Individualized Prediction of Drug Response and Rational Combination Therapy in NSCLC Using Artificial

Elizabeth A Coker1,2,3, Adam Stewart4,5, Bugra Ozer1,3

  • 1Department of Data Science, The Institute of Cancer Research, London, United Kingdom.

Insights

Acute protein changes from targeted anticancer drugs predict non-small cell lung cancer (NSCLC) sensitivity. Proteomics data offers a novel approach to guide personalized NSCLC treatment selection.

Area of Science:

  • Oncology
  • Pharmacology
  • Proteomics

Background:

  • Targeted anticancer drugs are crucial for non-small cell lung cancer (NSCLC) treatment.
  • Predicting drug sensitivity and combination therapy effectiveness remains a challenge.
  • Current biomarkers (gene mutations) have limitations in predicting treatment response.

Purpose of the Study:

  • To investigate if acute protein perturbation by targeted anticancer drugs can predict drug sensitivity in NSCLC.
  • To develop machine-learning models for predicting single-agent and combination therapy response.
  • To assess the feasibility of using ex vivo patient samples for drug response assays.

Main Methods:

  • Assayed dynamic changes in 52 phosphoproteins after 1-hour drug exposure in 35 NSCLC cell lines.
  • Evaluated drug sensitivities and synergy of 252 drug combinations.
  • Developed orthogonal machine-learning approaches for response prediction.
  • Validated predictions against experimental data and compared with gene mutation-based predictions.

Main Results:

  • Machine-learning models accurately predicted drug sensitivity quartiles (AUC 0.79-0.78), outperforming gene mutation-based predictions (AUC 0.5).
  • Model predictions for drug combinations showed a significant bias towards high synergy scores (P = 0.0243).
  • Demonstrated feasibility of performing these proteomic assays on patient-derived NSCLC cells from pleural effusions.

Conclusions:

  • Acute proteomic profiling can predict targeted drug sensitivity and combination therapy response in NSCLC.
  • Proteomics-based predictions offer a promising complement to traditional genetic biomarkers.
  • This approach provides a proof of concept for using ex vivo drug exposure to inform clinical treatment selection.