TopoBench: A Framework for Benchmarking Topological Deep Learning

L. Telyatnikov, Guillermo Bernardez, Marco Montagna, Mustafa Hajij, Martin Carrasco, Pavlo Vasylenko, Mathilde Papillon, Ghada Zamzmi, Michael T. Schaub, Jonas Verhellen, Pavel Snopov, Bertran Miquel-Oliver, Manel Gil-Sorribes, Alexis Molina, V. Guallar, Theodore Long, Julian Suk, Patryk Rygiel, Alexander Nikitin, Giordan Escalona, M. Banf, Dominik Filipiak, Max Schattauer, Liliya Imasheva, Álvaro Martínez, Halley Fritze, Marissa Masden, Valentina S'anchez, M. Lecha, Andrea Cavallo, C. Battiloro, Matt Piekenbrock, M. Tec, George Dasoulas, N. Miolane, Simone Scardapane, Theodore Papamarkou

Abstract

This work introduces, an open-source library designed to standardize benchmarking and accelerate research in topological deep learning (TDL). decomposes TDL into a sequence of independent modules for data generation, loading, transforming and processing, as well as model training, optimization and evaluation. This modular organization provides flexibility for modifications and facilitates the adaptation and optimization of various TDL pipelines. A key feature of is its support for transformations and lifting across topological domains. Mapping the topology and features of a graph to higher-order topological domains, such as simplicial and cell complexes, enables richer data representations and more fine-grained analyses. The applicability of is demonstrated by benchmarking several TDL architectures across diverse tasks and datasets.

Benchmark Results

Tested on 5 datasets 4 baseline values reported too low

Tables show presented results in global context (best 3 papers + GCN + MLP).

Heterophily Benchmark

Minesweeper

Global leaderboard →

#	Model	ROC-AUC ↑	Paper
1	BuNN code ↗ Hyb global	0.9899 ±0.0016	Bundle Neural Networks for message diffusion on graphs code ↗
2	DC-GNN GNN global	0.9850 ±0.0021	Differentiable Cluster Graph Neural Network
3	Tango -GPS global	0.9839 ±0.0054	TANGO: Graph Neural Dynamics via Learned Energy and Tangential Flows
36	GAT GNN baseline actual #36 vs paper #93 57 places worse	0.9391 ±0.0035 best reported	actual #36 vs paper #93 57 places worse paper 0.8964 ±0.0043 best at pub: Graph Neural Networks for Graphs with Heterophily: A Survey (2022) best today: Relieving the Over-Aggregating Effect in Graph Transformers (NeurIPS 2025) paper 0.8964 ±0.0043 Δ at pub 0.0237 Δ today 0.0427
42	GraphSAGE code ↗ GNN global	0.9358 ±0.0005	Gumbel-MPNN: Graph Rewiring with Gumbel-Softmax code ↗
68	GCN GNN baseline actual #68 vs paper #106 38 places worse	0.9150 ±0.0006 best reported	actual #68 vs paper #106 38 places worse paper 0.8752 ±0.0042 best at pub: Graph Neural Networks for Graphs with Heterophily: A Survey (2022) best today: Gumbel-MPNN: Graph Rewiring with Gumbel-Softmax (2025) paper 0.8752 ±0.0042 Δ at pub 0.0223 Δ today 0.0398
82	SCN GNN baseline	0.9032 ±0.0011	This paper
95	CCCN GNN baseline	0.8942	This paper
98	SCCNN GNN baseline	0.8900	This paper
101	CWN GNN baseline	0.8862 ±0.0004	This paper
104	GIN GNN baseline	0.8782 ±0.0034	This paper
112	EDGNN GNN baseline	0.8452 ±0.0005	This paper
117	AST GT baseline	0.8114 ±0.0005	This paper
121	UniGNN2 GNN baseline	0.7802	This paper
156	MLP Trad global	0.5099 ±0.0147	Graph Homophily Booster: Rethinking the Role of Discrete Features on Heterophilic Graphs

Tolokers

Global leaderboard →

#	Model	ROC-AUC ↑	Paper
1	Polynormer-r code ↗ GT global	0.8591 ±0.0074	Polynormer: Polynomial-Expressive Graph Transformer in Linear Time code ↗
2	DC-GNN GNN global	0.8588 ±0.0081	Differentiable Cluster Graph Neural Network
3	TANGO-GPS Hyb global	0.8566 ±0.0101	TANGO: Graph Neural Dynamics via Learned Energy and Tangential Flows
28	GAT GNN baseline actual #28 vs paper #28	0.8443 ±0.0100	actual #28 vs paper #28 paper 0.8443 ±0.0100 prev best: Graph Neural Networks for Graphs with Heterophily: A Survey (2022) Improvement within reported uncertainty; not highlighted. paper 0.8443 ±0.0100 Δ at pub 0.0073
50	GCN GNN baseline actual #50 vs paper #62 12 places worse	0.8364 ±0.0067 best reported	actual #50 vs paper #62 12 places worse paper 0.8302 ±0.0071 best at pub: Graph Neural Networks for Graphs with Heterophily: A Survey (2022) best today: Beyond Oversquashing: Understanding Signal Propagation in GNNs Via Observables (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.8302 ±0.0071 Δ at pub 0.0062 Δ today 0.0062
60	AST GT baseline	0.8326 ±0.0010	This paper
76	SAGE GNN global	0.8243 ±0.0044	Beyond Oversquashing: Understanding Signal Propagation in GNNs Via Observables
89	GIN GNN baseline actual #89 vs paper #89	0.8072 ±0.0119	actual #89 vs paper #89 paper 0.8072 ±0.0119 prev best: Data Augmentation for Supervised Graph Outlier Detection via Latent Diffusion Models (LoG 2023) paper 0.8072 ±0.0119 Δ at pub 0.0667
124	EDGNN GNN baseline	0.7753 ±0.0001	This paper
125	UniGNN2 GNN baseline	0.7735 ±0.0003	This paper
139	MLP code ↗ Trad global	0.7362 ±0.0020	Gumbel-MPNN: Graph Rewiring with Gumbel-Softmax code ↗

Other Graph Benchmarks

Amazon2M

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	MA-GCL code ↗ GNN global	0.8883 ±0.0030	MA-GCL: Model Augmentation Tricks for Graph Contrastive Learning code ↗
2	CCA-SSG code ↗ GNN global	0.8842 ±0.0030	MA-GCL: Model Augmentation Tricks for Graph Contrastive Learning code ↗
3	ARIEL code ↗ GNN global	0.8827 ±0.0020	MA-GCL: Model Augmentation Tricks for Graph Contrastive Learning code ↗
9	GAT GNN baseline actual #9 vs paper #32 23 places worse	0.8693 ±0.0030 best reported	actual #9 vs paper #32 23 places worse paper 0.5017 ±0.0059 best at pub: MA-GCL: Model Augmentation Tricks for Graph Contrastive Learning (AAAI 2022) paper 0.5017 ±0.0059 Δ at pub 0.3676
10	GCN GNN baseline actual #10 vs paper #32 22 places worse	0.8651 ±0.0050 best reported	actual #10 vs paper #32 22 places worse paper 0.4956 ±0.0055 best at pub: MA-GCL: Model Augmentation Tricks for Graph Contrastive Learning (AAAI 2022) paper 0.4956 ±0.0055 Δ at pub 0.3695
30	CWN GNN baseline	0.5190 ±0.0015	This paper
31	AST GT baseline	0.5050 ±0.0027	This paper
32	CCCN GNN baseline	0.5026 ±0.0017	This paper
33	GIN GNN baseline	0.4916 ±0.0102	This paper
34	UniGNN2 GNN baseline	0.4906 ±0.0008	This paper
35	EDGNN GNN baseline	0.4818 ±0.0009	This paper
38	SAGE GNN global	0.4648	Language Models as Messengers: Enhancing Message Passing in Heterophilic Graph Learning

Hypergraph Benchmarks

Cora-CA

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	HAN (full batch) code ↗ GNN global	0.8404 ±0.0102	Hypergraph Benchmarks code ↗
2	AllSetTransformer GT baseline actual #2 vs paper #6 4 places worse	0.8363 ±0.0147 best reported	actual #2 vs paper #6 4 places worse paper 0.8219 ±0.0261 best at pub: Hypergraph Benchmarks (ICLR 2021) Difference within reported uncertainty; not counted as underperformance. paper 0.8219 ±0.0261 Δ at pub 0.0144
3	UniGCNII code ↗ GNN global	0.8360 ±0.0114	Hypergraph Benchmarks code ↗
6	EDGNN GNN baseline	0.8236 ±0.0072	This paper
9	UniGNN2 GNN baseline	0.8171 ±0.0142	This paper
20	GCN code ↗ GNN global	0.7516 ±0.0065	Node Embedding from Neural Hamiltonian Orbits in Graph Neural Networks code ↗
21	MLP code ↗ Trad global	0.7431 ±0.0189	Hypergraph Benchmarks code ↗

DBLP-CA

Global leaderboard →

#	Model	Accuracy ↑	Paper
2	AllSetTransformer GT baseline	0.9198 ±0.0018	This paper
1	HealHGNN code ↗ GNN global	0.9198 ±0.0031	Heterophily-Agnostic Hypergraph Neural Networks with Riemannian Local Exchanger code ↗
3	ED-HNN code ↗ GNN global	0.9190 ±0.0019	Heterophily-Agnostic Hypergraph Neural Networks with Riemannian Local Exchanger code ↗
14	EDGNN GNN baseline	0.9083 ±0.0025	This paper
15	UniGNN2 GNN baseline	0.9072 ±0.0023	This paper
20	MLP code ↗ Trad global	0.8483 ±0.0022	Hypergraph Benchmarks code ↗

SOTA performance (not significantly different from the best) Baseline / comparison (this paper) Global context (other papers) Best at pub — reported result worse than the best known when published (best reported in green) Best at pub — reported value improved over the best known when published