I understand the objective to derive molecular properties from an algorithm of atomic components, science is still a very long ways off. Perhaps you may think of it as similar to sitting a very large number of monkeys at typewriters in hopes of randomly producing one of Shakespeare's plays. It is just very improbable. Even if we know a great deal more about structural elements, we cannot determine which amino acids catalyze a reaction from the primary sequence of a protein. I think if you were to add vector properties for structures and reactions, you would find the mathematical possibilities will become very very large. I have a reasonable amount of experience in predicting the products of reactions, yet there are many reactions that I cannot reliably predict the products. If (organic) chemistry were sufficiently predictable, chemists would have devised a system of robots to carry out the reactions. There are some areas of chemistry where this can be done, but not in a general sense.
I think the way to approach your question is to try to find problems that one might find an answer to. That is, you probably won't find many people discussing their monkey results in trying to write plays because it is too improbable. You can find people are searching for ways to predict, mp, bp, bond lengths, activation energies, bond energies, preferred conformations, self assembly systems, etc, etc. There is nothing wrong with the question. You need to have some knowledge as to how it might be answered. For example, you might ask whether one could predict some properties of bimolecular combinations. If you look at the range of properties, this will begin to teach you how formidable that problem might be. Don't give up just yet, limit the types of atoms, can you make a prediction? That is how science operates.