Deep learning-based screening strategy for rapid identification of annilsoem cell-entry inhibitors

Authors

• Pen Gao
• Wei Zheng
• Min Shen

Abstract

Liquid-liquid phase separation (LLPS) has been proposed to cause diseases such as cancer and neurodegeneration by concentrating proteins at abnormal subcellular loci to form large protein assemblies at micrometer-scale. Compound screens have been proposed and carried out to identify small molecules that can reverse or promote liquid-liquid phase separation for the proteins of interest. However, limitations of imaging-based screening methods restrict the scale of the compound screen. Here we use graph convolutional network (GCN) based method to facilitate the small molecule discovery process and identified small molecule candidate that can reduce the nuclear LLPS of TDP-43, a protein that phase transits in several common neurodegenerative diseases. We demonstrated that GCN-based deep learning algorithm is suitable for spatial information extraction from molecular graph; thus, these kinds of methods may own high potential to identify small molecule candidates with high conformation consistency. Furthermore, we validated that these candidates change the LLPS of TDP-43 without affecting its normal nuclear function. Taken together, a combination of imaging-based screen and GCN-based deep learning method greatly improve the speed and accuracy of compound screen for LLPS.

Scientific Focus Area: Computational Biology

This page was last updated on Tuesday, August 6, 2024