From Protein Structure to Function with Bioinformatics.pdf

116 P. Tompa5.2.3 Low Sequence Complexity and DisorderAnother manifestation of the repetitive nature of IDPs is low sequence complexityof their polypeptide chains. Application of an entropy function (Shannon 1948) toamino acid sequences of proteins (Wootton 1994a, b) has shown that globular proteinsappear mostly to be in a high-entropy (complexity) state, whereas in manyother proteins long regions apparently of low complexity can be observed. As muchas 25% of all amino acids in Swiss-Prot are in low-complexity regions, and 34% ofall proteins have at least one such segment (Wootton 1994a, b). The exact relationshipof low complexity and disorder has been addressed in two studies. First, therelationship of alphabet size (number of amino acids) and complexity to the capacityof folding was studied (Romero et al. 1999). It was found that SwissProt proteinscover the entire possible range of alphabet size (1–20) and entropy range (K = 0.0–4.5),whereas globular domains only occupied a limited region (alphabet = 10–20,K = 3.0–4.2). Regions corresponding to lower values (down to alphabet size = 3and K = 1.5) mostly correspond to structured, fibrous proteins, such as coiled-coils,collagens and fibroins. It was concluded that a minimal alphabet size of 10 andentropy near 2.9 are necessary and sufficient to define a sequence that can fold intoa globular structure. By extending these studies to IDPs (Romero et al. 2001), it wasshown that the complexity distribution of disordered proteins is shifted to lowervalues, but significantly overlaps with that of ordered proteins. Overall, disorderedand low-complexity regions correlate and are abundant in proteomes, but lowcomplexityand disorder should not be treated as synonyms.5.3 Prediction of DisorderBased on the noted compositional bias, about 25 predictors of disorder have beendeveloped (see Table 5.1 (Ferron et al. 2006; Dosztanyi et al. 2007) ). The best predictorsapproach the accuracy of the best secondary structure prediction algorithms,and the principles of comparing their performance have already been laid down.5.3.1 Prediction of Low-Complexity RegionsAs shown by the aforementioned studies, low sequence complexity differs from disorder,yet prediction of low complexity regions can be considered as a first reasonableapproach to assessing disorder, or at least the lack of globularity. The entropyfunction of Shannon (Shannon 1948), adapted to the case of protein sequences(Wootton 1994a, b) forms the basis of the SEG program routinely used to identifysequentially biased fragments of low compositional complexity measures. This practicehas a definite value in delineating non-globular regions of proteins.