MARML: Motif-Aware Deep Representation Learning in Multilayer Networks.

Area of Science:

• Computational biology
• Network science
• Data science

Background:

• High-throughput data analysis increasingly relies on network models.
• Graph embedding methods aid interpretation but struggle with complex, cross-domain interactions.
• Multilayer networks integrate biological data but lack methods for embedding diverse interaction types.

Purpose of the Study:

• To introduce Motif-aware deep representation learning in multilayer (MARML) networks.
• To address the challenge of embedding nodes with different interaction types in multilayer networks.
• To enhance the analysis of complex biological systems.

Main Methods:

• MARML utilizes recurring motif patterns, topological information, and attributive data as node features.
• The method learns network representations by integrating these diverse data sources.
• Validation was performed on various multilayer network datasets.

Main Results:

• MARML effectively incorporates higher-order connections across hierarchies using motif information.
• Learned features demonstrated high accuracy in link prediction and link differentiation tasks.
• The method successfully distinguished between existing and disconnected triplets.

Conclusions:

• MARML enhances the understanding of complex biological systems.
• Integrating intrinsic node attributes and topological structures improves network representation.
• The proposed method offers a novel solution for analyzing multilayer biological networks.