A Pan-Cancer Ex Vivo Drug Screen Atlas for Functional Precision Oncology

Area of Science:

• Oncology
• Translational Medicine
• Computational Biology

Background:

• Patient-derived organoids and ex vivo models better predict therapeutic responses compared to traditional immortalized cell lines.
• High costs and technical challenges have hindered the development of comprehensive pan-cancer ex vivo datasets for drug response analysis.
• Existing cell line databases exhibit biases, potentially undervaluing certain drug targets relevant to cancer biology.

Purpose of the Study:

• To establish a large-scale, harmonized ex vivo drug screening atlas across diverse cancer types.
• To identify tissue-specific drug sensitivities and their correlation with molecular profiles.
• To reveal and address systematic biases present in conventional cancer cell line screening models.

Main Methods:

• Development of the Pan-PreClinical (PPC) project, a drug screen atlas comprising 2.1 million experiments.
• Utilizing 1,982 ex vivo samples and 3,100 drugs across 134 cancer indications from 26 studies.
• Application of a contrastive Bayesian model for data harmonization and analysis.

Main Results:

• Identification of 303 tissue-specific drug sensitivities within the PPC atlas.
• Demonstration that ex vivo drug sensitivities predict clinically relevant molecular profiles.
• Uncovered systematic biases in cell line screens, favoring proliferation targets over cell-cell communication targets across 55 cancer subtypes.

Conclusions:

• The PPC project provides a foundational ex vivo dataset and computational platform for advancing cancer biology research.
• The findings highlight the importance of ex vivo models for accurate drug response prediction in oncology.
• Addressing biases in current models is crucial for developing more effective and targeted cancer therapies.