Apparently the RECAP implementation in RDKit cleaves bonds within ring motifs in the case of amines, e.g. in the case of imatinib or in the example given in the original paper.
I tried to modify the SMARTS pattern a bit in order to reproduce the cisapride example, without success. The original algorithm does not seem to work in a step-wise manner leading to a tree-like result. That is why a propyl group appears as a resulting fragment in the example, which is within the molecule in the first place but would be a terminal group in the RDKit implementation and therefore not be cleaved of course. Cisapride: c...@h]1cn(CCCOc2ccc(F)cc2)c...@h]1nc(=O)c3cc(Cl)c(N)cc3OC Imatinib: CN1CCN(Cc2ccc(cc2)C(=O)Nc3ccc(C)c(Nc4nccc(n4)c5cccnc5)c3)CC1 example from paper: http://pubs.acs.org/isubscribe/journals/jcisd8/38/i03/figures/ci970429if00003.html RDKit fragments (Leaves) of cisapride: CCCOc1ccc(F)cc1 C(=O)c1c(OC)cc(N)c(Cl)c1 CCC(N)C(OC)C CC(OC)C(N)CC Adrian