A Gaussian Mixture-Model Exploiting Pathway Knowledge for Dissecting Cancer Heterogeneity.

Area of Science:

• Computational Biology
• Systems Biology
• Cancer Research

Background:

• Heterogeneous cancer tissues present challenges in treatment due to diverse cell subpopulations.
• Understanding cellular responses to drugs is crucial for effective cancer therapy design.

Purpose of the Study:

• To develop a systematic approach for dissecting heterogeneous cancer tissues using pathway knowledge and multivariate Gaussian mixture models.
• To evaluate the efficacy of an Expectation-Maximization (EM) algorithm for estimating mixture composition from cell-by-cell measurements.

Main Methods:

• Applied pathway knowledge to a multivariate Gaussian mixture model.
• Utilized downstream transcription factors as observables and upstream drugs.
• Employed an Expectation-Maximization (EM) algorithm with hill-climbing for mixture composition estimation.
• Validated the approach using cell-by-cell data from dynamic cell imaging of multiple cancer cell lines under different drug treatments.

Main Results:

• The EM-based approach successfully estimated the composition of experimental mixture sets.
• Demonstrated the methodology's application on hourly data for A2058, HCT116, and SW480 cell lines under untreated, Lapatinib-treated, and Temsirolimus-treated conditions.
• Showcased the ability to compare the killing rates of different drugs on heterogeneous cancer tissues.

Conclusions:

• The developed methodology provides a robust framework for analyzing complex cellular systems like heterogeneous tumors.
• This approach has significant implications for designing more efficient and targeted anti-cancer drugs.