Artificial neural networks are fitted to molecular dynamics trajectories using the Behler-Parrinello method of atom-centered symmetry functions in order to obtain analytical interatomic potentials. Molecular dynamics trajectories are generated using the Atomic Simulation Environment (ASE) and the neural networks are initialized and trained using the Atomistic Machine-Learning Package (AMP). AMP is interfaced with ASE through the Calculator interface, which is a black box that accepts atomic numbers and atomic positions and calculates the energy and, if implemented, forces and stresses. Neural network potentials are constructed for copper and silicon in equilibrium crystal structures, and are evaluated on the potential energy, energy conservation, radial distribution function and mean squared displacement, as well as the absolute errors of the potential energies and force components on the test trajectories. We find the neural networks are able to reproduce the crystal structures, but obtain negative results for the ability to conserve energy, leading to an increase in kinetic energy and translational momen- tum over time, with negative implications for long-term numerical stability. Recommendations for future work include better sampling algorithms for sampling likely configurations out of equilibrium, testing different numerical optimization algorithms and a more efficient implementation of the Behler- Parrinello symmetry functions for facilitating faster training and deployment of different architectures on available training data, as well as on new input data.