This technology is a machine learning algorithm for predicting the outcome of clinical trials using tissue-specific toxicity scores of drugs.
Currently available methods to predict drug toxicity typically use either chemical structure or transcriptome profiling-based approaches. However, neither of these approaches are effective at correctly predicting the outcome. Part of the reason for these limitations is the inherent difficulty in predicting toxicity due to the presence of numerous interconnected factors governing how a drug interacts with the body. As such, methods that offer substantial improvements in the ability to predict the failures and successes of clinical trials are needed, particularly those that integrate personal data, such as transcriptomics.
This technology is a random forest model to predict clinical trial success using tissue-specific toxicity scores of drugs. These tissue toxicity scores are calculated systematically using mRNA expression, genetic variation tolerance, and pharmacological pathways. The model is built and trained using data from ClinicalTrials.gov, and predictive performance is evaluated using out-of-bag probability. As a result, this technology offers an effective, and potentially powerful, method for enabling more efficient drug development.
This algorithm has been found to outperform chemical structure and transcriptome-based prediction methods, and has predicted the failure of several drugs currently undergoing trials.
Patent Pending (US20210134402)
IR CU18390
Licensing Contact: Kristin Neuman