Machine learning on syngeneic mouse tumor profiles to model clinical immunotherapy response

Area of Science:

• Immunology
• Computational Biology
• Oncology

Background:

• Immune checkpoint blockade (ICB) therapy shows limited efficacy in many cancer patients, necessitating better patient stratification methods.
• Primary human cancer data present challenges like heterogeneity and limited accessibility, hindering controlled treatment studies.
• Syngeneic mouse tumor models offer a controlled environment for investigating ICB treatments and identifying response predictors.

Purpose of the Study:

• To develop a machine learning framework for modeling tumor immunity and predicting ICB response using syngeneic mouse tumor data.
• To validate the predictive capability of the developed model on human immunotherapy trial data.
• To identify key factors influencing ICB response and resistance across different cancer types.

Main Methods:

• Utilized transcriptomic and experimental data from over 700 ICB-treated/control syngeneic mouse tumors.
• Developed a machine learning framework to model tumor immunity and predict ICB response.
• Applied the model to human immunotherapy trial datasets and The Cancer Genome Atlas (TCGA) data.

Main Results:

• The machine learning model successfully predicted clinical ICB response when projected onto human immunotherapy trial data.
• The model accurately predicted ICB-responsive and resistant cancer types in TCGA, aligning with clinical observations.
• Feature analysis identified specific factors significantly associated with ICB response and resistance.

Conclusions:

• The computational framework, built on mouse tumor data, reliably stratifies patients for ICB therapy.
• The study provides insights into mechanisms of ICB resistance.
• The developed framework holds potential for broad applications in cancer treatment and disease studies.