Predicting Nucleolar Proteins Using Support-Vector Machines

Abstract

The intra-nuclear organisation of proteins is based on possibly transient interactions with morphologically defined compartments like the nucleolus. The fluidity of trafficking challenges the development of models that accurately identify compartment membership for novel proteins. A growing inventory of nucleolar proteins is here used to train a support-vector machine to recognise sequence features that allow the automatic assignment of compartment membership. We explore a range of sequence-kernels and find that while some success is achieved with a profile-based local alignment kernel, the problem is ill-suited to a standard compartment-classification approach.