AI tool dramatically reduces computing power needed to find protein-binding molecules

A new machine learning tool called DrugCLIP dramatically speeds up the identification of small-molecule candidates that can bind specific proteins, requiring significantly less computational power compared to previous methods.
Researchers at Tsinghua University developed the DrugCLIP framework, which uses a deep-learning algorithm to predict binding affinities, making it up to 10 million times faster than traditional docking methods.
The method involves representing protein pockets and small molecules as vectors in high-dimensional space, calculating the scalar product to predict binding affinity, and validating the top candidates with AlphaFold3.
Using the new tool, researchers identified potential drug candidates for the serotonin 2A receptor and the norepinephrine transporter, with one molecule targeting the norepinephrine transporter showing promising chemical effectiveness.
The team discovered a new molecule that they hope to take to clinical trials, although they caution about the uncertainty of clinical effectiveness.
Nicholas Polizzi at Harvard Medical School sees the potential of high-throughput, genome-wide screening for identifying ligands with minimal off-target effects, though he raises concerns regarding the generalizability of the deep-learning algorithm.

AI tool dramatically reduces computing power needed to find protein-binding molecules | News