Quantifying tumor specificity using Bayesian probabilistic modeling for drug target discovery and prioritization.

Area of Science:

• Oncology
• Bioinformatics
• Pharmacology

Background:

• Developing new cancer therapeutics requires predicting potentially lethal "off target" effects, often identified through costly and risky testing.
• Current methods for prioritizing drug targets lack standardization, relying on manual inspection and ad-hoc thresholds, which are further complicated by data sensitivity and accuracy issues.
• Comprehensive molecular tissue atlases offer a path to predict toxicity by analyzing normal RNA and protein expression, but require robust analytical tools.

Approach:

• We introduce a Bayesian Tumor Specificity (BayesTS) score, a probabilistic method to quantify tumor specificity.
• BayesTS integrates multiple molecular evidence types, including RNA-Seq and protein expression, while accounting for inference uncertainty.
• The score was applied to 24,905 human genes across 3,644 normal tissue samples from GTEx and TCGA datasets.

Key Points:

• BayesTS accurately combines RNA, protein, and tissue distribution data, effectively managing inference uncertainty.
• The approach successfully prioritized known drug targets and de-emphasized those later linked to toxicity.
• Customizable tissue importance weights allow for clinically relevant prioritization, focusing on tissues like reproductive organs.

Conclusions:

• BayesTS offers a standardized, quantifiable approach to drug target prioritization in oncology.
• This method facilitates novel drug target discovery and can be extended to unconventional targets like splicing neoantigens.
• The BayesTS score and associated code are publicly available, promoting improved and safer oncology drug development.