Improvements to profile PSTMM for glycan recognition profile prediction

Glycans are biomolecules that are composed of various monosaccharides, and they are bound to proteins and lipids on the cell surface. Glycans are known as key players of biological phoenomena such as in the determination of blood type, cell adhesion, antigen-antibody reactions, virus infections, etc. It is known that more than half of the proteins in major protein structure database such as SWISS-PROT are glycosylated. Unlike proteins, however, glycans are tree structures which vary across organs, tissues, and even cells. Lectins are glycan-binding proteins that recognize monosaccharides at the non-reducing end, and it is believed that they may also recognize sugars further along the glycan chain. To capture these potential recognition structures, PSTMM (Probabilistic Sibling-dependent Tree Markov Model), which considers relationships between “sibling” monosaccharides within glycans, was developed. Furthermore, a profile version of PSTMM, called profile PSTMM, was developed [1], which has been implemented as a web tool last year. Profile PSTMM requires the definition of a “state model” which defines the structure of the profile to be learned from the training data. The state model of the profile PSTMM web tool is currently generated based on the maximum common subtree (MCST) of all input glycan structure. However, this results in a very small state model. In order to compensate for this, we are developing a method to multiply align all glycan structures and extract glycan substructure blocks based on a method similar to ClustalW. We introduce this new method of multiple tree alignment here.