Reassessing pharmacogenomic cell sensitivity with multilevel statistical models

Area of Science:

• Pharmacogenomics
• Computational Biology
• Cancer Research

Background:

• Pharmacogenomic experiments systematically test drug effects on cell lines across various concentrations to link genomic markers with treatment sensitivity.
• Quantifying drug response in cell lines is crucial but challenging due to low signal-to-noise ratios from biological and experimental variability.
• Replicated pharmacogenomic datasets offer opportunities to increase statistical power for robust analysis.

Purpose of the Study:

• To develop a hierarchical mixture model for estimating drug-specific cell sensitivity and classifying drug effects as broad or targeted.
• To propose a unified approach for calculating the probability of cell susceptibility to a drug under targeted effects or effect sizes under broad effects.
• To demonstrate the utility of the proposed model through case studies and analysis of publicly available pharmacogenomic data.

Main Methods:

• Formulation of a hierarchical mixture model to estimate drug-specific mixture distributions for cell sensitivity.
• Development of a unified approach to yield posterior probabilities of cell susceptibility (targeted effect) or relative effect sizes (broad effect).
• Application of the model to assess pairwise agreements across pharmacogenomic datasets and identify drug-specific sensitivities.

Main Results:

• The study confirms moderate pairwise agreement between many publicly available pharmacogenomic datasets.
• Analysis identified specific cell lines (harboring EML4-ALK or NPM1-ALK gene fusions) sensitive to the drug crizotinib.
• Significantly down-regulated cell-matrix pathways were associated with crizotinib sensitivity in these cell lines.

Conclusions:

• The proposed hierarchical mixture model provides a robust framework for analyzing pharmacogenomic data, enhancing the estimation of drug sensitivity.
• The unified approach allows for nuanced assessment of drug effects, distinguishing between targeted and broad mechanisms.
• The findings highlight the potential of pharmacogenomic data integration and analysis for identifying novel drug-response relationships and therapeutic targets.