Supplementary Note
For the article “Evolutionary conrvation of motif constituents within the yeast protein interaction network” by S. Wuchty, Z.N. Oltvai and A.-L. Barabási.
Table of content
成人大学A.Databas
B.Effects of data incompleteness and errors
B.1 Errors in protein-protein interactions (Supplementary Table 1)
B.2 Errors in ortholog protein assignment (Supplementary Table 2)
B.3 The effect of protein complexes (Supplementary Table 3)
俄罗斯奥运会
A. Databas
Protein Interaction Databas es: Large-scale two-hybrid screens, which allow the identification of potential protein-protein interactions between open reading frames predicted from the S. cerevisiae ge
nomic quence1,2 are an integral part of proteomics rearch. Yet, the quality of two-hybrid data is significantly affected by high rates of fal positives and fal negatives3, indicated also by the fact that the results obtained by different groups overlap only to a limited extent4. Moreover, many identified interactions rely on positive signals from a single technique and result from indirect obrvations.
A variety of databas provides efficient and easy access to the rapidly increasing knowledge about various protein interaction process. The MIPS databa5 collects genetic, biochemical and cell biological knowledge of various organisms which was extracted from the literature. BioKnowledge library is a composition of protein-specific information collected from the scientific literature6. Schwikowski et al. gathered all available protein interaction data of yeast, providing a comprehensive graph of protein-protein interactions7.
In our study, we ud the databa of interacting proteins (DIP) which is bad on extensive literature arches, and aims to provide the best curated collection of all functional linkages of proteins obtained by experimental methods. The majority of protein-protein interaction data relies on yeast two-hybrid- and co-immunoprecipitation experiments. 84% of the interactions are detected by only one single experiment, whereas 16% are confirmed by more than one experimental method. DI
P records nearly 3000 proteins which are involved in approximately 9000 interactions8. Since it is manually curated, DIP provides high quality interaction data by minimizing the total number of fal positive and negative interactions.
Orthologous Sequence Information:M ethods of finding orthologous pairs of quences often utilize pairwi BLAST comparisons of whole proteomes. Each protein reprents
queries against the entire proteome of the other species. Symmetrical best hits in the BLAST arches, emphasizing expectation values smaller than 10-3, were considered to be orthologous. The databa of clusters of orthologous genes (COG) was compiled9 utilizing such orthologous quence pairs of mainly prokaryotic organisms.
Our choice of orthologous protein quence information is the InParanoid databa10 which, similarly to COG, runs an all-versus-all BLAST arch with two ts of quences. Sequence pairs with mutual best hits are detected and rve as central main ortholog pairs around which further orthologs from both species are clustered in later steps. The initial assumption is that quences from the same species that are more similar to the main ortholog than to any quence from other species are ‘in-paralogs’, belonging to the same group of orthologs. In contrast to COG, the quality o
f the resulting orthologous clusters is examined and incread by a final bootstrap analysis10. Furthermore, InParanoid provides comprehensive pairwi comparative orthologous information between S. cerevisiae and H. sapiens, D. melanogaster, C. elegans, M. musculus and A. thaliana, which are abnt in COG.
B.Effects of data incompleteness and errors
B.1 Errors in protein-protein interactions (Supplementary Table 1)
It is well known that protein-protein interaction and ortholog databas are incomplete and contain fal positives3. In order to address the potential effects of this technology-derived noi we investigated the effects of data incompleteness and incorrectness on our results. We mimick fal positives by the addition of extra 10% or 20% of interactions between randomly picked protein pairs that were previously abnt in the yeast protein interaction network. In turn, the random removal of 10% and 20% of known interactions accounts for data incompleteness (fal negatives). Subquently, we recalculated the conrvation rate of the motifs for the artificially altered databa (Supplementary Table 1). The measurements indicate that the conrvation rate of the motifs is only marginally affected, and the trend towards the incread degree of conrvation of larger motifs rem
荷兰足球ains unchanged. In particular, while data incompleteness (fal negatives) has negligible influence, since fal positives randomize the network, the motif conrvation is slightly, -but not significantly-, affected.
隔离室图片B.2 Errors in ortholog protein assignment (Supplementary Table 2)
Similar results are obtained for the introduction of fal positive and negative orthologs. For this we removed 10% or 20% of the assigned ortholog proteins in order to simulate the effects of fal negative ortholog data. Accounting for fal positives, we randomly added 10% or 20% more proteins to the original t of orthologous proteins. The results of the subquent recalculation of the motif’s conrvation rate are compiled in Supplementary Table 2, indicating that the addition of fal positive orthologs leaves the conrvation rates almost unaltered. In contrast, fal negatives lead to decread conrvation rates, since the removal of known orthologs significantly lowers the statistics. However, the basic trend obrved in
每日晨读the paper remains valid: Despite the systematic lowering in the conrvation rate for fal negatives, the highly connected motifs are more likely conrved than their less cohesive counterparts. Although flawed interaction and orthologs data obviously affect the exact quantities, we conclude tha
t qualitative features of the networks, such as trends towards evolutionary conrvation of the motifs remain robust.
古代寓言
B3: The effect of protein complexes (Supplementary Table 3):
Protein complexes, in which each protein interacts with almost all other complex partners, might bias our investigation and could be viewed as the sole origin of the obrved conrvation of the highly connected motifs. In the following, we show that the prence of protein complexes cannot explain exclusively the obrved conrvation effects. To eliminate the effects of protein complexes we removed all proteins that are part of known complexes bad on the Swissprot databa11 (31 proteins and 564 interactions) and recalculated the conrvation rates of motifs remaining in the diluted network. Although we obrve, as expected, a common decrea of conrvation rates, the basic trend towards higher conrvation rates of strongly connected motifs remains unaltered. In comparison to the random reference ts, natural conrvation ratios of the real motifs are still veral orders of magnitude higher.
征服妈妈>逛花街protein interaction data. As in Supplementary Table 1, the third column d enotes the number of motifs found in the interaction network of 3,128 yeast proteins, obtained by counting subgraphs of two to five nodes. Our t of orthologs embraces again all 678 proteins that have an orthologue in each proteome of human, mou, fly, worm and A. thaliana. The natural conrvation rate shows what fraction of the original yeast motifs are evolutionary fully conrved. In ord er to investigate the influence of partially flawed interaction data on the conrvation rate of motifs, 10% of the underlying interaction network’s edges were randomly chon and deleted mimicking fal negative interaction signals. Conrvation rates of motifs were subquently calculated (col. 5). The same procedure was repeated after randomly deleting 20% of the edges of the initial protein interaction network (col. 4). Accounting for fal positive signals, 10% (col. 6) and 20% (col. 7) more new ed ges, previously abnt from the interaction network were rand omly ad d ed and conrvation rates of motifs thus obtained. Although the numbers of newly added or removed edges are quite high, the overall trend that highly intraconnected motifs persist is not affected. Although it is commonly assumed that yeast protein interaction data are highly incomplete and flawed due to the shortcomings of the widely ud two-hybrid experimental t up, conrvation rates, in particular for the triangles #3, squares #9 and pentagons #11, appear to be quite robust.
