Multicoil-HMM : improved prediction of coiled-coil oligomer state from sequence

Abstract

The Multicoil-HMM algorithm offers improved prediction of coiled-coil oligomerization state. The algorithm combines the pairwise correlations of the Multicoil method with the flexibility of HMM methods. The resulting method incorporates predictors deemed important by a multinomial logistic regression to distinguish between the dimer, trimer and non-coiled coil oligomerization states. The Multicoil-HMM algorithm shows significantly improved oligomer state prediction over a retrained Multicoil algorithm, which is currently the state-of-the-art. The general strategy of using multinomial regression on predictors that can be simulated by HMMs while abandoning the probabilistic interpretation of HMMs may be useful in other machine learning applications.