Graph convolutional network approach to investigate potential selective Limk1 inhibitors

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• LIM kinase 1 (Limk1) is a crucial drug target, but selective inhibitors are scarce, hindering disease treatment.
• Developing novel scaffolds for selective Limk1 inhibitors is essential for advancing therapeutic options.

Purpose of the Study:

• To identify potential and selective Limk1 inhibitors using computational screening.
• To develop and validate a predictive model for Limk1 and ROCK2 inhibitory activity.

Main Methods:

• Molecular docking was employed to screen compounds from the Traditional Chinese Medicine (TCM) database.
• A three-dimensional graph convolutional network (3DGCN) was developed for predicting inhibitory activity against Limk1 and ROCK2.
• Molecular dynamics (MD) simulations were used to assess the stability and binding modes of potential inhibitors.

Main Results:

• The 3DGCN model demonstrated superior accuracy compared to baseline models, with averaged RMSE values of 0.721 for Limk1 and 0.852 for ROCK2.
• Over 80% of test molecules were predicted with an absolute error of 1.0 or less by the 3DGCN.
• MD simulations confirmed the stability of ligand-protein complexes, revealing binding modes for potential Limk1 inhibitors.

Conclusions:

• The study successfully identified potential selective Limk1 inhibitors, namely compounds 7549 and 2007_15649.
• The integrated computational approach combining docking, 3DGCN, and MD simulations provides a robust strategy for drug discovery.
• These findings pave the way for developing new treatments targeting Limk1-related diseases.