2/20/2026 | USA | technology | ✓ Verified - arxiv.org

Beyond Message Passing: A Symbolic Alternative for Expressive and Interpretable Graph Learning

#Graph Neural Network #SymGraph #1-Weisfeiler-Leman #Structural Hashing #Topological Role Aggregation #Rule‑Based Explainability #CPU Training Speedup #Drug Discovery

📌 Key Takeaways

SymGraph replaces continuous message passing with discrete symbolic operations such as structural hashing and topological role aggregation.
Theoretical analysis shows it can surpass the 1‑WL expressivity limit that traditional message‑passing GNNs cannot exceed.
Empirical benchmarks demonstrate state‑of‑the‑art accuracy across a suite of graph‑learning tasks.
Training speeds reach 10‑ to 100‑fold acceleration on CPU-only hardware, eliminating the need for GPU‑heavy differentiation.
The framework generates interpretable rule‑based explanations with finer semantic granularity than prior rule‑based methods.

📖 Full Retelling

WHO: A team of eight researchers—Chuqin Geng, Li Zhang, Haolin Ye, Ziyu Zhao, Yuhe Jiang, Tara Saba, Xinyu Wang, and Xujie Si—who published together. WHAT: They present SymGraph, a symbolic framework that replaces continuous message‑passing in Graph Neural Networks (GNNs) with discrete structural hashing and role‑based aggregation. WHERE: The work was submitted to arXiv, an open‑access pre‑print repository. WHEN: The pre‑print was made public on 18 February 2026. WHY: The authors aim to overcome fundamental constraints of standard GNNs—the 1‑Weisfeiler‑Leman (1‑WL) expressivity barrier—and to enhance interpretability and computational efficiency, particularly for high‑stakes fields such as drug discovery.

🏷️ Themes

Graph Neural Networks, Symbolic Machine Learning, Expressivity Limits, Interpretability, Computational Efficiency

Entity Intersection Graph

No entity connections available yet for this article.

Deep Analysis

Why It Matters

SymGraph offers a symbolic alternative to conventional message‑passing GNNs, overcoming the 1‑WL expressivity barrier and providing fine‑grained interpretability. This makes it especially valuable for high‑stakes domains like drug discovery, where model transparency and speed are critical.

Context & Background

GNNs are widely used in drug discovery but are often black‑box.
Standard message‑passing GNNs are limited by the 1‑WL expressivity barrier.
SymGraph replaces continuous message passing with discrete hashing and role‑based aggregation, achieving higher expressiveness and interpretability.

What Happens Next

Researchers will likely benchmark SymGraph against existing GNNs across diverse graph tasks, exploring its applicability beyond drug discovery. The framework’s CPU‑only speedups may encourage adoption in resource‑constrained settings, and future work may extend its symbolic rules to other domains such as chemistry and biology.

Frequently Asked Questions

How does SymGraph differ from traditional message‑passing GNNs?

Unlike conventional GNNs that rely on continuous message passing and are limited by the 1‑WL test, SymGraph uses discrete structural hashing and topological role‑based aggregation to achieve higher expressiveness and generate interpretable symbolic rules.

What are the practical implications for drug discovery?

SymGraph’s faster training on CPUs and its rule‑based explanations can accelerate the screening of molecular graphs, provide clearer insights into structure‑activity relationships, and improve trust in predictive models used in drug design.

Original Source

              --> Computer Science > Machine Learning arXiv:2602.16947 [Submitted on 18 Feb 2026] Title: Beyond Message Passing: A Symbolic Alternative for Expressive and Interpretable Graph Learning Authors: Chuqin Geng , Li Zhang , Haolin Ye , Ziyu Zhao , Yuhe Jiang , Tara Saba , Xinyu Wang , Xujie Si View a PDF of the paper titled Beyond Message Passing: A Symbolic Alternative for Expressive and Interpretable Graph Learning, by Chuqin Geng and 7 other authors View PDF HTML Abstract: Graph Neural Networks have become essential in high-stakes domains such as drug discovery, yet their black-box nature remains a significant barrier to trustworthiness. While self-explainable GNNs attempt to bridge this gap, they often rely on standard message-passing backbones that inherit fundamental limitations, including the 1-Weisfeiler-Lehman (1-WL) expressivity barrier and a lack of fine-grained interpretability. To address these challenges, we propose SymGraph, a symbolic framework designed to transcend these constraints. By replacing continuous message passing with discrete structural hashing and topological role-based aggregation, our architecture theoretically surpasses the 1-WL barrier, achieving superior expressiveness without the overhead of differentiable optimization. Extensive empirical evaluations demonstrate that SymGraph achieves state-of-the-art performance, outperforming existing self-explainable GNNs. Notably, SymGraph delivers 10x to 100x speedups in training time using only CPU execution. Furthermore, SymGraph generates rules with superior semantic granularity compared to existing rule-based methods, offering great potential for scientific discovery and explainable AI. Comments: 23 pages, 9 pages Subjects: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI) Cite as: arXiv:2602.16947 [cs.LG] (or arXiv:2602.16947v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2602.16947 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission histo...
            

Read full article at source

Source

arxiv.org