Sepsis is a fatal organ dysfunction characterized by the simultaneous hyperinflammation and immunosuppression. Nowadays, the early precision intervention of sepsis is challenging. Ferroptosis is involved in the development of sepsis. The current study aimed to find out the signature genes of sepsis with network topology analysis and machine learning, and further provide the potential natural compounds for sepsis with virtual screening and in vitro validation. In this study, five genes namely CGAS, DPP4, MAPK14, PPARG and TXN were identified as ferroptosis-related signature genes for sepsis by network topological analysis, machine learning algorithms, and external datasets verification. The results of immune infiltration analysis confirmed these genes were significantly associated with the infiltration abundance of some immune cells including neutrophil, macrophage, plasmacytoid dendritic cell and activated dendritic cell. Moreover, coniferin, 5-O-caffeoylshikimic acid, and psoralenoside were initially identified as the natural inhibitors of CGAS by virtual screening. However, further in vitro study on macrophages revealed coniferin and psoralenoside had better inhibitory activities on CGAS. In summary, the present study pointed out the importance of CGAS in sepsis, and discovered novel natural inhibitors of CGAS.