Molecular Graph-Based Deep Learning Algorithm Facilitates an Imaging-Based Strategy for Rapid Discovery of Small...

Area of Science:

• Biochemistry
• Computational Biology
• Neuroscience

Background:

• Biomolecular condensates are implicated in diseases like cancer and neurodegeneration.
• Conventional imaging screens for compounds affecting condensates have limitations in scale.

Purpose of the Study:

• To develop a scalable computational method for identifying small molecules targeting biomolecular condensates.
• To find compounds that reduce the phase separation of TAR DNA-binding protein 43 (TDP-43).

Main Methods:

• Utilized a graph convolutional network (GCN)-based deep learning algorithm.
• Applied the GCN model for spatial information extraction from molecular graphs.
• Combined imaging-based screening with the GCN computational approach.

Main Results:

• Identified small molecule candidates that reduce nuclear liquid-liquid phase separation of TDP-43.
• Demonstrated the suitability of GCN for spatial molecular information extraction.
• Validated that candidate compounds do not affect TDP-43 splicing function.

Conclusions:

• The GCN-based deep learning method is a promising approach for identifying novel small molecule scaffolds.
• Combining imaging and GCN methods significantly improves compound screening speed and accuracy for biomolecular condensates.
• This integrated approach aids in developing therapeutics for diseases linked to aberrant protein phase transitions.