Classification of FLT3 inhibitors and SAR analysis by machine learning methods

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• FMS-like tyrosine kinase 3 (FLT3) is a crucial target in anti-cancer therapy.
• Understanding structure-activity relationships (SAR) is vital for developing effective FLT3 inhibitors.

Purpose of the Study:

• To perform a comprehensive SAR study on a large dataset of FLT3 inhibitors.
• To build predictive models for FLT3 inhibition activity.
• To identify key structural fragments and scaffolds associated with FLT3 inhibition.

Main Methods:

• Collected and curated a dataset of 3867 FLT3 inhibitors.
• Utilized MACCS, ECFP4, and TT molecular fingerprints for feature representation.
• Developed 36 classification models using Support Vector Machine (SVM), Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Deep Neural Networks (DNN).
• Applied K-Means clustering to group inhibitors and analyzed SAR using RF and ECFP4 fingerprints.

Main Results:

• A Deep Neural Network model using TT fingerprints achieved the highest prediction accuracy (85.83%) and Matthews Correlation Coefficient (MCC) of 0.72.
• Identified key active fragments including 2-aminopyrimidine, 1-ethylpiperidine, and alkynyl groups.
• Discovered significant FLT3 inhibition activity associated with three specific scaffolds across different subsets.

Conclusions:

• The study successfully developed a highly accurate predictive model for FLT3 inhibition.
• Identified critical structural motifs that can guide the design of novel and potent FLT3 inhibitors.
• Provides valuable insights into the SAR of FLT3 inhibitors for future drug development efforts.