Generating Potential RET-Specific Inhibitors Using a Novel LSTM Encoder-Decoder Model.

Area of Science:

• Oncology
• Computational Chemistry
• Drug Discovery

Background:

• Receptor tyrosine kinase RET is crucial for nervous system development.
• Abnormal RET activation drives cancers like thyroid and non-small-cell lung cancer.
• Existing RET inhibitors are often multi-kinase, lacking specificity.

Purpose of the Study:

• To develop a novel molecular generation model for creating RET-specific inhibitors.
• To overcome limitations of current multi-kinase inhibitors.
• To accelerate the discovery of targeted cancer therapeutics.

Main Methods:

• Utilized fragment-based drug design (FBDD) and a long short-term memory (LSTM) encoder-decoder model.
• Trained the model for high molecular assembly accuracy (98.4%).
• Employed transfer learning to generate a RET-specific molecule library, followed by virtual screening, molecular dynamics, and binding energy calculations.

Main Results:

• Generated a library of RET-specific candidate molecules.
• Identified five novel RET inhibitor candidates.
• Two molecules demonstrated favorable binding affinities, synthesizability, and high selectivity.

Conclusions:

• The developed model effectively generates novel, receptor-specific molecules.
• This AI-driven approach offers a rapid and efficient method for discovering potential drug candidates.
• The identified molecules show promise as selective RET inhibitors for cancer treatment.