LIMO: Latent Inceptionism for Targeted Molecule Generation.

Area of Science:

• Computational Chemistry
• Artificial Intelligence in Drug Discovery
• Molecular Modeling

Background:

• Drug discovery is challenged by the slow, resource-intensive generation of high-affinity molecules.
• Current methods using reinforcement learning or deep generative models are often too slow for physics-based binding affinity calculations.

Purpose of the Study:

• To develop a novel AI method, Latent Inceptionism on Molecules (LIMO), for accelerated generation of drug-like molecules with high binding affinity.
• To significantly improve the speed and efficiency of identifying potent drug candidates.

Main Methods:

• LIMO utilizes a variational autoencoder latent space and sequential neural network property prediction.
• Employs an inceptionism-like technique for faster gradient-based reverse-optimization of molecular properties.
• Validates results using both docking and molecular dynamics simulations.

Main Results:

• LIMO demonstrates competitive performance on benchmark tasks.
• Significantly outperforms state-of-the-art methods in generating high-affinity drug-like compounds.
• Achieved nanomolar binding affinity against two protein targets, with one compound showing exceptional affinity (6 x 10^-14 M) for the human estrogen receptor.

Conclusions:

• LIMO offers a substantial advancement in accelerating the discovery of potent drug candidates.
• The method shows promise for identifying novel therapeutics with significantly improved binding affinities.
• Generated compounds exhibit binding affinities exceeding typical early-stage candidates and many approved drugs.