SADLN: Self-attention based deep learning network of integrating multi-omics data for cancer subtype recognition

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Cancer subtype recognition is crucial for personalized treatment.
• Existing deep learning methods for multi-omics integration often overlook data distribution and sample relationships.
• Simple data concatenation limits the effectiveness of current deep learning approaches.

Purpose of the Study:

• To develop an advanced deep learning framework for effective multi-omics data integration in cancer subtype recognition.
• To address limitations of existing methods by considering differential omics data distributions and sample interdependencies.
• To propose the Self-Attention Based Deep Learning Network (SADLN) for enhanced cancer subtyping.

Main Methods:

• Developed SADLN, a unified framework incorporating encoder, self-attention, decoder, and discriminator.
• Employed self-attention mechanisms to adaptively model sample relationships.
• Utilized Gaussian Mixture Model (GMM) for cancer subtype identification based on learned integrated representations.

Main Results:

• SADLN demonstrated superior performance in cancer subtype recognition across ten TCGA cancer datasets.
• The proposed method outperformed ten existing comparative approaches.
• The self-attention mechanism effectively captured complex inter-omics and inter-sample relationships.

Conclusions:

• SADLN offers an effective and robust approach for integrating multi-omics data for cancer subtype recognition.
• The framework's ability to model sample relationships and data distributions leads to improved accuracy.
• This work highlights the potential of self-attention mechanisms in bioinformatics for complex biological data analysis.