Molecular design based on Q-learning and maximum likelihoodestimation

Aiming at the complex problems during developing new drugs in pharmaceutical factories, such as the optimization of the hydrophobic constant of a molecule and the water solubility of a molecule, a molecular design model based on Q-leaming and maximum likelihood estimation was developed in this paper. The proposed model introduces the idea of reinforcement learning and designs the learning criterion based on Q-leaming method. By constantly updating the state action value function to select the optimal action under the current strategy, the problem of grammar errors in smiles is overcome. Compared with the deep reinforcement learning based on RNN and the traditional Q-leaming method, the experimental results show that this model can effectively optimize the chemical properties of the molecules on the basis of avoiding the grammatical errors of the smiles, and is more efficient.