Researchers at Hangzhou Lin’an Traditional Chinese Medicine Hospital built a three-stage computational pipeline that screened 22,823 compounds for dual activity against CDK4 and CDK6, then reported the results in Molecular Diversity.
The first stage was ligand-based machine learning. Working from ChEMBL bioactivity data covering 265 CDK4 compounds and 402 CDK6 compounds, the team encoded each molecule as an ECFP4 fingerprint and benchmarked several algorithms. A Bayesian Ridge regressor won, reaching cross-validated R-squared values of 0.731 for CDK4 and 0.721 for CDK6.
From scores to a molecule
Docking the survivors against both kinases narrowed the field to three candidates for bench testing. One, HY-18,623, returned an IC50 of 3.5 nanomolar against CDK4 and 17.4 nanomolar against CDK6.
Two-hundred-nanosecond molecular dynamics runs then showed the compound holding persistent hydrogen bonds with hinge residues Val96 and Val101, the anchoring interaction typical of kinase inhibitors.
Approved CDK4/6 drugs such as palbociclib and ribociclib reshaped breast cancer treatment, yet resistance and side effects keep demand alive for chemically different scaffolds. The team released its data and simulation code publicly.
