Limits...
A probabilistic generative model for GO enrichment analysis.

Lu Y, Rosenfeld R, Simon I, Nau GJ, Bar-Joseph Z - Nucleic Acids Res. (2008)

Bottom Line: This makes it hard to determine if the identified significant categories represent different functional outcomes or rather a redundant view of the same biological processes.Our model accommodates noise and errors in the selected gene set and GO.When used with microarray expression data and ChIP-chip data from yeast and human our method was able to correctly identify both general and specific enriched categories which were overlooked by other methods.

View Article: PubMed Central - PubMed

Affiliation: Computer Science Department, Carnegie Mellon University, 5000 Forbes Ave, Pittsburgh, PA, 15213 USA.

ABSTRACT
The Gene Ontology (GO) is extensively used to analyze all types of high-throughput experiments. However, researchers still face several challenges when using GO and other functional annotation databases. One problem is the large number of multiple hypotheses that are being tested for each study. In addition, categories often overlap with both direct parents/descendents and other distant categories in the hierarchical structure. This makes it hard to determine if the identified significant categories represent different functional outcomes or rather a redundant view of the same biological processes. To overcome these problems we developed a generative probabilistic model which identifies a (small) subset of categories that, together, explain the selected gene set. Our model accommodates noise and errors in the selected gene set and GO. Using controlled GO data our method correctly recovered most of the selected categories, leading to dramatic improvements over current methods for GO analysis. When used with microarray expression data and ChIP-chip data from yeast and human our method was able to correctly identify both general and specific enriched categories which were overlooked by other methods.

Show MeSH
Comparison using GO for yeast. Performance comparison of GenGO (blue curve) with four other methods on data generated using the yeast GO database. We use p to represent the fraction of genes that are identified from an active GO category (true positive rate for a category, see Materials and methods section) and q to represent the fraction genes that are selected but do not belong to any active category. (a) Selecting one category with p = 0.9, q = 0.01; (b) Selecting one category with p = 0.5, q = 0.15; (c) and (d) Same as (a) and (b) but using two categories; (e) and (f) same with five categories. Note that even when the noise is substantial (using 50% of genes in selected categories and 15% of all other genes, bottom row) GenGO is still able to accurately recover most of the correct categories. See Supplementary Figure 1 for a more detailed figure.
© Copyright Policy - creative-commons
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC2553574&req=5

Figure 2: Comparison using GO for yeast. Performance comparison of GenGO (blue curve) with four other methods on data generated using the yeast GO database. We use p to represent the fraction of genes that are identified from an active GO category (true positive rate for a category, see Materials and methods section) and q to represent the fraction genes that are selected but do not belong to any active category. (a) Selecting one category with p = 0.9, q = 0.01; (b) Selecting one category with p = 0.5, q = 0.15; (c) and (d) Same as (a) and (b) but using two categories; (e) and (f) same with five categories. Note that even when the noise is substantial (using 50% of genes in selected categories and 15% of all other genes, bottom row) GenGO is still able to accurately recover most of the correct categories. See Supplementary Figure 1 for a more detailed figure.

Mentions: We used precision/recall curves to compare GenGO with four other methods (see Materials and methods section). These included ‘Classic’ (hypergeometric test) and the three other methods listed above. The results are plotted in Figure 2 (yeast) and Figure 3 (human). For all settings, the performance of GenGO dominates all other methods. When the noise level is low, the performance of GenGO is close to optimal (top rows in Figures 2 and 3). When the noise level is high, the performance drops for all methods, though GenGO is still the best. Even with high noise and multiple categories (as is the case for most real experiments) GenGO can achieve 80% precision for high recall levels (60–80%). As for the other methods, in most cases ‘Weight’ is the second best and ‘Classic’ is usually the worst, indicating that all methods previously proposed for the task indeed improve upon the standard usage of GO.Figure 2.


A probabilistic generative model for GO enrichment analysis.

Lu Y, Rosenfeld R, Simon I, Nau GJ, Bar-Joseph Z - Nucleic Acids Res. (2008)

Comparison using GO for yeast. Performance comparison of GenGO (blue curve) with four other methods on data generated using the yeast GO database. We use p to represent the fraction of genes that are identified from an active GO category (true positive rate for a category, see Materials and methods section) and q to represent the fraction genes that are selected but do not belong to any active category. (a) Selecting one category with p = 0.9, q = 0.01; (b) Selecting one category with p = 0.5, q = 0.15; (c) and (d) Same as (a) and (b) but using two categories; (e) and (f) same with five categories. Note that even when the noise is substantial (using 50% of genes in selected categories and 15% of all other genes, bottom row) GenGO is still able to accurately recover most of the correct categories. See Supplementary Figure 1 for a more detailed figure.
© Copyright Policy - creative-commons
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC2553574&req=5

Figure 2: Comparison using GO for yeast. Performance comparison of GenGO (blue curve) with four other methods on data generated using the yeast GO database. We use p to represent the fraction of genes that are identified from an active GO category (true positive rate for a category, see Materials and methods section) and q to represent the fraction genes that are selected but do not belong to any active category. (a) Selecting one category with p = 0.9, q = 0.01; (b) Selecting one category with p = 0.5, q = 0.15; (c) and (d) Same as (a) and (b) but using two categories; (e) and (f) same with five categories. Note that even when the noise is substantial (using 50% of genes in selected categories and 15% of all other genes, bottom row) GenGO is still able to accurately recover most of the correct categories. See Supplementary Figure 1 for a more detailed figure.
Mentions: We used precision/recall curves to compare GenGO with four other methods (see Materials and methods section). These included ‘Classic’ (hypergeometric test) and the three other methods listed above. The results are plotted in Figure 2 (yeast) and Figure 3 (human). For all settings, the performance of GenGO dominates all other methods. When the noise level is low, the performance of GenGO is close to optimal (top rows in Figures 2 and 3). When the noise level is high, the performance drops for all methods, though GenGO is still the best. Even with high noise and multiple categories (as is the case for most real experiments) GenGO can achieve 80% precision for high recall levels (60–80%). As for the other methods, in most cases ‘Weight’ is the second best and ‘Classic’ is usually the worst, indicating that all methods previously proposed for the task indeed improve upon the standard usage of GO.Figure 2.

Bottom Line: This makes it hard to determine if the identified significant categories represent different functional outcomes or rather a redundant view of the same biological processes.Our model accommodates noise and errors in the selected gene set and GO.When used with microarray expression data and ChIP-chip data from yeast and human our method was able to correctly identify both general and specific enriched categories which were overlooked by other methods.

View Article: PubMed Central - PubMed

Affiliation: Computer Science Department, Carnegie Mellon University, 5000 Forbes Ave, Pittsburgh, PA, 15213 USA.

ABSTRACT
The Gene Ontology (GO) is extensively used to analyze all types of high-throughput experiments. However, researchers still face several challenges when using GO and other functional annotation databases. One problem is the large number of multiple hypotheses that are being tested for each study. In addition, categories often overlap with both direct parents/descendents and other distant categories in the hierarchical structure. This makes it hard to determine if the identified significant categories represent different functional outcomes or rather a redundant view of the same biological processes. To overcome these problems we developed a generative probabilistic model which identifies a (small) subset of categories that, together, explain the selected gene set. Our model accommodates noise and errors in the selected gene set and GO. Using controlled GO data our method correctly recovered most of the selected categories, leading to dramatic improvements over current methods for GO analysis. When used with microarray expression data and ChIP-chip data from yeast and human our method was able to correctly identify both general and specific enriched categories which were overlooked by other methods.

Show MeSH