LRTM: Left-Right Transition Matrices for Molecular Association Prediction

Area of Science:

• Computational biology
• Bioinformatics
• Network science

Background:

• Molecular associations are crucial for biological processes, diagnostics, and drug development.
• Existing computational methods for predicting molecular associations are often domain-specific and require complex preprocessing.
• A generalized approach for accurate molecular association prediction remains a challenge.

Purpose of the Study:

• To propose a generalized computational method for predicting molecular associations.
• To introduce Left-Right Transition Matrices (LRTM) for modeling molecular bipartite networks.
• To evaluate the performance and generalizability of the LRTM algorithm.

Main Methods:

• Constructed two transition matrices from a diffusion model perspective to capture undirected graph information propagation.
• Modeled transition probabilities of links within molecular bipartite networks.
• Applied LRTM for molecular association prediction and link prediction tasks.

Main Results:

• The LRTM algorithm demonstrated superior performance compared to existing methods in extensive experiments.
• LRTM showed potential for cross-task prediction, indicating its generalizability.
• Case studies confirmed LRTM's effectiveness in practical biological applications.

Conclusions:

• Left-Right Transition Matrices provide a powerful and generalized approach for molecular association prediction.
• The method overcomes limitations of domain-specific models and complex preprocessing.
• LRTM offers a valuable tool for advancing biological exploration and drug discovery.