Abstract: To achieve rapid mining of key factors influencing the contents of chemical composition in tobacco leaves and to improve the predictive ability of the models for chemical components in tobacco leaves, a method for correlation mining and chemical content prediction in tobacco leaves based on causal discovery was developed. Taking the meteorological and chemical component data of tobacco leaves in Henan Province from 2010 to 2020 as materials, the causal relationship between meteorological factors and chemical components in tobacco leaves was mined by combining the Graphical Lasso method with the Information Geometric Causal Inference (IGCI) causal inference method. In addition, a causal weight transformation mechanism was proposed based on the causal adjacency matrix obtained from the causal relationship to improve the predictability of the models. The results showed that there was a correlation between the meteorological factors and chemical components in tobacco leaves studied, and the correlation results were consistent with the results of agricultural experiments. A causal weight transformation mechanism was established, and the prediction accuracy of the deep learning models was improved for over 83% of the samples in predicting the chemical components in the leaves of cv. Yunyan 87 and cv. Zhongyan 100. This method provides support for mining factors influencing chemical components in tobacco leaves and predicting chemical component contents.