Prediction of sequence signals for lipid post-translational modifications: insights from case studies

In silico annotation techniques for post-translational modifications (PTMs) are important to generate biologically meaningful descriptions for sequences of experimentally uncharacterized proteins. Having previously contributed with predictors for lipid PTMs, we summarize our methodological experience. Rather than only looking for the sequence pattern in substrate sequences, a strategy aimed at creating a generalized model of substrate protein/enzyme interaction appears more appropriate since the number of known substrate sequences is small, and some of them are not sufficiently verified experimentally. Such a physical approach (in contrast to a mere textual analysis of substrate sequences) can also take into account other, heterogeneous biological data (mutations of substrate sequences, kinetic data, enzyme sequences/structures) with simple analytical expressions in the score function. Several lipid PTMs are encoded in the form of a small sequence region (with pronounced amino acid type preferences) that is connected to the substrate protein by a linker region with many conformationally flexible, hydrophilic residues. A score function composed of terms penalizing sequence properties known to be incompatible with productive substrate protein/enzyme complexes essentially unselects inappropriate queries. Also, we estimate the number of nonredundant sequences necessary for robust profile computation with statistical criteria, a number that is not reached in most cases of PTM prediction. Finally, we discuss the usage of evolutionary information in evaluating the functional importance of predicted PTMs in cases of motif conservation within sequence families.