Artificial neural networks (ANN) are one of the main tools used in machine learning. ANN have been used in variety of fields, including computer vision, speech recognition, medical diagnosis, playing board games and also computational chemistry. Here we present a new ANN for deriving complex collective variables (CVs) just from atomistic coordinates. This ANN takes a molecular trajectory (a real or an artificial one) and a set of CVs, which are (by their nature) difficult to calculate from atomistic coordinates. The output is a series of mathematical transformations starting with positions of atoms and ending with the value of this complex CV. Everything is produced in the form of Plumed syntax and can be directly used for metadynamics simulations using popular Plumed package. Solvent accessible surface area, distances calculated by Dijkstras algorithm and Rosetta score were tested as examples of complex CVs. The correlation between the values of original CVs and values encoded by ANN was above 0.90, in case of Rosetta score, and above 0.99, in case of two other CVs. Our new ANN is freely available in the form of Python code on GitHub (https://github.com/spiwokv/anncolvar). In future, we would like to investigate the performance of the above mentioned collective variables.