In unison: regularization of protein secondary structure predictions that makes use of multiple sequence alignments.

We present a method whose purpose is to post-process thefuzzy results of secondary structure prediction methodsthat use multiple sequence alignments, in order to obtain‘realistic’ secondary structures, i.e., secondary structureelements whose length is greater than or equal to somepredefined minimum length. This regularization helps withinterpretation of the secondary structure prediction.Keywords: filtering/multiple sequence alignment/prediction/realistic secondary structureIntroductionThe prediction of a protein’s three-dimensional (3D) structurefrom its amino acid sequence is an important, albeit unsolved,problem of molecular biology. The development of newmethods, such as threading algorithms (see Lemer et al., 1995and references therein) might represent a crucial step towardsthis long sought after goal. Threading methods can be describedsuccinctly as 3D alignments. An amino acid sequence isaligned (threaded) onto the 3D core of a protein family (thecore is defined as conserved secondary structure elements,α-helices and β-strands, in the 3D structures of the proteinfamily) and one searches the particular alignment thatminimizes an empirical energy function. In practice perform-ing this 3D alignment is a difficult problem. Lathrop (1994)has shown that this is a NP-complete problem. A way ofmaking this problem more tractable is to include the secondarystructure prediction in the threading algorithm (J.F.Gibrat,unpublished results). This considerably reduces the alignmentspace that needs to be explored.In this context it is necessary to develop methods that,obviously, provide as accurate as possible secondary structurepredictions.Current protein secondary structure prediction methodspredict three residue states e: the two regular secondarystructures, α-helix (e5 H) and β-strand (e5 b), and the rest,random coils or loops (e5 c or ‘.’). The predictive power ofthe methods is expressed in terms of the ‘precision’ or ‘quality’index Q