Limits...
Identification of protein complexes from co-immunoprecipitation data.

Geva G, Sharan R - Bioinformatics (2010)

Bottom Line: The framework aims at identifying sets of preys that significantly co-associate with the same set of baits.In application to an array of datasets from yeast, our method identifies thousands of protein complexes.Comparing these complexes to manually curated ones, we show that our method attains very high specificity and sensitivity levels (∼ 80%), outperforming current approaches for protein complex inference.

View Article: PubMed Central - PubMed

Affiliation: School of Computer Science, Tel Aviv University, Tel Aviv, Israel.

ABSTRACT

Motivation: Advanced technologies are producing large-scale protein-protein interaction data at an ever increasing pace. A fundamental challenge in analyzing these data is the inference of protein machineries. Previous methods for detecting protein complexes have been mainly based on analyzing binary protein-protein interaction data, ignoring the more involved co-complex relations obtained from co-immunoprecipitation experiments.

Results: Here, we devise a novel framework for protein complex detection from co-immunoprecipitation data. The framework aims at identifying sets of preys that significantly co-associate with the same set of baits. In application to an array of datasets from yeast, our method identifies thousands of protein complexes. Comparing these complexes to manually curated ones, we show that our method attains very high specificity and sensitivity levels (∼ 80%), outperforming current approaches for protein complex inference.

Availability: Supplementary information and the program are available at http://www.cs.tau.ac.il/~roded/CODEC/main.html.

Show MeSH
A comparison of protein complex identification approaches on the data of Gavin (2002). See legend of Figure 2 for details.
© Copyright Policy - creative-commons
Related In: Results  -  Collection

License 1 - License 2
getmorefigures.php?uid=PMC3008648&req=5

Figure 4: A comparison of protein complex identification approaches on the data of Gavin (2002). See legend of Figure 2 for details.

Mentions: Our final comparison was to the Local Modeling method (Scholtens et al., 2005). The available implementation of the method could not run on the datasets of Gavin et al. (2006) and Krogan et al. (2006) due to their relatively large size. Hence, we used a smaller data set as a test case (Gavin, 2002), containing 455 bait proteins and 1364 prey proteins. The protein complexes inferred by Local Modeling are partitioned into three categories: complexes that are supported by multiple baits (marked as MBME), complexes that are supported by a single bait (marked as SMBH) and complexes that contain two baits where only one of the baits identifies the other bait as its prey. We focused on the 272 MBME complexes, which represent the highest confidence predictions. As can be seen in Table 3 and Figure 4, CODEC outperforms local modeling, attaining higher specificity and sensitivity. When including in the Local Modeling solution also the SMBH complexes (336 in total) the sensitivity increased to 93%, at the price of a decrease in specificity (to 69%). Overall, these results are comparable to those of CODEC, although providing a slightly worse F-measure (79% compared with CODEC's 82.5%).Fig. 4.


Identification of protein complexes from co-immunoprecipitation data.

Geva G, Sharan R - Bioinformatics (2010)

A comparison of protein complex identification approaches on the data of Gavin (2002). See legend of Figure 2 for details.
© Copyright Policy - creative-commons
Related In: Results  -  Collection

License 1 - License 2
Show All Figures
getmorefigures.php?uid=PMC3008648&req=5

Figure 4: A comparison of protein complex identification approaches on the data of Gavin (2002). See legend of Figure 2 for details.
Mentions: Our final comparison was to the Local Modeling method (Scholtens et al., 2005). The available implementation of the method could not run on the datasets of Gavin et al. (2006) and Krogan et al. (2006) due to their relatively large size. Hence, we used a smaller data set as a test case (Gavin, 2002), containing 455 bait proteins and 1364 prey proteins. The protein complexes inferred by Local Modeling are partitioned into three categories: complexes that are supported by multiple baits (marked as MBME), complexes that are supported by a single bait (marked as SMBH) and complexes that contain two baits where only one of the baits identifies the other bait as its prey. We focused on the 272 MBME complexes, which represent the highest confidence predictions. As can be seen in Table 3 and Figure 4, CODEC outperforms local modeling, attaining higher specificity and sensitivity. When including in the Local Modeling solution also the SMBH complexes (336 in total) the sensitivity increased to 93%, at the price of a decrease in specificity (to 69%). Overall, these results are comparable to those of CODEC, although providing a slightly worse F-measure (79% compared with CODEC's 82.5%).Fig. 4.

Bottom Line: The framework aims at identifying sets of preys that significantly co-associate with the same set of baits.In application to an array of datasets from yeast, our method identifies thousands of protein complexes.Comparing these complexes to manually curated ones, we show that our method attains very high specificity and sensitivity levels (∼ 80%), outperforming current approaches for protein complex inference.

View Article: PubMed Central - PubMed

Affiliation: School of Computer Science, Tel Aviv University, Tel Aviv, Israel.

ABSTRACT

Motivation: Advanced technologies are producing large-scale protein-protein interaction data at an ever increasing pace. A fundamental challenge in analyzing these data is the inference of protein machineries. Previous methods for detecting protein complexes have been mainly based on analyzing binary protein-protein interaction data, ignoring the more involved co-complex relations obtained from co-immunoprecipitation experiments.

Results: Here, we devise a novel framework for protein complex detection from co-immunoprecipitation data. The framework aims at identifying sets of preys that significantly co-associate with the same set of baits. In application to an array of datasets from yeast, our method identifies thousands of protein complexes. Comparing these complexes to manually curated ones, we show that our method attains very high specificity and sensitivity levels (∼ 80%), outperforming current approaches for protein complex inference.

Availability: Supplementary information and the program are available at http://www.cs.tau.ac.il/~roded/CODEC/main.html.

Show MeSH