Comparison of three related methods to select T cell-presented sequences of protein antigens

A comparison of three methods to predict T cell-presented sequences within antigenic proteins led to the view that recurrent hydrophobic residues might nucleate excised peptides as alpha-helices against hydrophobic surfaces. Such helices could be protease-protected structures on their way to desetope binding. The compared methods were: the amphipathicity algorithm of DeLisi and Berzofsky [Proc. natn. Acad. Sci. U.S.A. 82, 7048-7052. (1985)] as modified by Margalit et al. [J. Immun. 138, 2213-2229. (1987)] the strip of-helix hydrophobicity algorithm (SOHHA) of Stille et al. [Molec. Immun. 24, 1021-1027. (1987)] and the motifs algorithm of Rothbard and Taylor [EMBO J. 7, 93-100. (1988)]. Correct prediction was defined at two levels of stringency: (1) the predicted sequence overlapped the experimentally reported sequence when the ratio of the intersection of both to the union of both greater than or equal to 0.5 or (2) the sequences touched when there was a non-empty intersection of both sequences. We determined the sensitivity (correct predictions/number of reported T cell-presented sequences) and efficiency (correct predictions/number of predictions) at each level of stringency. In terms of overlap, the SOHHA was more sensitive (0.43) than the amphipathicity (0.29) (not significant) and motifs (0.0, 0.0) (p less than 0.05) predictions and more efficient (0.35) than the amphipathicity (0.14) and motifs (0.0, 0.0) predictions. At the less stringent criterion touching, the amphipathicity method (0.71) was as sensitive as motif Rothbard-4 (0.79) and more sensitive than SOHHA (0.57) and motif Rothbard-5 (0.43). At that criterion, the SOHHA was more efficient (0.47) than the amphipathicity (0.36) and motifs (0.25, 0.40) methods. We hypothesize that the comparability of these approaches reflected the common, predominant influence of recurrent hydrophobicity in their predictions.