Information distance explains MHC II supertypes

The classification of MHC II molecules into supertypes is an active field of research with significance in the development of epitope-based vaccines. Existing methods based on binding data are restricted by the scarcity of such data. For this reason, methods based on clustering partial or pseudo-sequences have become popular. However, these methods are inept in determining the number of supertypes, and they also fail to include a few MHC II serotypes. In this work, we incorporate co-evolution information between residues to calculate the distance between sequences, and adopt the Bayesian information criterion with a dynamic programming method to determine the number of supertypes. The resultant algorithm is able to include more serotypes, achieving a 90.6% of accuracy according to the serotypes assigned by World Health Organisation.