Suggestions for applying machine learning to 3D data?

Im rather new to machine learning. I'm applying neural networks to a problem only because thats the tool I have. I am open to other machine learning approaches.

I have 3D positions of atoms and Im looking at what happens when a few are changed. Thus far my approach is going to be to grid out the 3D space Im interested in and assign a vector to each grid cell that counts the number of each type of atom in that cell then I will construct my input vector using these cell vectors and an additional three-vector that consists of the 3D coordinates for the center of each cell. At the end of this input vector will be another small vector that represents the changes I made and the target vector will be a simple number that represents the measured consequence of that change.

This is mostly off the top of my head with little experience in machine learning. If some of you have suggestions or know of a different method I should be looking at please let me know.