Network classifiers and it achieves 87 cross-validation accuracy on balanced data with equal number of ordered and disordered residues. We used the VL3E predictor to predict Swiss-Prot proteins with long disordered regions. Each with the 196,326 Swiss-Prot proteins was labeled as putatively disordered if it contained a predicted intrinsically disordered area with 40 consecutive amino acids and as putatively ordered otherwise. For notational convenience, we introduce ABL1 Proteins web disorder operator d such that d(si) = 1 if sequence si is putatively disordered, and d(si) = 0 if it really is putatively ordered. Connection involving long disorder prediction and protein length The likelihood of labeling a protein as putatively disordered increases with its length. To account for this length dependency, we estimated the probability, PL, that VL3E predicts a disordered region longer than 40 consecutive amino acids in a SwissProt protein sequence of length L. Probability PL was determined by partitioning all SwissProt proteins into groups determined by their length. To minimize the effects of sequence redundancy, every single sequence was weighted because the inverse of its loved ones size; if sequence si was assigned to TribeMCL cluster c (si), we calculated ni as the total variety of SwissProt sequences assigned to this cluster and set its weight to w(si) = 1/ni. In this manner, each cluster is provided exactly the same influence in estimation of PL, no matter its size. To estimate PL, all SwissProt sequences with length between L-l and L+l were grouped in set SL = si, L-l siL+l. The probability PL was estimated asNIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author ManuscriptWindow size l permitted us to manage the smoothness of PL function. Within this study we applied window size equal to 20 in the sequence length, l = 0.1 . We show the resulting curve in Figure 1 collectively using the very same results when l = 0. Extracting disorder-and order-related Swiss-Prot key phrases For each of the 710 SwissProt keyword phrases occurring in much more than 20 SwissProt proteins, we set to figure out if it can be enriched in putatively disordered or ordered proteins. To get a keyword KWj, j = 1…710, we very first grouped all SwissProt proteins annotated with all the keyword to Sj. ToJ Proteome Res. Author manuscript; offered in PMC 2008 September 19.Xie et al.Pagetake into consideration sequence redundancy, every single sequence si Sj was weighted based on the SwissProt TribeMCL clusters. If sequence si was assigned to cluster c(si), we calculated nij as the total quantity of sequences from Sj that belonged to that cluster and set its weight to wj(i) = 1/nij. Then, the Checkpoint Kinase 2 (Chk2) Proteins Storage & Stability fraction of putatively disordered proteins from Sj was calculated asNIH-PA Author Manuscript NIH-PA Author Manuscript Final results NIH-PA Author ManuscriptThe query is how nicely this fraction fits the null model that may be according to the length distribution PL. Let us define random variable Yj aswhere XL is often a Bernoulli random variable with P(XL = 1) = 1 – P(XL = 0) = PL. In other words, Yj represents a distribution of fraction of putative disorder among randomly selected SwissProt sequences together with the same length distribution as these annotated with KWj. If Fj is within the left tail in the Yj distribution (i.e. the p-value P(Yj Fj) is near 1), the keyword is enriched in ordered sequences, although if it really is in the right tail (i.e. the p-value P(Yj Fj) is near 0) it can be enriched in disordered sequences. We denote all key phrases with p-value 0.05 as disorder-related and these with p-value 0.95.