Predicting the glass-transition temperatures (Tg) of glass-forming polymers is of critical importance as it governs the thermophysical properties of polymeric materials. The cheminformatics approaches based on machine learning algorithms are becoming very useful in predicting the quantitative relationships between key molecular descriptors and various physical properties of materials. In this work, we developed a modeling framework by integrating the cheminformatics approach and coarse-grained molecular dynamics (CG-MD) simulations to predict Tg of a diverse set of polymers. The developed machine learning-based QSPR model identified the most prominent molecular descriptors influencing the Tg of a hundred of polymers. Informed by the QSPR model, CG-MD simulations are performed to further delineate mechanistic interpretation and systematic dependence of these influential molecular features on Tg by investigating three major CG model parameters, namely the cohesive interaction, chain stiffness, and grafting density. The CG-MD simulations reveal that the higher intermolecular interaction and chain stiffness increase the Tg of CG polymers, where their relative influences are coupled with the existence of side chains grafted on the backbone. This synergistic modeling framework provides valuable insights into the roles of key molecular features influencing the Tg of polymers, paving the way to establishing a materials-by-design framework for polymeric materials via molecular engineering. Display omitted •A novel modeling framework was developed by integrating cheminformatics and coarse-grained molecular dynamics to predict the Tg of polymers.•The molecular features from machine learning model are grouped into three MD parameters - cohesive energy, chain stiffness, and grafting density.•The coarse-grained MD complements the machine learning model for a mechanistic and systematic interpretation of key molecular features.•The mutual and competing influence of cohesive energy vs. chain stiffness and grafting density on Tg is analyzed using coarse-grained MD