ICML 2022 | limo: a new method for rapid generation of targeted molecules

Today, I'd like to introduce a team from the University of California in the United States published on ICML2022 Articles on . In this paper, a new method which can accelerate the formation of molecules is proposed LIMO Model .LIMO A mutation auto coder is used to generate the potential representation of molecules , And through the network to predict the properties of molecules , To achieve faster gradient based inverse optimization of molecular attributes . Comprehensive experiments show that ,LIMO Be competitive on benchmark tasks , It is obviously superior to the current state-of-the-art technology in the new task of generating drug like compounds with high binding ability , And the binding force of the two protein targets reaches the nanomolar range . The authors use a more accurate calculation of the absolute binding free energy based on molecular dynamics , The results show that the generated molecules are based on docking , It also shows that a drug like compound generated by the model can predict the human estrogen receptor K D value （ Measurement of binding affinity ） Far more than the early typical candidate drugs and most FDA Affinity of approved drugs to their respective targets .

 

Thesis link ： https://arxiv.org/abs/2206.09010