University of Pittsburgh researchers have developed DeepRx, a deep-learning method for predicting anti-cancer drug sensitivity using omics data from individual tumors. Cancer is a heterogeneous disease, and precision oncology aims to identify and target tumor-specific aberrations. Traditional single-gene-based therapeutic indications (SGTI) are often inaccurate. DeepRx leverages deep learning models to transform raw omics data into features reflecting the states of cellular signaling systems, significantly improving the prediction of drug sensitivity. This approach has the potential to benefit millions of patients by expanding the application of existing anti-cancer drugs.
Description
DeepRx utilizes deep learning models, specifically deep belief networks, to analyze omics data from cancer cell lines. The method involves transforming raw transcriptomic features into binary features that capture the covariance structures of co-differentially expressed genes. The deep learning model consists of multiple hidden layers, each representing different levels of abstraction. By combining genomic data with transcriptomics-derived features, DeepRx can predict drug sensitivity with high accuracy. The technology outperforms traditional SGTI approaches, offering a more reliable and comprehensive method for precision oncology.
Applications
• Predicting anti-cancer drug sensitivity
• Precision oncology
• Personalized treatment plans
• Cancer research
Advantages
DeepRx offers several advantages by significantly improving the prediction of drug sensitivity compared to traditional SGTI approaches. It leverages deep learning to capture the activation states of cellular signaling pathways, providing a more accurate and comprehensive analysis. This method can expand the application of existing anti-cancer drugs to a larger patient population, potentially benefiting millions of patients. The technology is highly predictive, with models achieving high sensitivity and positive predictive value.
Invention Readiness
The software for DeepRx is developed and validated. The technology is ready for further development and commercialization, with demonstrated high accuracy in predicting drug sensitivity using omics data.
