A novel atom-pair hologram (APH) and its application in peptide QSARs

Thirty-six kinds of atomic fragments are divided in terms of common structural characteristics of organic compounds. Based upon that, according to occurring frequencies of atom-pair fragments at different bond distances, multi-order atom-pair frequency matrix is constructed and based upon that, atom-pair hologram (APH), a novel coding technique, is proposed. Applying APH into three panels of peptides (i.e. 48 bitter tasting dipeptides, 20 thromboplastinm inhibitors and 31 bradykinin potentiating pentapeptides), correlation coefficient r2 and cross-validation q2 of the resulting three PLS models are 0.980, 0.903; 0.993, 0.975 (0.993, 0.957) and 0.974, 0.869, respectively. In comparison, APH is potent to extracting structural information of pharmaceutical molecules and to prominently relating with biological activities. Moreover, APH has great merits in quick coding, easy operation, no requires for experimental parameters, etc. Besides, characterizing peptide analogues by APH help overcome demerits inherited in traditional sequence characteristic methods which are based on single amino acid descriptors.