Visible neural networks for multi-omics integration: a critical review

Area of Science:

• Computational biology
• Bioinformatics
• Artificial intelligence in medicine

Background:

• Biomarker discovery and drug response prediction are crucial for personalized medicine.
• Biologically informed neural networks (BINNs), also known as visible neural networks (VNNs), integrate prior biological knowledge (e.g., gene ontologies, pathway databases) into neural network architectures.
• These sparse models enhance interpretability and focus on biologically meaningful functions.

Purpose of the Study:

• To systematically review and analyze the current landscape of BINNs/VNNs in scientific literature.
• To identify key trends in architectural design, data sources, and evaluation methodologies for BINNs/VNNs.
• To highlight the potential of BINNs/VNNs for advancing personalized medicine.

Main Methods:

• Conducted a systematic review, identifying 86 relevant research papers.
• Analyzed the selected papers to extract information on architectural design, data sources, and evaluation methods.
• Synthesized findings to highlight trends and identify areas for improvement.

Main Results:

• Demonstrated a growing adoption and implementation of BINNs/VNNs in recent literature.
• Observed a lack of standardized terminology, computational tools, and benchmarks in the field.
• Highlighted the heterogeneity in architectural design and evaluation methodologies.

Conclusions:

• BINNs/VNNs show significant promise for integrating biological insights into predictive models for personalized medicine.
• Addressing the current lack of standardization and development of computational tools is essential for the widespread adoption and advancement of BINNs/VNNs.
• Further research and community efforts are needed to establish best practices and facilitate progress in this domain.