Molecular Properties Prediction Based On Machine Learning

Prediction of molecular properties is an essential problem and has attracted much attention in relevant files.It could accelerate research progress in the application of drug design,substance discovery and food development.However,traditional studies based on density functional theory(DFT)in physics are proved to be time-consuming for predicting large number of molecules.Besides,in existing datasets,most molecules are unlabeled.In this paper,firstly,we proposed a generalizable Multilevel Graph Convolutional Neural Network(MGCN)for molecular property prediction.Specifically,we represent each molecule as a graph to preserve its internal structure.Moreover,the well-designed hierarchical graph neural network,directly extracts features from the conformation and spatial information followed by the multilevel interactions.As a consequence,the mul-tilevel overall representations can be utilized to make the prediction.Extensive ex-periments on several datasets demonstrate the effectiveness of our model and prove that MGCN is generalizable and transferable.Secondly,we proposed a novel Active Semi-supervised Graph Neural Network(ASGN)by incorporating both labeled and un-labeled molecules.It adopts a teacher-student framework consisting of two sub-models that these model work iteratively.Some techniques like weight transfer are used to connect two models for accelerating training.Moreover,we proposed a novel active learning strategy to select informative data to improve the labeling efficiency.We con-duct experiments under various experimental settings,the results show the remarkable performance and interpretability of our ASGN framework.At last,to show the power of machine learning in predicting molecular properties,we advocate the use of MGCN to demonstrate our model is capable of accurately predicting various electronic prop-erties of an important class of Organic Semiconductors(OSCs).We construct a work flow from data generation,model training,properties prediction to the application to model the UV-Vis absorption spectra of OSCs in dichloromethane.The results show that,the proposed model could achieve a good agreement between the calculated and experimental spectra.