Decoding fibrosis: Transcriptomic and clinical insights via AI-derived collagen deposition phenotypes in MASLD

Area of Science:

• Pathology
• Bioinformatics
• Artificial Intelligence

Background:

• Histological assessment is crucial for liver disease multi-omics studies.
• Conventional fibrosis staging and metrics like collagen proportionate area (CPA) lack resolution and fail to capture tissue architecture.
• Existing AI approaches are proprietary and inaccessible for academic research.

Purpose of the Study:

• To present a novel, interpretable AI-based framework for characterizing liver fibrosis from picrosirius red (PSR)-stained slides.
• To improve the sensitivity and biological specificity of downstream transcriptomic and proteomic analyses.

Main Methods:

• Development of a novel, interpretable AI framework for liver fibrosis characterization.
• Identification of data-driven collagen deposition phenotypes (CDPs) capturing distinct morphologies.
• Comparison of CDP performance against CPA and traditional fibrosis scores in multi-omics analyses.

Main Results:

• The AI framework identified distinct CDPs, significantly enhancing sensitivity and biological specificity in multi-omics analyses compared to CPA and traditional scores.
• Pathway analysis linked CDPs 4 and 5 to active extracellular matrix remodeling.
• CDP correlations highlighted associations with liver functional status and demonstrated prognostic value in a discovery cohort.

Conclusions:

• The novel AI framework provides a more sensitive and specific method for liver fibrosis assessment using PSR-stained slides.
• The identified CDPs offer deeper insights into liver disease biology and patient prognosis.
• Freely available models and tools promote transparent and reproducible multi-omics pathology research.