In response to the increasing concern over antibiotic resistance and the limitations of traditional methods in antibiotic discovery, we introduce a novel machine learning based method named MFAGCN, which predicts the antimicrobial efficacy of molecules by integrating MACCS molecular fingerprints and molecular graph representations as input features, with a focus on molecular functional groups. MFAGCN incorporates an attention mechanism to assign different weights to the importance of information from different neighboring nodes. Comparative experiments with baseline models on two public datasets demonstrate MFAGCN's superior performance. Additionally, structural similarity analyses with known antibiotics are conducted to prevent the rediscovery of established antibiotics. This approach enables researchers to rapidly screen molecules with potent antimicrobial properties and facilitates the identification of functional groups that influence antimicrobial performance, providing valuable insights for further antibiotic development.