Antibacterial secondary metabolites such as tannins and their derivatives are found in the Vernonia amygdalina Del. Antibiotic resistance can develop due to overuse, reducing the efficacy of drugs to prevent and treat infections. This research aims to use the Quantitative Structure-Activity Relationship (QSAR) and the semi-empirical method Austin Model 1 (AM1) to design a modified novel compound from African leaves that has improved antibacterial activity. This research includes a descriptor calculation of QSAR using AM1 MOE on typical compounds from African leaves, and calculation results are chosen based on a multilinear regression statistical analysis. The model equation represents the three primary parameters of QSAR, which are electronic, hydrophobic, and steric parameters, which will be used to measure modified compounds. Molecular docking using Autodock Tools (The Scripps Research Institute, USA), and analysis of results of docking Autodock Tools using Discovery Studio 3.5 Client. The best QSAR model obtained is LogEC50 = (0.829 x LogP) - (1,302 x AM1_HOMO) - (0.339 x AM1_dipole) - (5,128 x mr) + (0.145 x vol) - (11,355). The results showed that EC50 prediction of modified hamamelitannin has the best activity with the lowest ΔGbind -9.0 kcal/mol and inhibition constant of 0.249 μM. In summary, the novel compound's design calculation has better antibacterial activity, as indicated by a lower EC50, than fosfomycin or compounds without modification. The modified hamamelitannin compound was found to have better antibacterial activity (prediction EC50 = 0.1933 μM) than the original (experimental EC50 = 145.50 μM).