Directed graph transformers meet metabolic networks

Alexandru Oarga, Jorge Júlvez

Keywords: Graph Transformers, Genome-Scale Metabolic Models, Metabolic Networks, Essential Reactions, Petri Nets, Biological Networks, Systems Biology

TL;DR: We propose a data-driven approach, based on a directed graph transformer architecture, to identify essential reactions in genome-scale metabolic models and build a large-scale dataset to showcase the effectiveness.

Abstract: Technical advances in sequencing have allowed the reconstruction of genome-scale metabolic models (GEMs) for a wide range of microorganisms. These models have been particularly useful for the prediction of essential genes and reactions, which are potential targets for antimicrobial therapies. However, current methods for essentiality prediction are computationally limited and are not able to accommodate the increasingly available data. Motivated by the success of data-driven approaches in other domains, this work introduces the metabolic transformer, a model designed for holistic identification of essential reactions in genome-scale models, entirely trained on synthetic knock-out data. It is demonstrated that the problem of essential reaction prediction can be theoretically formulated as the identification of redundant nodes in directed bipartite graphs. This reveals the limitations of message-passing schemes and motivates the development of a novel graph transformer architecture specifically tailored for metabolic networks. The proposed architecture is capable of addressing the essential reaction identification problem by capturing both the directionality and global structure of metabolic networks. To demonstrate the effectiveness of our approach, we composed a large-scale dataset of genome-scale models reconstructed from real microorganisms.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 11249