Proteins are central to virtually all biological processes, and their structural diversity underlies a wide range of functions. Among them, multifunctional proteins (i.e., those associated with multiple enzymatic activities) pose unique challenges for computational analysis. In this work, we apply a suite of machine learning and deep learning techniques to predict multifunctionality roles starting from protein 3D structures. Proteins are represented as residue contact networks, enabling the use of graph machine learning approaches. We explore feature-based methods grounded in simplicial complexes analysis alongside end-to-end Graph Neural Networks implementing recent message-passing schemes. Our models address the task of multi-label classification of the first level Enzyme Commission numbers of multifunctional proteins. Evaluation across a strictly multifunctional subset of the human proteome associated with repeated stratified validation demonstrates that structural graph representations effectively capture signals of multifunctionality, shedding light on how protein architecture encodes a diverse range of biochemical roles.
Decoding functional multiplicity: graph learning approaches to multifunctional proteins / Cervellini, M., De Santis, E., Rizzi, A., Martino, A.. - (2026). (2026 International Joint Conference on Neural Networks (IJCNN) Maastricht; Netherlands ) [10.1109/IJCNN69854.2026.11729755].
Decoding functional multiplicity: graph learning approaches to multifunctional proteins
Mattia CervelliniPrimo
;Enrico De SantisSecondo
;Antonello RizziPenultimo
;
2026
Abstract
Proteins are central to virtually all biological processes, and their structural diversity underlies a wide range of functions. Among them, multifunctional proteins (i.e., those associated with multiple enzymatic activities) pose unique challenges for computational analysis. In this work, we apply a suite of machine learning and deep learning techniques to predict multifunctionality roles starting from protein 3D structures. Proteins are represented as residue contact networks, enabling the use of graph machine learning approaches. We explore feature-based methods grounded in simplicial complexes analysis alongside end-to-end Graph Neural Networks implementing recent message-passing schemes. Our models address the task of multi-label classification of the first level Enzyme Commission numbers of multifunctional proteins. Evaluation across a strictly multifunctional subset of the human proteome associated with repeated stratified validation demonstrates that structural graph representations effectively capture signals of multifunctionality, shedding light on how protein architecture encodes a diverse range of biochemical roles.| File | Dimensione | Formato | |
|---|---|---|---|
|
Cervellini_Decoding_2026.pdf
solo gestori archivio
Note: Camera Ready Article
Tipologia:
Documento in Post-print (versione successiva alla peer review e accettata per la pubblicazione)
Licenza:
Tutti i diritti riservati (All rights reserved)
Dimensione
248.71 kB
Formato
Adobe PDF
|
248.71 kB | Adobe PDF | Contatta l'autore |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.


