FR-BINN: Biologically Informed Neural Networks for Enhanced Biomarker Discovery and Pathway Analysis

Area of Science:

• Biomedical Informatics
• Computational Biology
• Cancer Research

Background:

• Chronic inflammation is linked to cancer risk, but the molecular differences between cancer-prone and non-cancer-prone inflammatory diseases are unclear.
• Understanding these differences is crucial for developing targeted cancer prevention and treatment strategies.

Purpose of the Study:

• To develop a biologically informed neural network (FR-BINN) for predicting disease outcomes and interpreting molecular mechanisms in chronic inflammatory diseases.
• To identify key genes and biological pathways associated with cancer risk in chronic inflammation.

Main Methods:

• Developed FR-BINN, a neural network incorporating Fenton reaction (FR)-related biological priors.
• Utilized multiple interpretability methods and large language models for gene identification and biomarker discovery.
• Validated findings using independent datasets and analyzed patterns in energy metabolism, oxidative stress, and pH regulation.

Main Results:

• FR-BINN demonstrated superior classification performance and provided biologically interpretable insights.
• High consistency was observed across different explainable AI techniques, revealing distinct patterns in cancer-prone vs. non-cancer-prone diseases.
• Identified genes like NCOA1 and SDHB associated with cancer susceptibility, and highlighted differences in energy metabolism, oxidative stress, and pH regulation.

Conclusions:

• FR-BINN effectively distinguishes cancer-prone from non-cancer-prone chronic inflammatory diseases, offering valuable biological insights.
• The framework facilitates the identification of potential biomarkers and therapeutic targets for inflammation-associated tumorigenesis.
• Distinct metabolic and oxidative stress patterns are key differentiators between these disease categories.