Multiple Heterogeneous Networks Representation With Latent Space for Synthetic Lethality Prediction

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Computational synthetic lethality (SL) is vital for cancer therapy development.
• Current feature representation methods struggle to capture complex gene correlations and geometric structures.
• A need exists for advanced models of gene latent spaces.

Purpose of the Study:

• To propose a novel method, Latent Space using matrix Tri-Factorization (LSTF), for improved gene representation.
• To enhance the identification of synthetic lethality (SL) gene pairs.
• To incorporate manifold subspace regularization for capturing latent geometric structures.

Main Methods:

• Developed Latent Space using matrix Tri-Factorization (LSTF) for gene representation.
• Applied manifold subspace regularization with gene PPI functional and GO semantic embeddings.
• Identified SL gene pairs through reconstruction of associations in the latent space.

Main Results:

• LSTF demonstrated superior performance compared to existing state-of-the-art methods.
• Experimental results validated the effectiveness of the proposed method.
• Case studies confirmed the accuracy of predicted SL associations.

Conclusions:

• LSTF offers a robust approach for modeling gene latent spaces.
• The method enhances the identification of synthetic lethality gene pairs for cancer medicine.
• This advancement holds promise for targeted cancer therapy development.