Fig. 2.

Overview of our local structure prediction method. Starting with the target amino acid sequence as input, a single track HMM is constructed and updated by the SAM program during an iterative search for homologous sequences. After convergence, the SAM HMM is used to generate a multiple sequence alignment of the homologous sequences that were found. This alignment is input to the predict-2nd multilayer neural network which generates predictions of the letters of a local structure alphabet for each position in the input alignment.