Interpretable deep learning for improving cancer patient survival based on personal transcriptomes.

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Precision medicine aims to personalize treatment by considering individual patient variability.
• Cancer research generates vast amounts of data, including genomic and drug prescription information.
• Predicting patient survival is crucial for treatment planning and drug development.

Purpose of the Study:

• To develop an interpretable deep learning model (CancerIDP) for predicting cancer patient survival.
• To utilize drug prescriptions and personal transcriptomes as input features for survival prediction.
• To identify potential alternative medicines that could improve patient survival outcomes.

Main Methods:

• Development of an interpretable neural network model named CancerIDP.
• Input features include patient drug prescriptions and transcriptome data.
• Model performance evaluated using classification accuracy and Pearson correlation for survival time prediction.

Main Results:

• CancerIDP achieved 96% accuracy in distinguishing between short-lived and long-lived cancer patients.
• A high Pearson correlation of 0.937 was observed between predicted and actual months-to-death.
• The model identified alternative medicines that could potentially benefit 27.4% of patients, increasing median survival by 3.9 months.

Conclusions:

• The interpretable deep learning model CancerIDP effectively predicts cancer patient survival.
• Personalized medicine approaches using CancerIDP can identify alternative treatments to improve survival.
• The model's interpretability facilitates mechanistic studies for novel cancer drug development.