Molecular Fingerprints 2D: Morgan/ECFP Circular Hashing

Before a neural network or a similarity search can compare two molecules, each one first has to become a fixed-length vector of numbers — and the dominant way to do that in cheminformatics is the Morgan / Extended-Connectivity Fingerprint (ECFP) algorithm. This 2D simulator lays out two randomly generated molecular graphs with a live spring-mass force-directed physics simulation — every atom repels every other atom, every bond behaves as a damped spring toward its rest length — the same class of algorithm used to draw real 2D skeletal structures, then runs the real circular-substructure hashing procedure on each: every atom's identifier is folded together with its neighbors' identifiers, radius by radius, and every identifier produced along the way is hashed into a bit of a fixed-length fingerprint vector. Click any atom on molecule A to run a real breadth-first search over the bond graph and shade every atom by its exact hop-distance from the click — the same atoms whose hashes were combined to produce that identifier — and watch the two bit vectors and their Tanimoto similarity update live as you change the radius, the fingerprint length, or regenerate either molecule.