Machine Learning-Guided Differentiation Therapy Targets Cancer Stem Cells in Colorectal Cancers.

Area of Science:

• Oncology
• Computational Biology
• Genomics

Background:

• Solid tumor differentiation therapy is limited by heterogeneity, hindering computational approaches.
• Colorectal cancers (CRCs) serve as a model to demonstrate a machine learning (ML) strategy for targeting cancer stem cells (CSCs).

Purpose of the Study:

• To develop and validate an ML-based approach for tracking, differentiating, and selectively targeting CSCs in CRCs.
• To identify therapeutic targets and strategies to reinstate CDX2 expression, a key marker of differentiated colonocytes lost in aggressive CRCs.

Main Methods:

• A transcriptomic network was constructed and validated using extensive human and mouse gene expression data.
• ML algorithms prioritized therapeutic targets and strategies, focusing on CDX2 re-expression.
• A clinical-grade drug targeting CDX2 was tested in vitro, in vivo xenografts, and patient-derived organoids (PDOs).

Main Results:

• The drug successfully induced CDX2 expression and crypt differentiation, demonstrating cytotoxicity selective for CDX2-negative cells across all models.
• Multivariate analyses in PDOs confirmed the link between therapeutic efficacy and the CDX2 biomarker.
• A 50-gene signature of therapeutic response, validated across ~1700 CRCs, suggests a potential ~50% reduction in mortality/recurrence risk.

Conclusions:

• The ML approach precisely targets CSCs and objectively assesses clinical impact.
• This study exemplifies ML's utility in generating clinical insights for personalized cancer medicine.
• CDX2 re-instatement therapy shows promise for improving patient outcomes in colorectal cancer.