skip to primary navigationskip to content


Structure prediction and machine learning for variable pathogens

The hemagglutinin (HA) glycoprotein of influenza is known to be a major factor in determining virus species specificity (avian, swine or human) and pathogenicity.  Mutations in HA arise frequently during genome replication, allowing the virus to escape the host's adaptive immune response and to infect the same individual multiple times.  The number of amino acids in the HA binding region and diversity of mutations observed make it difficult to test all the possible changes using current experimental screening methods. A method that could predict the receptor binding behaviour of potential future strains would be of tremendous benefit to virology and human health. This is highlighted by the current global pandemic situation. Extensive studies of HA have already been conducted using molecular dynamics simulations, and we propose using basin-hopping global optimisation to identify novel binding conformations and minimise the time required to model and quantitatively assess a particular protein-ligand combination.  Investigations have been conducted into the interactions of various human, swine and avian HAs with (2,3) and (2,6) linked 3-ring SLeX ligands. The initial phase of the proposed research would be to expand the set of ligands modelled to reflect glycan diversity in different species and different sections of the human respiratory tract. The work will involve collaboration with experimental groups involved in influenza research and influenza surveillance efforts worldwide. Whilst the immediate focus of the research will be upon predicting developments in the pandemic H1N1 strain, due to the change in the prevalence of different influenza strains, it is difficult to predict which strains may be of interest in the longer term.