Developing and validating a drug recommendation system based on tumor microenvironment and drug fingerprint

Area of Science:

• Computational biology
• Genomics
• Pharmacogenomics

Background:

• Tumor heterogeneity complicates cancer treatment selection due to variable patient responses.
• Personalized cancer therapy offers a strategy to improve treatment effectiveness and precision.

Purpose of the Study:

• To develop a personalized drug recommendation model using genomic profiles.
• To optimize therapeutic outcomes by predicting drug sensitivity.

Main Methods:

• A content-based filtering algorithm and multilayer perceptron (MLP) models were used to predict IC50 values.
• Trained and validated on GDSC and CCLE datasets, with clinical application assessed using TCGA data.
• Tumor microenvironment (TME) and drug fingerprints characterized patient and drug features, respectively.

Main Results:

• The model achieved high predictive accuracy (R=0.914 training, R=0.902 test).
• Robust predictions for cytotoxic drugs (Docetaxel, Cisplatin), with weaker performance for targeted therapies.
• Successful prediction of clinical outcomes in TCGA data, including association with progression-free survival (PFS).

Conclusions:

• The model shows strong performance across preclinical datasets, indicating potential for real-world personalized cancer therapy.
• This approach bridges preclinical drug sensitivity data (IC50) with clinical response endpoints (BOR), offering a tool for optimizing patient-specific treatments.