D3EGFR: a webserver for deep learning-guided drug sensitivity prediction and drug response information retrieval for...

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Epidermal growth factor receptor (EGFR) mutations are key drivers in non-small-cell lung cancer (NSCLC).
• Variable drug sensitivities to targeted therapies present a challenge for precision medicine in NSCLC.
• Clinical decision support for drug recommendations in EGFR-mutated NSCLC is highly needed.

Purpose of the Study:

• To develop a comprehensive database (D3EGFRdb) of clinicopathologic characteristics and drug responses for patients with EGFR mutations.
• To create a deep learning-based prediction model (D3EGFRAI) for accurate drug sensitivity prediction in EGFR-driven NSCLC.
• To establish a novel platform for clinical-level drug response prediction for all approved small molecule drugs in EGFR-mutated lung cancer.

Main Methods:

• Literature mining was employed to compile the D3EGFRdb database, including data from 1339 patients with EGFR mutations.
• A deep learning model, D3EGFRAI, was developed using the D3EGFRdb database for drug sensitivity prediction.
• The D3EGFRAI model was validated using patient data from the database and internal hospital cohorts.
• In silico mutation scanning of critical residues within drug-binding pockets was conducted to assess potential impacts on drug sensitivity.

Main Results:

• The D3EGFRdb database contains clinicopathologic characteristics and drug responses for 1339 patients with EGFR mutations.
• The D3EGFRAI model achieved prediction accuracies of 0.81 for patients in D3EGFRdb and 0.85 for patients from the hospitals.
• The study explored potential future mutations and their effects on drug sensitivity.
• D3EGFR represents the first platform offering clinical-level drug response predictions for all approved small molecule drugs in EGFR-mutated lung cancer.

Conclusions:

• D3EGFRdb and D3EGFRAI provide a valuable resource for understanding EGFR mutation-driven NSCLC and predicting drug responses.
• The developed AI model demonstrates high accuracy in predicting drug sensitivity, supporting personalized treatment strategies.
• This platform facilitates clinical decision-making by offering reliable drug recommendations for NSCLC patients with specific EGFR mutations.
• The D3EGFR platform is freely accessible, promoting advancements in precision medicine for lung cancer treatment.