Comparison of average cosine similarities between candidate and detected molecules (intergroup, circles) and among detected molecules themselves (intragroup, squares), shown as a function of atom number and charge state (anions: −1; neutral: 0; cations: 1). The similarity is defined as the cosine similarity between ECFP vectors for pairs of molecules. For each molecule, the similarity to all molecules in the detected molecule set is calculated and averaged; these averages are then grouped and averaged again over molecules with the same number of atoms.