**3. Palmitoyl-Cysteine prediction**

Prior to the discovery of PATs, attempts were made to define stretches of amino acids that were preferred for palmitoylation. Palmitoylation near the N-terminus, following myristoy‐ lation, is among the predictable places for palmitoylation to occur provided there is one or more nearby cysteines. Navarro-Lérida et al (2002) fused a myristoylation motif (MGCTLS) to GFP with a short intervening sequence containing cysteines at various locations. These authors found a preference for cysteine palmitoylation at positions 3, 9, 15 and (to a much lesser degree) 21 residues away from the N-terminal methionine, but intervening residues were not evaluated. Commonalities in the composition of amino acid residues surrounding palmitoylated cysteines have been noted among members of the family of yeast amino acid permeases [74].

As more palmitoylated proteins and specific palmitoyl-cysteines are discovered, the task of predicting which adjacent amino acids provide a favorable environment for palmitoylation becomes easier. Algorithms trained with data from identified palmitoyl cysteines and adja‐ cent amino acid residues are now able to provide predictions of the statistical likelihood that a cysteine of interest may be palmitoylated [75-78]. CSS-Palm 2.0, which was designed to predict potential palmitoylation sites, has been published [75]. The algorithm was trained to recognize potential palmitoyl-cysteines using a dataset of 263 experimentally determined palmitoylation sites from 109 distinct proteins. Interestingly, CSS-Palm 2.0 also successfully predicted most (~75%) of the same novel palmitoyl-cysteines in yeast proteins previously identified by Roth. et al [74] as well as palmitoyl-cysteines predicted by Roth et al., to be pal‐ mitoylated but not experimentally determined. This rate of success in both cases suggests that CSS-PALM 2.0 is more conservative at calling a site, potentially resulting in a greater rate of false negative results but is reasonably accurate nonetheless. This algorithm should prove useful when prioritizing which cysteine(s), often among multiple potential cysteines of a candidate palmitoyl protein, to analyze experimentally.

Patterns of amino acid residues surrounding palmitoyl-cysteines have emerged from these analyses. A diagram of favored residues generated by an early version of CSS-Palm 2.0 (NBA-Palm) [76] shows that leucines and additional cysteines are more commonly observed around palmitoyl-cysteines. The subsequent versions of NBA-palm used significantly im‐ proved predictive tests, but the rough sequence of preferred residues remains. An important aspect that cannot yet be considered when attempting to predict cysteine palmitoylation with these algorithms is the complexity of the PAT-substrate recognition that is encoded by residues outside of those that immediately surround the palmitoyl-cysteine; the higher or‐ der components of the recognition sites.
