The inter-atomic potential based on artificial neural network (ANN) is very promising tool for atomic modeling. Using high quality training set constructed by First principles calculation based on Density functional theory (DFT), ANN potential would become sophisticated replica of DFT. However, the transferability of ANN potential depends on the DFT training dataset. In this study, we perform molecular dynamics (MD) calculations for molten iron, which is not directly included in the training data. We also examine the results to discuss the applicability of the constructed potential.