PROBABILISTIC LEARNING OF TREATMENT TREES IN CANCER

Area of Science:

• Oncology
• Computational Biology
• Biostatistics

Background:

• Identifying synergistic cancer treatments and their mechanisms is crucial for effective therapy.
• Patient-derived xenografts (PDX) offer a valuable preclinical model for evaluating multiple treatment combinations on human tumor samples.
• Existing methods lack robust frameworks for analyzing complex relationships within PDX data.

Purpose of the Study:

• To propose a novel Bayesian probabilistic tree-based framework (Rx-tree) for analyzing PDX data.
• To infer hierarchical treatment relationships and quantify mechanistic similarity between treatments.
• To identify potential synergistic treatment combinations for cancer therapy.

Main Methods:

• Developed a Bayesian probabilistic tree-based framework (Rx-tree) utilizing Dirichlet Diffusion Trees.
• Derived a closed-form marginal likelihood for computationally efficient posterior inference using a two-stage algorithm.
• Introduced a new metric for mechanistic similarity between treatments, accounting for estimation uncertainty.

Main Results:

• The proposed Rx-tree framework accurately recovers treatment structures and similarities in simulation studies.
• Analysis of a PDX dataset showed high concordance between estimated treatment similarities and known biological mechanisms across five cancer types.
• Identified novel, potentially effective combination therapies with synergistic pathway regulation.

Conclusions:

• The Rx-tree framework provides a robust and computationally efficient method for analyzing PDX data.
• This approach facilitates the discovery of synergistic treatment combinations and elucidation of their underlying biological mechanisms.
• The findings offer promising avenues for future clinical investigations in cancer therapy.