I'm a patent holder for a pharmaceutically relevant class of compounds. There are two types of patents typically seen in pharma. One is procedural (how the drug is synthesized, isolated, formulated or administered) and another is composition of matter (the atoms in the molecule, their relative coordinates and the lengths and types of bonds that connect them). This program takes a target compound and does two things. 1) produces a library of similar compounds by structural diversification (making substitutions of atoms or small groups of atoms in the target compound with other atoms or groups which are known to behave similarly) with restrictions based on what has been patented under composition of matter. 2) It then takes each molecule and looks for ways it can be synthesized. It does this by breaking apart the molecule piece by piece until it obtains commercially available building blocks. These piecewise separations correspond to synthetic a steps in the other direction, which the program screens for literature precedence. It does this for every possible combination of piecewise separations until it finds a set of viable pathways from available compounds to the target. This is called retro synthesis. It then filters the potential synthetic routes for those that are covered by procedural patents until it has a list of non patented, commercially viable synthesis plans.

As you can imagine there are thousands of ways a moderately complex molecule could be deconstructed and the bank of known reactions is more than any one person can really grasp. That and the myriad patent literature and how cryptic and dense it can be make the problem particularly suited for algorithmic treatment. This is only being done now because cataloging all the research and patent literature (which goes back to the late 1800s) and digitally formatting in a way that allows it to be computationally analyzed and processed requires manual translation of each report by a human being. Not many people have the expertise required for this and those that do usually prefer less menial work. So it has taken this long to amass a digital library of sufficient size to give one confidence that the answers it provides are comprehensive and that further searching would be pointless.