GLIP-QwenCategory=Ours, LLM Backbone=Qwen, Supervision Strategy=Semi-Supervised2026.06 | 59.5 | — | |
| TEA-GLM-QwenCategory=Graph-LLM Method, LLM Backbone=Qwen, Supervision Strategy=Semi-Supervised2026.06 | 58.99 | — | |
| GraphCLCategory=Graph Pretrain Method, Supervision Strategy=Semi-Supervised2026.06 | 56.48 | — | |
| DGICategory=Graph Pretrain Method, Supervision Strategy=Semi-Supervised2026.06 | 56.32 | — | |
| GLIP-MistralEvaluation Protocol=5-shot, Model Category=Ours, Backbone=Mistral2026.06 | 56.17 | — | |
| GraphGPT-QwenCategory=Graph-LLM Method, LLM Backbone=Qwen, Supervision Strategy=Semi-Supervised2026.06 | 55.76 | — | |
| GraphGPT-MistralEvaluation Protocol=5-shot, Model Category=Graph-LLM Method, Backbone=Mistral2026.06 | 55.23 | — | |
| TEA-GLM-MistralEvaluation Protocol=5-shot, Model Category=Graph-LLM Method, Backbone=Mistral2026.06 | 55.18 | — | |
| GLIP-QwenEvaluation Protocol=5-shot, Model Category=Ours, Backbone=Qwen2026.06 | 55.14 | — | |
| LLaGA-MistralEvaluation Protocol=5-shot, Model Category=Graph-LLM Method, Backbone=Mistral2026.06 | 54.95 | — | |
| DGIEvaluation Protocol=5-shot, Model Category=Graph Pretrain Method2026.06 | 54.76 | — | |
| GraphGPT-MistralCategory=Graph-LLM Method, LLM Backbone=Mistral-7B, Supervision Strategy=Semi-Supervised2026.06 | 54.76 | — | |
| LLaGA-MistralCategory=Graph-LLM Method, LLM Backbone=Mistral-7B, Supervision Strategy=Semi-Supervised2026.06 | 54.55 | — | |
| SimGRACEEvaluation Protocol=5-shot, Model Category=Graph Pretrain Method2026.06 | 54.38 | — | |
| GraphMAEEvaluation Protocol=5-shot, Model Category=Graph Pretrain Method2026.06 | 54.36 | — | |
| GraphGPT-QwenEvaluation Protocol=5-shot, Model Category=Graph-LLM Method, Backbone=Qwen2026.06 | 54.18 | — | |
| GraphMAECategory=Graph Pretrain Method, Supervision Strategy=Semi-Supervised2026.06 | 54.11 | — | |
| LLaGA-QwenEvaluation Protocol=5-shot, Model Category=Graph-LLM Method, Backbone=Qwen2026.06 | 54.07 | — | |
| SimGRACECategory=Graph Pretrain Method, Supervision Strategy=Semi-Supervised2026.06 | 54.07 | — | |
| GraphCLEvaluation Protocol=5-shot, Model Category=Graph Pretrain Method2026.06 | 53.6 | — | |
| LLaGA-QwenCategory=Graph-LLM Method, LLM Backbone=Qwen, Supervision Strategy=Semi-Supervised2026.06 | 53.32 | — | |
| TEA-GLM-QwenEvaluation Protocol=5-shot, Model Category=Graph-LLM Method, Backbone=Qwen2026.06 | 51.65 | — | |
| BLIP-2 (LoRA)Category=VLM, requires LLM/VLM at inference=true, protocol=LoRA2026.05 | — | 86 | |
| GPT-4o (ICL)Category=LLM, requires LLM/VLM at inference=true, protocol=ICL2026.05 | — | 56 | |
| Janus-Pro 7B (ICL)Category=VLM, requires LLM/VLM at inference=true, protocol=ICL2026.05 | — | 78 | |
| MolSight (SigLIP2, S6)Category=Image-only, Configuration=single-model results2026.05 | — | 82 | |
| MolSight-BestCategory=Image-only, Configuration=best curriculum-trained architecture per task2026.05 | — | 82 | |
| Uni-MolCategory=3D GNN, requires 3D conformer generation=true2026.05 | — | 78 |