This letter presents results on improving protein-ligand binding affinity predictions using molecular dynamics simulations with machine learning potentials, specifically a hybrid neural network potential and molecular mechanics methodology (NNP/MM). Relative binding free energies (RBFE) were computed using the Alchemical Transfer Method (ATM) and validated against established benchmarks, showing significant improvements over conventional MM force fields like GAFF2. The study evaluated a series of targets from Wang's and Schindler's datasets. Due to limitations of ANI-2x, a subset of targets was selected. The workflow involved parameterizing ligands with GAFF2, preparing complex systems with HTMD, and performing energy minimization, thermalization, and equilibration. Simulations were run using both GAFF2 and NNP/MM approaches, with the latter using an NNP for the small molecule and MM for the rest. The classical RBFE simulations (GAFF2) were run in triplicate for each ligand pair, while NNP/MM simulations were performed with a 1fs timestep. The results showed that NNP/MM outperformed pure MM runs in both relative and absolute measures, with lower MAE and higher correlation coefficients. However, NNP/MM calculations are slower than conventional MM calculations. The study found that NNP/MM significantly improved the accuracy of binding free energy predictions, particularly for ligands that were poorly predicted by GAFF2. The results demonstrated that NNP/MM can predict a higher percentage of ligands with MAE below 1 and 1.5 kcal/mol. However, the current NNP is limited to neutral molecules and a limited set of elements, restricting its application. Future work should focus on expanding the applicability of NNPs to include charged ligands and improving computational performance. The study also evaluated the impact of different timesteps on the accuracy of RBFE calculations, finding no significant difference between 2fs and 4fs timesteps. The results highlight the potential of NNP/MM for accurate RBFE calculations and the importance of further advancing NNPs to encompass a broader range of molecular species.This letter presents results on improving protein-ligand binding affinity predictions using molecular dynamics simulations with machine learning potentials, specifically a hybrid neural network potential and molecular mechanics methodology (NNP/MM). Relative binding free energies (RBFE) were computed using the Alchemical Transfer Method (ATM) and validated against established benchmarks, showing significant improvements over conventional MM force fields like GAFF2. The study evaluated a series of targets from Wang's and Schindler's datasets. Due to limitations of ANI-2x, a subset of targets was selected. The workflow involved parameterizing ligands with GAFF2, preparing complex systems with HTMD, and performing energy minimization, thermalization, and equilibration. Simulations were run using both GAFF2 and NNP/MM approaches, with the latter using an NNP for the small molecule and MM for the rest. The classical RBFE simulations (GAFF2) were run in triplicate for each ligand pair, while NNP/MM simulations were performed with a 1fs timestep. The results showed that NNP/MM outperformed pure MM runs in both relative and absolute measures, with lower MAE and higher correlation coefficients. However, NNP/MM calculations are slower than conventional MM calculations. The study found that NNP/MM significantly improved the accuracy of binding free energy predictions, particularly for ligands that were poorly predicted by GAFF2. The results demonstrated that NNP/MM can predict a higher percentage of ligands with MAE below 1 and 1.5 kcal/mol. However, the current NNP is limited to neutral molecules and a limited set of elements, restricting its application. Future work should focus on expanding the applicability of NNPs to include charged ligands and improving computational performance. The study also evaluated the impact of different timesteps on the accuracy of RBFE calculations, finding no significant difference between 2fs and 4fs timesteps. The results highlight the potential of NNP/MM for accurate RBFE calculations and the importance of further advancing NNPs to encompass a broader range of molecular species.