We are using the Computer Graphics Laboratory (CGL) to create an unsupervised BCM-based neural network for finding principal patterns relating amino acid sequence to phi-psi bond angles. These patterns, when constrained by a limited number of nodes in the model, represent """"""""secondary structure"""""""" archetypes ranked by prepredictability rather than based on bond-angle sequences. A second project is to write software to identify, based both on sequence and three-dimensional structure, the general superfamily to which a particular protein belongs. This technique is based on statistically analyzing distantly related members of superfamilies using a novel algorithm invented here. A third project involves analyzing the general structure of neurotrophins, the small protein growth factors that are critical to neural development. The structure of one neurotrophin was determined by our lab and we hope to relate its activity and binding specificity to structural similarities and differences with the other neurotrophins.
Showing the most recent 10 out of 508 publications