Selective Cleaning Enhances Machine Learning Accuracy for Drug Repurposing: Multiscale Discovery of MDM2 Inhibitors.

Area of Science:

• Oncology
• Pharmacology
• Computational Biology

Background:

• Cancer treatment necessitates novel therapeutic strategies, including the reactivation of tumor suppressor genes like p53.
• MDM2 is a key negative regulator of p53, making it a critical target for cancer therapy.
• Drug repurposing offers an efficient pathway to identify new therapeutic agents.

Purpose of the Study:

• To identify novel MDM2 inhibitors for cancer treatment using a drug repurposing strategy.
• To develop and apply an advanced data-cleaning algorithm to improve machine learning model accuracy for predicting drug potency.
• To prioritize candidate compounds through integrated virtual screening and molecular dynamics simulations.

Main Methods:

• Screening of over 24,000 clinically tested molecules against MDM2.
• Development of a selective data-cleaning algorithm to enhance bioactivity dataset quality.
• Machine learning model development for pIC50 prediction, validated by structure-based virtual screening and molecular dynamics simulations.

Main Results:

• A novel data-cleaning algorithm significantly improved machine learning model performance (RMSE reduced by 21.6%, R² = 0.87).
• Atorvastatin and two CB1 antagonists (MePPEP, otenabant) were identified as promising MDM2 inhibitor candidates.
• In silico analyses indicated potential multi-target mechanisms involving MDM4 and BCL2 for enhanced p53 restoration.

Conclusions:

• The developed AI-driven workflow, featuring a novel data-curation strategy, enables efficient drug repurposing for challenging cancer targets.
• Identified compounds show promise for reactivating p53 and warrant further investigation as anti-cancer therapeutics.
• This approach demonstrates the power of integrating advanced data preprocessing with AI for accelerated drug discovery.