DanteR: DanteR

In DanteR: Proteomics data analysis tool

Description

DAnTE is a front-end tool for downstream proteomic data analysis.

Usage

dante()

Arguments

No arguments

Details

dante extends the capabilities of the data pipeline by allowing the user to apply many data analysis algorithms.

Value

NULL

Note

Functions Within DanteR

File Menu

DAnTE has an advanced data loading mechanism. It can load either abundance data (or expression ratios) or Sequest output files to process Peptide count data.

Data tables are treated as crosstabs where columns generally correspond to data sets and rows correspond to an observation class - for example, peptide intensities, or gene observation.

Data tables may be linked to tables containing metadata (data about data) for their row and column names.

Abundance Data

DAnTE can preserve the peptide-protein mapping information if the user decides to load protein information. The data file can be in the following formats: Excel, Excel 2007, Access, CSV, TSV and SQLite DB.

The data file must have

* A header row * A unique row ID column (ex: Mass Tag or Probeset ID) * Abundance values for each row in the next columns with each column corresponding to a dataset * Optional Columns containing row metadata IDs to create a separate row metadata table.

Peptide Count Data File -> Open -> Spectral Count Data will bring an interface to load sequest output files. There are three methods to load the SEQUEST output files: 1. Concatenated *.out files (*_out.txt) or Synopsis files (*_syn.txt) from a local folder. These files are generated using Peptide File Extractor (available at http://omics.pnl.gov/software/.

2. SEQUEST *.out files residing in a local folder. Dataset names are obtained using the file names (dataset names are extracted from the file names truncated at the first "." Ex: File name: dataset1_0305-07.5.5.1.out -> Dataset name: dataset1_0305-07).

3. Directly from the PNNL DMS archive, either *_out.txt or *_syn.txt files (on PNNL computers only). You need to have at least two unique SEQUEST results sets to build a table of spectral counts. Loading from *_syn.txt files will be the fastest.

Metadata Metadata, or data about data, is important for keeping track of data organization and analysis. Much like Access, DanteR defines links between the crosstab of interest and tables containing row metadata and column metadata.

Column Metadata (Factors) Factors capture the underlying grouping structure in the data. For example, gender can be treated as a factor with two enumeration levels Male/Female. Treatment can be treated as a factor with as many enumeration levels (Burn/Sham). Factors can either be loaded as a csv file or defined in the 'Metadata' menu.

Pre-Process Menu

Log Transform User can log transform the data (either base 2 or 10) with options of having a additive or multiplicative bias.

Linear Regression Based Normalization

Linear regression method tries to fit a regression line for each dataset within a selected factor (ex. Replicate) against a reference (see 'Define Factors' for more information on factors).

Loess Normalization Loess normalization (local regression estimates) tries to fit simple models to localized subsets of the data. Everything else follows as in the linear regression method. The size of the local subset is controlled by the span value (Window width) and Callister et.al.(2006) report that a value of 0.4 works best for LC-MS proteomics data.

Quantile Normalization This method implements the Quantile Normalization method proposed by Bolstad BM, Irizarry RA, Astrand M, Speed TP. (See Bioinformatics. 2003 Jan 22;19(2):185-93)

Median Absolute Deviation (MAD) Adjustment This method tries to adjust the MAD of each dataset so that all the data sets will have the same spread of abundance intensity (i.e. comparable box plots). This is achieved by choosing a suitable scaling factor for each dataset to adjust its abundance values. See Yang, Y. H., S. Dudoit, et al. (2002). Nucleic Acids Res 30(4): e15.

Central Tendancy Adjustment Mean/Median of the dataset (in a column) is subtracted from (or divided by) each, resulting datasets which will have mean/median centered at zero. User can choose to subtract/divide either mean or the median and also to set the new mean/median at zero. If the checkbox 'New Center at Zero' is unchecked, all means/medians are set to the maximum in all the datasets.

Imputing Missing Values There are some simple ways of imputing missing values and also some advanced methods. The screenshot above shows some of the simplest ways one can impute missing data.

Mean/Median Mean or the median of the entire dataset is used to replace all missing values in the dataset.

Substitue a Constant Fill all the missing data with the same constant value such as half of the maximum detected value.

Row Mean within a Factor This method fills the missing values within a row with the mean value of row observations by the selected factor. This approach is recommended when significant difference are expected among factor levels.

k-Nearest Neighbor (kNN) Method Replaces missing data with the mean of the k-nearest neighbor peptides (See: Troyanskaya O. and Cantor M. and Sherlock G. and Brown P. and Hastie T. and Tibshirani R. and Botstein D. and Altman RB. - Missing value estimation methods for DNA microarrays. Bioinformatics. 2001 Jun;17(6):520-5.)

Weighted kNN Method Replaces missing data with a weighted mean of the k-nearest neighbor peptides. The weights are inversely proportional to the distances from the neighboring peptides.

SVDimpute method SVDimpute algorithm works iteratively until the change in the estimated solution falls below a certain threshold. Each step the eigenpeptides of the current estimate are calculated and used to determine a new estimate. An optimal linear combination is found by regressing the incomplete variable against the k most significant eigenpeptides. See: Troyanskaya O. and Cantor M. and Sherlock G. and Brown P. and Hastie T. and Tibshirani R. and Botstein D. and Altman RB. - Missing value estimation methods for DNA microarrays. Bioinformatics. 2001 Jun;17(6):520-5.

Rollup Menu

RRollup - Reference Peptide Based Scaling Note that this method correctly works only with log transformed data.

A reference peptide which has the most presence across all the datasets, is chosen from the group of peptides that belong to a protein. If there are multiple candidates, the most abundant one is chosen. Then the ratios of peptide abundances with respect to the reference are computed (since the data is assumed to be in log scale, the differences are used) and their median is used as a scaling factor. Protein abundance is obtained as the median of the resulting peptide abundances.

Minimum Presence of at least one Peptide for a Protein: Peptides that have too many missing values below this percentage are dropped.

Exclude peptides from scaling if they are at least not present in this many datasets: Within a group of peptides for a specific protein, the ones that do not overlap well (controlled by this value) are not scaled but they are kept to calculate the final protein abundance.

Include Single peptide/protein matches (i.e. 'One-Hit-Wonders'): Protein with only one observed peptide will be included in the final list of proteins. The rational behind this is that if a particular protein may have only one peptide but it may be quite abundant and present throughout giving some strong confidence on the presence of the protein.

Plot Menu Histograms Blue curve shows the density distribution if shifted to mean zero and the red curve shows the density distribution in the data.

QQ Plot A normal Qquantile-Quantile Plot is a graphical method for diagnosing how well the data fits to a comparison distribution (a normal distribution is used here). In the case of substantial deviations from linearity, one can assume that the normality assumption is violated.

Correlation Plots Correlations between the column data can be plotted in many different ways. For heatmap style and 2D box style, the correlation range can be adjusted for display purposes.

Correlation Plots

Box Plots

Principal Component Analysis (PCA) and Partial Least Squares Analysis (PLS)

You can select a factor for the PCA/PLS plot to be colored accordingly. There are options to plot either scores plot or loadings plot (see biplot option). User can also choose to plot the screeplot.

The resulting weights of the PCA and PLS analysis on each of the orthogonal component are also reported in a Data Table. The weights reported correspond to a p-value obtained by fitting an empirical distribution function to the original weights. A low p-value denotes a significant feature.

MA Plot MA plot is a scatterplot with transformed axes. The X-axis represents the average log intensity from 2 datasets while Y-axis represents the log-ratios. MA plot is especially useful for the detection of the intensity-dependent effects in datasets.

Protein Rollup Plots These are similar to the ones shown in 'Peptides and Protein Rollup Methods'

Cluster Heatmaps A simple heatmap display control is added with hierachical and k-means clustering options. The data rows to be used for the plot can be either specified in the dialog box below or can be selected in the datatable. If a factor is selected, a colorbar will be added on top of the heatmap denoting the grouping information of datasets. Cluster ordering for Hierarchical clusters and the cluster assignments for k-means clustering will be displayed in a separate data table "Clusters".

Statistics Menu Define Factors Factors will identify the groups within the datasets. For an example, the experiment may have Male and Female individuals, or the samples are from healthy individuals and cancerous patients, or the order they were run in batches (blocking factors) can be used as factors. See File Menu for loading factors from a file.

Shapiro-Wilks Test for Normality Shapiro-Wilks Test is used to check the normality of the data. This is supposed to be better than the Kolmogorov-Smirnov test and other similar goodness of fit tests. Royston (1995) reports that the null hypothesis (i.e. data is normally distributed) can be rejected for p-value < 0.1 (see: Patrick Royston (1995) Remark AS R94: A remark on Algorithm AS 181: The W test for normality. Applied Statistics, 44, 547–551).

ANOVA There is a comprehensive ANOVA scheme included in DAnTE. This model can account for unbalanced data using marginal sums of squares in the case of 2-way ANOVA (or n-Way) and can account for random effects such as LC column effects etc through a REML based multi level model. User can also check interactions in a higher order (n-Way) ANOVA. The output will be the p-values and the corresponding q-values that denote the FDR thresholds (see: Storey JD. (2003) The positive false discovery rate: A Bayesian interpretation and the q-value. Annals of Statistics, 31: 2013-2035).

Non-parametric Tests Wilcoxon Rank Sum Test This test is equivalent to the Mann-Whitney test and can be applied in place of the t-test when the normality assumption does not hold. This test can be applied only for a factor with two levels. For factors with more than two levels the Kruskal-Walis test (see below) should be used instead.

Kruskal-Walis Test This test can be applied in place of the One-way ANOVA when the normality assumption does not hold.

Fold Changes

Fold changes can be calculated among two levels of a Factor. Data is assumed to be in log scale and the fold change is calculated by first avaraging the values in datasets for each factor level and subtracting one from the other noting that log(x/y) = log(x) - log(y).

If you uncheck the check box to denote that the data is in normal scale the usual ratios are obtained by division. For ratios smaller than 1, -1/ratio is reported as a negative fold change. Absolute fold change value is also reported. Menu location: Statistics -> Calculate Fold Changes.

p/q-value Filters Based on the resulting p-values or q-values the data can be filtered. The dialog box is shown below.

Miscellaneous Features Merging columns of a data table Columns can be merged based on a factor. Merging can be done as a sum, median, or as mean. If this table to be used for subsequent analysis, it has to be save and loaded to a new analysis session since the old factor definitions would no longer be valid. However the statistical plots can be done on this table.

Plot rows of a data table Select Rollup->Plot Rows and select rows or row metadata to plot.

Save data table with protein information If the data table needs to be saved with protein ID columns appended, right click the table from the tree view on the left to bring up the context menu.

Save/Open Session Choose File -> Save Session to save the entire analysis session so that it can be later retrieved by File -> Open Session.

Step-by-step Analysis Summary Abundance Data 1. First step would be to log transform the data. 2. You can investigate the various plots available in DAnTE to study the variation and to decide on the type of normalization method to apply. 3. Define factors so that the normalization methods can be applied to groups of datasets categorized by factors. 4. Apply normalization. There are few major methods available: Linear Regression, Loess and Quantile. These can be thought of as applying locally within a selected factor. In addition a global methods can also be applied. These are the Median Absolute Deviation (MAD) correction and the mean/median centering. Mean centering is recommended as the final step after MAD adjustment. 5. If the protein information is loaded, data can be rolled up to protein abundances using one of the three methods available. Note that the RRollup method assumes that the data is in log scale.

Peptide Count Data 1. Load the data using the wizard. 2. Investigate using the scatter plots available under the correlation plots to see for any systematic shift in the data and correct if any, using the linear regression method. 3. Log transform the data using an appropriate bias. 4. Use a non-parametric test (ex: Kruskal-Walis test) to select the significant features.

Author(s)

Tom Taverner <[e-mail]>

Examples

  dante()

DanteR documentation built on May 2, 2019, 6:11 p.m.