Knowledge-guided artificial intelligence technologies for decoding complex multiomics interactions in cells.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Cellular functions rely on complex molecular interactions across multiomics data.
• High-throughput technologies generate vast multiomics data, overwhelming conventional experimental methods.
• Modeling nonlinear multiomics interactions is challenging due to complex dependencies.

Purpose of the Study:

• To review deep learning models for multiomics interactions in bioinformatics.
• To categorize AI models based on how they incorporate prior biological knowledge.
• To highlight the benefits of knowledge-guided AI in multiomics data analysis.

Main Methods:

• Literature review of bioinformatics applications of deep learning models.
• Categorization of models by their knowledge-guidance mode (e.g., architecture, training).
• Analysis of AI model performance and interpretability in multiomics interaction studies.

Main Results:

• Deep learning excels at modeling nonlinear relationships in large-scale multiomics data.
• Few existing AI models explicitly integrate prior biological knowledge.
• Knowledge-guided AI can improve model efficiency, interpretability, and reduce spurious findings.

Conclusions:

• Artificial intelligence, particularly deep learning, shows significant promise for modeling complex multiomics interactions.
• Integrating domain knowledge into AI models is crucial for enhancing their biological relevance and reducing data requirements.
• Further development of knowledge-guided AI is essential for advancing bioinformatics and understanding cellular mechanisms.