Introduction

Immunoglobulin (Ig) domains are found in hundreds of proteins throughout the human genome with various functions. These domains are often crucial components of protein-protein or protein-ligand interactions. Characterizing interacting domains is an important pharmacological task since many drug candidates target these regions to prevent binding.

One the best studied immunoglobulin domains is that in antibodies. Antibodies are small, 150kDa proteins in the immune system that are used to target and neutralize pathogens. They consist of two chains, the light and the heavy chain. The light and heavy chain have two regions, the constant and variable region. This variable region binds antigens and undergoes recombination to increase the repertoire of binding events. This process creates exquisite specificity for ligands and dissociation constants in the picomolar range for antibody-epitope interactions.

Such tight binding events are highly desired in the pharmacological world and the ability to predict such events would reduce cost, time, and wasted resources for antibody development. Saving companies and researchers valuable funds and time. The human genome probably contains hundreds of Ig domains that are similar to the antibody variable region. Furthermore, these binding potentials have yet to be explored. Hidden Markov models (HMM) are often used to predict unobserved states in a system. By using observable outcomes to predict the probability that an unobserved variable is in one of two states (for our purposes we used two states, but there are higher level models). The model is first trained on a seed or test data that contains the desired state. This "trained model" can then be used to predict novel states such as immunoglobulin domains.

Another pharmaceutical application is manipulation of Ig domains to create modified binding to targets. Phylogenetic trees and clustering software can be used as predictive models to classify binding domains' affinities.