Differential Expression Analysis.

Author

Functional Genomics Center Zurich

Published

July 29, 2026

Overview

Samples12

Groups3

Proteins100

Visual abstract: quantified protein abundances from two sample groups are normalized and statistically compared to produce differential-expression results.

This report tests whether normalized protein abundances differ between experimental groups. It provides the sample design, quality diagnostics, and statistical evidence needed to interpret differential protein abundance.

Input: quantified protein abundances from 12 samples, analysed with DIANN. The detailed source-data reference is recorded in Session Info.

Analysis background

The differential expression analysis verifies if the difference between normalized empirical protein abundances measured in two groups is significantly non-zero. To make the test as sensitive and specific as possible, the methods used to measure (Taverna and Gaspari 2021) and estimate protein abundances (Grossmann et al. 2010) are optimized to minimize the biochemical and technical variance. In addition, these empirical abundances $A$ are further $log_2$ transformed and scaled to make them compatible with the statistical test procedure (Välikangas et al. 2018). Therefore, we obtain a scale free $\log_2$ transformed normalized protein abundances $\log_2(A)$ for a sample.

For unpaired experiments the difference $\Delta$ between group $a$ and $b$, for a specific protein is estimated by:

\[ \Delta = \frac{1}{n}\sum_{i=1}^n \log_2(A^a_i) - \frac{1}{m}\sum_{j=1}^{m}\log_2(A^b_j) \]

where $A^a_i$ - is the normalized protein abundance of sample $i$ in the group $a$ of $n$ samples, while $A^b_j$ is the protein abundance of sample $j$ in group $b$ of $m$ samples.

For paired experiments, the difference is estimated by:

\[ \Delta = \frac{1}{n}\sum_{i=1}^n \log_2(A^a_i) - \log_2(A^b_{i}) \]

where $n$ is the number of subjects, each treated with $a$ and $b$.

Of note, when comparing two samples $a$ and $b$ the difference of logarithms equals the logarithm of the ratio (exponent rule):

\[ \log(A^a) - \log(A^b) = \log(\frac{A^a}{A^b}) \]

It is called the $\log$-ratio or $\log$ fold-change (logFC).

The estimated differences $\Delta$ have an associated error $\epsilon$. Therefore, the differential expression analysis must test if the difference is significantly nonzero.

We run a set of functions implemented in the R package [prolfqua] (Wolski et al. 2023) to filter and normalize the data, generate visualizations, and to compute differential expression analysis. To further improve the power of the differential expression test the protein variances are moderated (Smyth 2004), i.e. the individual protein variances are updated using a variance prior estimated from all the proteins in the experiment.

Results

Table 1 shows the number of samples assigned to each group while Table 2 shows the names of the files assigned to the group.

Table 1: Nr of samples assigned to each group.

group_	# samples
A	4
B	4
Ctrl	4

Table 2: LC-MS samples annotation table. The content of the sampleName column is used as a short form to plot labels. The group to which a sample is assigned is shown in the column group.

sample	sampleName	group_	nr_1	nr_2
A_V1	A_V1	A	97	69
A_V2	A_V2	A	98	73
A_V3	A_V3	A	100	68
A_V4	A_V4	A	97	72
B_V1	B_V1	B	96	74
B_V2	B_V2	B	99	72
B_V3	B_V3	B	97	75
B_V4	B_V4	B	98	71
Ctrl_V1	Ctrl_V1	Ctrl	100	72
Ctrl_V2	Ctrl_V2	Ctrl	96	69
Ctrl_V3	Ctrl_V3	Ctrl	97	72
Ctrl_V4	Ctrl_V4	Ctrl	98	72

Peptide and Protein identification

The protein matrix is filtered using the following threshold:

Minimum number of peptides / protein: 1.

The overall number of proteins in this experiment is: 100. The percentage of contaminant proteins is:0 %. The percentage of decoy proteins among quantified entries is: 0 %.

We keep the contaminant proteins because, for some experiments, these contaminants are relevant. They are recognizable by identifiers starting with zz or CON, are flagged in the result tables, and are drawn as triangles in the volcano plot. Decoy sequences (identifiers starting with REV or rev) are retained in the raw and abundance exports for traceability, but they are excluded from the model fit so they cannot distort the variance moderation that is shared across proteins. They therefore carry no fold change, p-value, or FDR, and do not appear in the volcano plot or the differentially-expressed set. Their proportion among quantified entries (reported above) serves as an empirical false-discovery signal for the whole quantified data set — not for the differentially-expressed list.

Figure 1 shows the number of quantified proteins with one or more peptides per sample.

Figure 1: Number of quantified proteins per sample, counting proteins identified with at least one peptide; bar height is the protein count per sample.

Figure 2 shows the number of quantified proteins with two or more peptides per sample.

Figure 2: Number of quantified proteins per sample, counting only proteins identified with at least two peptides; bar height is the protein count per sample.

Missing Value Analysis

The absence of a protein measurement in a sample might be biologically relevant or might point to technical problems. Significant differences in the set of proteins observed in the samples within a group typically indicate either technical problems or excessive biological variability. If one sample out of ten has a different set of proteins, it is likely an outlier and can be removed from the analysis. If the differences between the groups are significant but within the groups are small, this might systematically bias the difference estimates, i.e., produce false-positive or false-negative test results.

A dichotomous view of the data can be constructed by transforming protein abundance estimates into present/absent calls (Figure 3). The heatmap shows only proteins with at least one missing value. There are 20 proteins with at least one missing value in the data, which is (20 %).

We expect that samples in the same group are more similar and cluster together, i.e., they are in the same branch of the dendrogram.

Figure 3: Protein abundance heatmap (rows indicate proteins, columns indicate samples) showing missing protein abundance estimates across the data set. Rows and columns are grouped based on the Minkowski distance using hierarchical clustering. White: Protein is observed, black: Protein is not observed.

Using Figure 4 we examine if we see the same proteins in each group. We say a protein is unobserved in the group if it is absent in all samples and is present otherwise. A significant overlap among groups allows more precise estimation of the protein abundance differences between the groups.

Figure 4: UpSet plot of protein presence across groups. Bar heights give the number of proteins observed in each group and in each group intersection; a protein counts as present in a group when observed in at least one of its samples.

Protein Abundance Analysis

The density plot (Figure 5 left panel) displays the protein abundance distribution for all data set samples. Major differences between samples could indicate that the individual protein abundance values are affected by technical biases. These biases might need to be corrected to separate them from biological effects. The right panel of Figure 5 shows the distribution of the transformed and scaled normalized empirical protein abundances. Normalization is applied to remove systematic differences in protein abundances due to different sample concentrations or amounts of sample loaded on a column. However, in the presence of a large proportion of missing data, normalization potentially amplifies systematic errors.

To do this the z-score of the $\log_2$ transformed protein abundances are computed. Because we need to estimate the protein differences on the original scale, we have to multiply the $z$-score by the average standard deviation of all the $N$ samples in the experiment. After normalization all samples have an equal mean and variance and a similar distribution.

Figure 5: Kernel density function showing the distribution of protein abundances in all samples. Left panel: Empirical protein abundance of all samples in the dataset. Right panel: Normalized empirical protein abundances of all samples in the dataset.

The median coefficient of variation (CV) of a group of samples or all samples in the dataset, can be used to compare the experiment with other experiments (Piehowski et al. 2013). For example, the median CV for high-performance liquid chromatography experiments ranges from 2% to 35% depending on the biological samples studied, the chromatography method used, label-free or labelled quantification (Taverna and Gaspari 2021).

Figure 6 shows the coefficients of variation (CV) for all proteins computed on non-normalized data. Ideally the within group CV should be smaller than the CV of all samples.

Figure 6: Coefficient of variation (CV) of non-normalized protein abundances, shown as a violin per group and across all samples (all); the overlaid point marks the median CV.

Table 3 shows the median CV of all groups and across all samples (all).

Table 3: Median coefficient of variation (CV) of non-normalized protein abundances, per group and across all samples (all).

what	A	B	Ctrl	All
CV	3.47	3.37	3.64	9.32

The protein abundance heatmap (Figure 7) groups the protein and samples using unsupervised hierarchical clustering. Distances between proteins and samples are computed using normalized protein abundances. Proteins with a large proportion of missing observations are not shown in this heatmap, because for these proteins no distance can be computed. Proteins and samples showing similar abundances are grouped and shown in adjacent rows and columns respectively.

Figure 7: Protein abundance heatmap (rows indicate proteins, columns indicate samples) showing the row scaled $\log_2$ transformed protein abundance value. Co-clustering (hierarchical complete linkage, euclidean distance) of samples and proteins was used. Cell color encodes the row z-score on a green-black-red scale (green: low, black: average, red: high); missing values are shown in light gray.

We use principal component analysis (PCA) to transform the high-dimensional space defined by all proteins into a two-dimensional one containing most of the information. Figure 8 shows the location of the samples according to the first and second principal component, which explain most of the variance in the data. Samples close in the PCA plot are more similar than those farther apart.

Figure 8: Plot of first and second principal component (PC1 and PC2) of principal component analysis (PCA). Normalized abundances were used as input.

Differential Expression Analysis

The method used to test for differential expression consists of several steps: First a model that explains the observed protein abundances using the grouping of the samples is fitted to each protein with the selected backend (the modelName/facade key; by default a linear model fitted with the R function lm):

normalized_abundance ~ group_.

Secondly, the difference between the groups is computed (Table 4).

Table 4: Name of difference (Contrast), and formula used to compute it.

name	contrast
AVsC	group_A - group_Ctrl
BVsC	group_B - group_Ctrl

and a null hypothesis significance test (NHST) is conducted, where the null hypothesis is that the protein is not differentially expressed (Faraway 2004).

If there are no abundances measured in one of the groups for some proteins, we assume the observations are missing because the protein abundance is below the detection limit. Therefore, we estimate the detection limit using the mean of the $1\%$ smallest group averages. Furthermore, to make it explicit how each estimate was produced, the estimate_type column (see table in Figure 9) records whether a row was fit from observed data, rescued via LOD imputation (lod_imputed), or filled from a group-mean fallback (missing_fallback); rescued/fallback rows are visualized with a distinct colour in Figure 10. The modelName column records the selected analysis method (the facade key, e.g. lm, rfit, limma).

Next, to increase the power of the analysis variance shrinkage is performed (Smyth 2004). Finally, the false discovery rate (FDR) using the Benjamini-Hochberg procedure is computed (Benjamini and Hochberg 1995). The FDR is the expected proportion of false discoveries in a list of proteins, and can be used to select candidates for follow up experiments. FDR thresholds commonly used are 5, 10 or 25%. By filtering the proteins using an FDR threshold of 10 % we can expect this proportion of false positives in the list and 90 % truly differentially expressed proteins. Because we do not know which of them are true positives follow up experiments are necessary.

The table (Figure 9) summarizes the differential expression analysis results by providing the following information:

protein_Id - unique protein identifier
description - information about the protein provided in the FASTA database
contrast - name of the comparison
modelName - selected analysis method (facade key, e.g. lm, rfit, limma)
estimate_type - how the estimate was produced: observed, lod_imputed, or missing_fallback
FDR - false discovery rate
diff - difference between groups.

The volcano plot Figure 10 helps to identify proteins with large differences among groups and a low FDR. The significance dimension is a $-\log_{10}$ transformed FDR, i.e., small values of FDR become large after transformation. Promising candidate proteins are found in the upper right and left sector of the plot.

Figure 9: Interactive, filterable table of per-protein, per-contrast differential expression results: protein ID, description, contrast, model, estimate type, log2 fold-change and FDR.

Figure 10: Volcano plot showing $-\log_{10}$ transformed FDR as function of the difference between groups. The red line indicates the $-log_{10}(FDR)$ of FDR = 0.1, while the green lines represent the difference of minus and plus 0.2. Points are coloured by estimate_type, so rows estimated by missing-value imputation or group-mean fallback are visually distinguished from observed fits.

Differentially Expressed Proteins

Here we use the FDR threshold of 0.1 and a difference threshold of 0.2 to select differentially expressed proteins. Table 5 summarizes the number of significant calls.

Table 5: Number of proteins called significant versus not significant per contrast, applying the fold-change and FDR thresholds (total, significant, not significant).

contrast	n	Significant	Not Significant
AVsC	100	40	60
BVsC	100	37	63

The table shown in Figure 11 lists all the significant proteins.

Figure 11: Significant proteins obtained by applying the difference and FDR thresholds.

Furthermore, Figure 12 shows a co-clustered heatmap of the row z-scored protein abundances of all significant calls. Proteins and samples with similar abundance profiles are grouped by unsupervised hierarchical clustering.

Figure 12: Heatmap of the proteins passing the FDR and difference thresholds. Cell colour encodes the row z-score of the $\log_2$ protein abundance (green: low, black: average, red: high; missing values in light gray). Proteins and samples are co-clustered by unsupervised hierarchical clustering (complete linkage, euclidean distance).

Figure 13: UpSet plot of significant proteins across contrasts. Bar heights give the number of proteins passing the FDR and fold-change thresholds in each contrast and in the contrast intersections.

Additional Analysis

The zip file contains an Excel file DE_Groups_vs_Controls.xlsx. All the figures can be recreated using the data in the excel file. The Excel file contains the following spreadsheets:

annotation - the annotation of the samples in the experiment
raw_abundances table with empirical protein abundances.
normalized_abundances table with normalized protein abundances.
raw_abundances_matrix A table where each column represents a sample and each row represents a protein and the cells store the empirical protein abundances.
normalized_abundances_matrix A table where each column represents a sample and each row represents a protein and the cells store the empirical protein abundances.
diff_exp_analysis A table with the results of the differential expression analysis. For each protein there is a row containing the estimated difference between the groups, the false discovery rate FDR, the 95% confidence interval, the posterior degrees of freedom.
missing_information - spreadsheet containing information if a protein is present (1) or absent in a group (0).
protein_variances - spreadsheet which for each protein shows the variance (var) or standard deviation (sd) within a group, the number of samples (n) and the number of observations (not_na) as well as the group average intensity (mean).

The data can be used to perform functional enrichment analysis (Monti et al. 2019). To compare the obtained results with known protein interactions we recommend the string-db.org (Szklarczyk et al. 2017), which is a curated database of protein-protein interaction networks for a large variety of organisms. To simplify the data upload to string-db we include text files containing the uniprot ids:

ORA_background.txt all proteins.
ORA_<contrast_name>.txt proteins accepted with the FDR and difference threshold.

Other web applications allowing to run over representation analysis (ORA) (Monti et al. 2019) are:

Furthermore, protein IDs sorted by t-statistic can then be subjected to gene set enrichment analysis (GSEA) (Subramanian et al. 2005). To simplify running GSEA, we provide the file:

GSEA_<contrast_name>.rnk

This file can be used with the webgestalt web application or used with the GSEA application from gsea-msigdb

For questions and improvement suggestions, with respect to this report, please contact protinf@fgcz.uzh.ch.

Session Info

Table 6: Compact report provenance, including the source input-data reference.

Field	Value
Workunit ID	23000
Order ID	6200
Project ID	3000
Project name	n/a
Creator	runner
Created at	2026-07-29 11:16:46 UTC
Input data	https://fgcz-bfabric.uzh.ch/bfabric/
Quantification software	DIANN
Model	lm_impute
prolfquapp version	2.6.1

R version 4.6.1 (2026-06-24)

Platform: x86_64-pc-linux-gnu

locale: LC_CTYPE=C.UTF-8, LC_NUMERIC=C, LC_TIME=C.UTF-8, LC_COLLATE=C.UTF-8, LC_MONETARY=C.UTF-8, LC_MESSAGES=C.UTF-8, LC_PAPER=C.UTF-8, LC_NAME=C, LC_ADDRESS=C, LC_TELEPHONE=C, LC_MEASUREMENT=C.UTF-8 and LC_IDENTIFICATION=C

attached base packages: stats, graphics, grDevices, utils, datasets, methods and base

other attached packages: dplyr(v.1.2.1)

loaded via a namespace (and not attached): RColorBrewer(v.1.1-3), jsonlite(v.2.0.0), shape(v.1.4.6.1), magrittr(v.2.0.5), jomo(v.2.7-6), farver(v.2.1.2), logistf(v.1.26.1), nloptr(v.2.2.1), rmarkdown(v.2.31), GlobalOptions(v.0.1.4), vctrs(v.0.7.3), minqa(v.1.2.8), progress(v.1.2.3), htmltools(v.0.5.9), S4Arrays(v.1.12.0), forcats(v.1.0.1), broom(v.1.0.13), cellranger(v.1.1.0), SparseArray(v.1.12.2), mitml(v.0.4-5), sass(v.0.4.10), bslib(v.0.11.0), htmlwidgets(v.1.6.4), plyr(v.1.8.9), cachem(v.1.1.0), plotly(v.4.12.1), mime(v.0.13), lifecycle(v.1.0.5), iterators(v.1.0.14), pkgconfig(v.2.0.3), Matrix(v.1.7-5), R6(v.2.6.1), fastmap(v.1.2.0), shiny(v.1.14.0), rbibutils(v.2.4.1), MatrixGenerics(v.1.24.0), clue(v.0.3-68), digest(v.0.6.39), dtplyr(v.1.3.3), colorspace(v.2.1-3), lobstr(v.1.2.1), S4Vectors(v.0.50.1), crosstalk(v.1.2.2), GenomicRanges(v.1.64.0), labeling(v.0.4.3), httr(v.1.4.8), abind(v.1.4-8), mgcv(v.1.9-4), compiler(v.4.6.1), withr(v.3.0.3), bit64(v.4.8.2), doParallel(v.1.0.17), pander(v.0.6.6), S7(v.0.2.2), backports(v.1.5.1), logger(v.0.4.2), UpSetR(v.1.4.1), prolfquasaint(v.0.1.5), pan(v.2.0), MASS(v.7.3-65), DelayedArray(v.0.38.2), rjson(v.0.2.23), optparse(v.1.8.2), tools(v.4.6.1), otel(v.0.2.0), httpuv(v.1.6.17), nnet(v.7.3-20), glue(v.1.8.1), promises(v.1.5.0), nlme(v.3.1-169), grid(v.4.6.1), cluster(v.2.1.8.2), generics(v.0.1.4), operator.tools(v.1.6.3.1), gtable(v.0.3.6), tzdb(v.0.5.0), formula.tools(v.1.7.1), preprocessCore(v.1.74.0), tidyr(v.1.3.2), data.table(v.1.18.4), hms(v.1.1.4), XVector(v.0.52.0), BiocGenerics(v.0.58.1), ggrepel(v.0.9.8), foreach(v.1.5.2), pillar(v.1.11.1), stringr(v.1.6.0), limma(v.3.68.4), later(v.1.4.8), circlize(v.0.4.18), splines(v.4.6.1), lattice(v.0.22-9), survival(v.3.8-6), bit(v.4.6.0), tidyselect(v.1.2.1), ComplexHeatmap(v.2.28.0), knitr(v.1.51), reformulas(v.0.4.4), gridExtra(v.2.3.1), prolfquapp(v.2.6.1), bookdown(v.0.47), IRanges(v.2.46.0), Seqinfo(v.1.2.0), SummarizedExperiment(v.1.42.0), stats4(v.4.6.1), xfun(v.0.60), prolfqua(v.1.7.0), Biobase(v.2.72.0), statmod(v.1.5.2), matrixStats(v.1.5.0), DT(v.0.34.0), stringi(v.1.8.7), lazyeval(v.0.2.3), yaml(v.2.3.12), boot(v.1.3-32), evaluate(v.1.0.5), codetools(v.0.2-20), tibble(v.3.3.1), BiocManager(v.1.30.27), cli(v.3.6.6), affyio(v.1.82.0), rpart(v.4.1.27), xtable(v.1.8-8), arrow(v.25.0.0), Rdpack(v.2.6.6), jquerylib(v.0.1.4), Rcpp(v.1.1.2), readxl(v.1.5.0), png(v.0.1-9), parallel(v.4.6.1), ggplot2(v.4.0.3), readr(v.2.2.0), assertthat(v.0.2.1), prettyunits(v.1.2.0), lme4(v.2.0-6), glmnet(v.5.0), viridisLite(v.0.4.3), scales(v.1.4.0), affy(v.1.90.0), purrr(v.1.2.2), crayon(v.1.5.3), writexl(v.1.5.4), GetoptLong(v.1.1.1), rlang(v.1.3.0), vsn(v.3.80.0) and mice(v.3.19.0)

References

Benjamini, Yoav, and Yosef Hochberg. 1995. “Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing.” Journal of the Royal Statistical Society: Series B (Methodological) 57 (1): 289–300.

Faraway, Julian J. 2004. Linear Models with r. Chapman; Hall/CRC.

Grossmann, Jonas, Bernd Roschitzki, Christian Panse, et al. 2010. “Implementation and Evaluation of Relative and Absolute Quantification in Shotgun Proteomics with Label-Free Methods.” Journal of Proteomics 73 (9): 1740–46.

Monti, Chiara, Mara Zilocchi, Ilaria Colugnat, and Tiziana Alberio. 2019. “Proteomics Turns Functional.” Journal of Proteomics 198: 36–44.

Piehowski, Paul D, Vladislav A Petyuk, Daniel J Orton, et al. 2013. “Sources of Technical Variability in Quantitative LC–MS Proteomics: Human Brain Tissue Sample Analysis.” Journal of Proteome Research 12 (5): 2128–37.

Smyth, Gordon K. 2004. “Linear Models and Empirical Bayes Methods for Assessing Differential Expression in Microarray Experiments.” Statistical Applications in Genetics and Molecular Biology 3 (1): 1–25.

Subramanian, Aravind, Pablo Tamayo, Vamsi K Mootha, et al. 2005. “Gene Set Enrichment Analysis: A Knowledge-Based Approach for Interpreting Genome-Wide Expression Profiles.” Proceedings of the National Academy of Sciences 102 (43): 15545–50.

Szklarczyk, Damian, John H Morris, Helen Cook, et al. 2017. “The STRING Database in 2017: Quality-Controlled Protein-Protein Association Networks, Made Broadly Accessible.” Nucleic Acids Research 45 (January): D362–68. https://doi.org/10.1093/nar/gkw937.

Taverna, Domenico, and Marco Gaspari. 2021. “A Critical Comparison of Three MS-Based Approaches for Quantitative Proteomics Analysis.” Journal of Mass Spectrometry 56 (1): e4669.

Välikangas, Tommi, Tomi Suomi, and Laura L Elo. 2018. “A Systematic Evaluation of Normalization Methods in Quantitative Label-Free Proteomics.” Briefings in Bioinformatics 19 (1): 1–11.

Wang, Jing, Suhas Vasaikar, Zhiao Shi, Michael Greer, and Bing Zhang. 2017. “WebGestalt 2017: A More Comprehensive, Powerful, Flexible and Interactive Gene Set Enrichment Analysis Toolkit.” Nucleic Acids Research 45 (July): W130–37. https://doi.org/10.1093/nar/gkx356.

Wolski, Witold E., Paolo Nanni, Jonas Grossmann, Maria d’Errico, Ralph Schlapbach, and Christian Panse. 2023. “Prolfqua: A Comprehensive R-Package for Proteomics Differential Expression Analysis.” Journal of Proteome Research 4 (22): 1092–104. https://doi.org/10.1021/acs.jproteome.2c00441.

Glossary

groups - different treatments, genotypes etc.
diff (difference) - it is the difference of the protein abundance estimate of two groups
FDR - false discovery rate

This report was generated from the Quarto template Grp2Analysis_V2_R6.qmd included in the prolfquapp R package (version 2.6.1).

--- title: "Differential Expression Analysis." author: "Functional Genomics Center Zurich" date: today format: html: toc-location: left grid: sidebar-width: 300px include-after-body: fgcz-plot-finder.html toc: true params: deanalyse_file: null vignette: > %\VignetteIndexEntry{Differential Expression Analysis} %\VignetteEngine{quarto::format} %\VignetteEncoding{UTF-8} bibliography: bibliography.bib --- ```{r setup, include = FALSE} knitr::opts_chunk$set( echo = FALSE, message = FALSE, warning = FALSE, out.width = "600px", fig.align = "center" ) deanalyse_file <- params$deanalyse_file if (is.null(deanalyse_file) || !nzchar(deanalyse_file)) { deanalyse_file <- file.path(tempdir(), "prolfquapp-example-deanalyse.rds") if (!file.exists(deanalyse_file)) { saveRDS(prolfquapp::example_deanalyse(), deanalyse_file) } } dea <- readRDS(deanalyse_file) cfg <- dea$prolfq_app_config is_saint_report <- identical(dea$default_model, "saint") report_provenance <- prolfquapp:::.report_provenance( project_spec = cfg$project_spec, software = cfg$software, model = dea$default_model ) empty_report_plot <- function(label) { ggplot2::ggplot() + ggplot2::annotate("text", x = 0, y = 0, label = label) + ggplot2::theme_void() } plot_or_empty <- function(plot, label) { if (is.null(plot)) { return(empty_report_plot(label)) } plot } sized_widget <- function(widget, width = 600) { htmltools::tags$div( style = sprintf("max-width: %dpx; margin: 0 auto;", width), widget ) } ``` # Overview ```{r overview-data-summary} #| results: asis cat(prolfquapp:::.report_overview_cards(dea$lfq_data_raw)) ``` <img src="`r knitr::image_uri(system.file("report-assets", "differential-expression.png", package = "prolfquapp"))`" alt="Visual abstract: quantified protein abundances from two sample groups are normalized and statistically compared to produce differential-expression results." style="width: 100%; max-height: 22rem; object-fit: contain;"> This report tests whether normalized protein abundances differ between experimental groups. It provides the sample design, quality diagnostics, and statistical evidence needed to interpret differential protein abundance. Input: quantified protein abundances from `r nrow(dea$lfq_data_raw$factors())` samples, analysed with `r report_provenance$software`. The detailed source-data reference is recorded in **Session Info**. ```{r introduction} #| results: asis #| eval: !expr "!get0('is_saint_report', ifnotfound = FALSE)" cat(r"( # Analysis background The differential expression analysis verifies if the __difference__ between normalized empirical protein abundances measured in two groups is significantly non-zero. To make the test as sensitive and specific as possible, the methods used to measure [@taverna2021critical] and estimate protein abundances [@grossmann2010implementation] are optimized to minimize the biochemical and technical variance. In addition, these empirical abundances $A$ are further $log_2$ transformed and scaled to make them compatible with the statistical test procedure [@valikangas2018systematic]. Therefore, we obtain a scale free $\log_2$ transformed normalized protein abundances $\log_2(A)$ for a sample. For unpaired experiments the difference $\Delta$ between group $a$ and $b$, for a specific protein is estimated by: $$ \Delta = \frac{1}{n}\sum_{i=1}^n \log_2(A^a_i) - \frac{1}{m}\sum_{j=1}^{m}\log_2(A^b_j) $$ where $A^a_i$ - is the normalized protein abundance of sample $i$ in the group $a$ of $n$ samples, while $A^b_j$ is the protein abundance of sample $j$ in group $b$ of $m$ samples. For paired experiments, the difference is estimated by: $$ \Delta = \frac{1}{n}\sum_{i=1}^n \log_2(A^a_i) - \log_2(A^b_{i}) $$ where $n$ is the number of subjects, each treated with $a$ and $b$. Of note, when comparing two samples $a$ and $b$ the difference of logarithms equals the logarithm of the ratio (exponent rule): $$ \log(A^a) - \log(A^b) = \log(\frac{A^a}{A^b}) $$ It is called the $\log$-ratio or $\log$ fold-change (logFC). The estimated differences $\Delta$ have an associated error $\epsilon$. Therefore, the differential expression analysis must test if the difference is significantly nonzero. We run a set of functions implemented in the R package _[prolfqua]_ [@prolfquawolski2023] to filter and normalize the data, generate visualizations, and to compute differential expression analysis. To further improve the power of the differential expression test the protein variances are moderated [@Smyth2004], i.e. the individual protein variances are updated using a variance prior estimated from all the proteins in the experiment. )") ``` # Results @tbl-samples shows the number of samples assigned to each group while @tbl-annotation shows the names of the files assigned to the group. ```{r} #| label: tbl-samples #| tbl-cap: "Nr of samples assigned to each group." #| eval: true tab <- data.frame(table(dea$lfq_data_raw$factors()[dea$lfq_data_raw$relevant_factor_keys()])) colnames(tab)[length(colnames(tab))] <- "# samples" knitr::kable(tab) ``` ```{r} #| label: tbl-annotation #| tbl-cap: !expr paste0("LC-MS samples annotation table. The content of the sampleName column ", "is used as a short form to plot labels. The group to which a sample ", "is assigned is shown in the column group.") #| eval: true ld <- dea$lfq_data_raw fx <- ld$factors() sr <- ld$get_Summariser() xx <- sr$hierarchy_counts_sample(nr_children = 1) xx$isotopeLabel <- NULL xx <- xx |> dplyr::rename(nr_1 = ld$hierarchy_keys()) x2 <- sr$hierarchy_counts_sample(nr_children = 2) x2$isotopeLabel <- NULL x2 <- x2 |> dplyr::rename(nr_2 = ld$hierarchy_keys()) fx <- dplyr::left_join(fx,xx, by = ld$sample_name()) fx <- dplyr::left_join(fx,x2, by = ld$sample_name()) knitr::kable(fx) ``` ## Peptide and Protein identification The protein matrix is filtered using the following threshold: - Minimum number of peptides / protein: `r cfg$processing_options$nr_peptides`. The overall number of proteins in this experiment is: `r dea$summary$totalNrOfProteins`. The percentage of contaminant proteins is:`r dea$summary$percentOfContaminants` %. The percentage of decoy proteins among quantified entries is: `r dea$summary$percentOfDecoys` %. We keep the contaminant proteins because, for some experiments, these contaminants are relevant. They are recognizable by identifiers starting with __zz__ or __CON__, are flagged in the result tables, and are drawn as triangles in the volcano plot. Decoy sequences (identifiers starting with __REV__ or __rev__) are retained in the raw and abundance exports for traceability, but they are **excluded from the model fit** so they cannot distort the variance moderation that is shared across proteins. They therefore carry no fold change, p-value, or FDR, and do not appear in the volcano plot or the differentially-expressed set. Their proportion among quantified entries (reported above) serves as an empirical false-discovery signal for the whole quantified data set — not for the differentially-expressed list. @fig-nr-per-sample shows the number of quantified proteins with one or more peptides per sample. ```{r} #| label: fig-nr-per-sample #| fig-cap: "Number of quantified proteins per sample, counting proteins identified with at least one peptide; bar height is the protein count per sample." #| fig-width: 10 #| fig-height: 7 sum <- dea$lfq_data_raw$get_Summariser() sum$plot_hierarchy_counts_sample(nr_children = 1) ``` ```{r prepNrPerSample2} #| include: false sum <- dea$lfq_data_raw$get_Summariser() p_nr_per_sample_2 <- tryCatch( sum$plot_hierarchy_counts_sample(nr_children = 2), error = function(e) NULL ) ``` @fig-nr-per-sample-2 shows the number of quantified proteins with two or more peptides per sample. ```{r} #| label: fig-nr-per-sample-2 #| fig-cap: "Number of quantified proteins per sample, counting only proteins identified with at least two peptides; bar height is the protein count per sample." #| fig-width: 10 #| fig-height: !expr if (is.null(p_nr_per_sample_2)) 1.2 else 7 plot_or_empty( p_nr_per_sample_2, "No features with two or more children are available for this data set." ) ``` ## Missing Value Analysis The absence of a protein measurement in a sample might be biologically relevant or might point to technical problems. Significant differences in the set of proteins observed in the samples within a group typically indicate either technical problems or excessive biological variability. If one sample out of ten has a different set of proteins, it is likely an outlier and can be removed from the analysis. If the differences between the groups are significant but within the groups are small, this might systematically bias the difference estimates, i.e., produce false-positive or false-negative test results. ```{r countProtWithNAs} res <- dea$lfq_data_raw$data_wide(as.matrix = TRUE)$data res[!is.na(res)] <- 0 res[is.na(res)] <- 1 allrows <- nrow(res) res <- res[apply(res,1, sum) > 0, , drop = FALSE] allNas <- nrow(res) ``` ```{r prepHeatmap} pl <- dea$lfq_data_raw$get_Plotter() nah <- tryCatch( pl$na_heatmap(), error = function(e) NULL ) ``` A dichotomous view of the data can be constructed by transforming protein abundance estimates into present/absent calls (@fig-na-heat). The heatmap shows only proteins with at least one missing value. There are `r allNas` proteins with at least one missing value in the data, which is (`r allNas/allrows * 100` %). We expect that samples in the same group are more similar and cluster together, i.e., they are in the same branch of the dendrogram. ```{r} #| label: fig-na-heat #| fig-cap: "Protein abundance heatmap (rows indicate proteins, columns indicate samples) showing missing protein abundance estimates across the data set. Rows and columns are grouped based on the Minkowski distance using hierarchical clustering. White: Protein is observed, black: Protein is not observed." #| fig-width: 7 #| fig-height: !expr if (is.null(nah)) 1.2 else 6 #| fig-dpi: 300 #| fig-alt: "" #| eval: true plot_or_empty( nah, "No missing values are available for the missing-value heatmap." ) ``` Using @fig-venn-proteins we examine if we see the same proteins in each group. We say a protein is unobserved in the group if it is absent in all samples and is present otherwise. A significant overlap among groups allows more precise estimation of the protein abundance differences between the groups. ```{r} #| label: fig-venn-proteins #| fig-cap: "UpSet plot of protein presence across groups. Bar heights give the number of proteins observed in each group and in each group intersection; a protein counts as present in a group when observed in at least one of its samples." #| fig-alt: "" pups <- prolfqua::upset_interaction_missing_stats(dea$lfq_data_raw, tr = 1) UpSetR::upset(pups$data, order.by = "freq", nsets = pups$nsets) ``` ## Protein Abundance Analysis The density plot (@fig-normalized left panel) displays the protein abundance distribution for all data set samples. Major differences between samples could indicate that the individual protein abundance values are affected by technical biases. These biases might need to be corrected to separate them from biological effects. The right panel of @fig-normalized shows the distribution of the transformed and scaled normalized empirical protein abundances. Normalization is applied to remove systematic differences in protein abundances due to different sample concentrations or amounts of sample loaded on a column. However, in the presence of a large proportion of missing data, normalization potentially amplifies systematic errors. ```{r normalization} #| results: asis if (cfg$processing_options$transform == "robscale") { cat( "To do this the z-score of the $\\log_2$ transformed protein", "abundances are computed.", "Because we need to estimate the protein differences on the", "original scale, we have to multiply the $z$-score by the average", "standard deviation of all the $N$ samples in the experiment.", "After normalization all samples have an equal mean and variance", "and a similar distribution.") } else if (cfg$processing_options$transform == "vsn") { cat("To do this the variance stabilizing normalization (vsn) was applied [@HuberVSN2002].") } else if (cfg$processing_options$transform == "none") { cat( "However, in some circumstances it is advisable not to normalize", "the data, e.g. in case of affinity purification experiments,", "or when the requirements of sufficient similarity among samples", "are not met.") } ``` ```{r} #| label: fig-normalized #| fig-dpi: 96 #| fig-cap: "Kernel density function showing the distribution of protein abundances in all samples. Left panel: Empirical protein abundance of all samples in the dataset. Right panel: Normalized empirical protein abundances of all samples in the dataset." #| fig-width: 8 #| fig-height: 5 #| fig-alt: "" #| eval: true pl <- dea$lfq_data_raw$get_Plotter() showlegend <- nrow(dea$lfq_data_raw$factors()) <= 12 p1 <- pl$intensity_distribution_density(legend = showlegend) plTransformed <- dea$lfq_data$get_Plotter() p2 <- plTransformed$intensity_distribution_density(legend = showlegend) sized_widget( prolfquapp::plotly_ggplot_subplot( p1, p2, showlegend = showlegend, width = 600, height = 450 ) ) ``` ```{r internalCalibration} #| results: asis internal_ids <- cfg$processing_options$internal if (length(internal_ids) > 0 && !all(is.na(internal_ids)) && !all(internal_ids == "")) { cat("In addition to the global normalization described above, **internal standard normalization** was applied.") cat(" The following protein(s) were used as internal reference standard(s):\n\n") cat(paste0("- `", internal_ids, "`", collapse = "\n"), "\n\n") cat("After transformation, the per-sample intensity of each reference", "protein was subtracted from all proteins in that sample", "(`center_to_reference`). ") cat("As a result, the reference protein(s) have an abundance of exactly 0 in the normalized data, ") cat("and all other protein abundances are expressed relative to the reference.\n\n") } ``` The median coefficient of variation (CV) of a group of samples or all samples in the dataset, can be used to compare the experiment with other experiments [@piehowski2013sources]. For example, the median CV for high-performance liquid chromatography experiments ranges from 2% to 35% depending on the biological samples studied, the chromatography method used, label-free or labelled quantification [@taverna2021critical]. @fig-sd-violin shows the coefficients of variation (CV) for all proteins computed on non-normalized data. Ideally the within group CV should be smaller than the CV of all samples. ```{r} #| label: fig-sd-violin #| fig-cap: "Coefficient of variation (CV) of non-normalized protein abundances, shown as a violin per group and across all samples (all); the overlaid point marks the median CV." #| fig-height: 3 #| fig-width: 7 #| fig-alt: "" #| eval: true stR <- dea$lfq_data_raw$get_Stats() pA <- stR$violin() + ggplot2::theme_classic() + ggplot2::labs(tag = "A") pA ``` @tbl-cv shows the median CV of all groups and across all samples (all). ```{r} #| label: tbl-cv #| tbl-cap: "Median coefficient of variation (CV) of non-normalized protein abundances, per group and across all samples (all)." #| eval: true resR <- stR$stats_quantiles() C <- dplyr::bind_rows(CV = resR$wide |> dplyr::filter(probs == 0.5) |> round(digits = 2)) C <- C |> dplyr::mutate( what = c("CV"), .before = 1 ) C$probs <- NULL knitr::kable(C) ``` ```{r generateHeatmaps} #| include: false pl <- dea$lfq_data$get_Plotter() ph <- pl$heatmap() ``` The protein abundance heatmap (@fig-heatmap) groups the protein and samples using unsupervised hierarchical clustering. Distances between proteins and samples are computed using normalized protein abundances. Proteins with a large proportion of missing observations are not shown in this heatmap, because for these proteins no distance can be computed. Proteins and samples showing similar abundances are grouped and shown in adjacent rows and columns respectively. ```{r} #| label: fig-heatmap #| fig-cap: "Protein abundance heatmap (rows indicate proteins, columns indicate samples) showing the row scaled $\\log_2$ transformed protein abundance value. Co-clustering (hierarchical complete linkage, euclidean distance) of samples and proteins was used. Cell color encodes the row z-score on a green-black-red scale (green: low, black: average, red: high); missing values are shown in light gray." #| fig-width: 7 #| fig-height: 8 #| fig-dpi: 300 #| fig-alt: "" #| eval: true ph ``` We use principal component analysis (PCA) to transform the high-dimensional space defined by all proteins into a two-dimensional one containing most of the information. @fig-pca shows the location of the samples according to the first and second principal component, which explain most of the variance in the data. Samples close in the PCA plot are more similar than those farther apart. ```{r} #| label: fig-pca #| fig-dpi: 96 #| fig-cap: "Plot of first and second principal component (PC1 and PC2) of principal component analysis (PCA). Normalized abundances were used as input." #| fig-alt: "" #| fig-width: 7 #| fig-height: 7 pca_widget <- dea$lfq_data$get_Plotter()$pca_plotly() pca_widget$width <- 600 pca_widget$height <- 500 sized_widget(pca_widget) ``` # Differential Expression Analysis The method used to test for differential expression consists of several steps: First a model that explains the observed protein abundances using the grouping of the samples is fitted to each protein with the selected backend (the `modelName`/facade key; by default a linear model fitted with the R function _lm_): `r dea$formula`. Secondly, the difference between the groups is computed (@tbl-contrasts). ```{r} #| label: tbl-contrasts #| tbl-cap: "Name of difference (Contrast), and formula used to compute it." df <- data.frame(name = names(dea$contrasts), contrast = dea$contrasts) rownames(df) <- NULL knitr::kable(df) ``` and a null hypothesis significance test (NHST) is conducted, where the null hypothesis is that the protein is not differentially expressed [@faraway2004linear]. If there are no abundances measured in one of the groups for some proteins, we assume the observations are missing because the protein abundance is below the detection limit. Therefore, we estimate the detection limit using the mean of the $1\%$ smallest group averages. Furthermore, to make it explicit how each estimate was produced, the `estimate_type` column (see table in @fig-table-all) records whether a row was fit from `observed` data, rescued via LOD imputation (`lod_imputed`), or filled from a group-mean fallback (`missing_fallback`); rescued/fallback rows are visualized with a distinct colour in @fig-volcano. The `modelName` column records the selected analysis method (the facade key, e.g. `lm`, `rfit`, `limma`). Next, to increase the power of the analysis variance shrinkage is performed [@Smyth2004]. Finally, the false discovery rate (FDR) using the Benjamini-Hochberg procedure is computed [@benjamini1995controlling]. The FDR is the expected proportion of false discoveries in a list of proteins, and can be used to select candidates for follow up experiments. FDR thresholds commonly used are 5, 10 or 25%. By filtering the proteins using an FDR threshold of `r cfg$processing_options$FDR_threshold*100` % we can expect this proportion of false positives in the list and `r (1 - cfg$processing_options$FDR_threshold) * 100` % truly differentially expressed proteins. Because we do not know which of them are true positives follow up experiments are necessary. The table (@fig-table-all) summarizes the differential expression analysis results by providing the following information: ```{r} hkey <- dea$lfq_data$hierarchy_keys() ``` - `r if(hkey[1] == "protein_Id"){ paste0(hkey[1], " - unique protein identifier")} else {hkey[1]}` - description - information about the protein provided in the FASTA database - contrast - name of the comparison - modelName - selected analysis method (facade key, e.g. `lm`, `rfit`, `limma`) - estimate_type - how the estimate was produced: `observed`, `lod_imputed`, or `missing_fallback` - FDR - false discovery rate - diff - difference between groups. The volcano plot @fig-volcano helps to identify proteins with large differences among groups and a low FDR. The significance dimension is a $-\log_{10}$ transformed FDR, i.e., small values of FDR become large after transformation. Promising candidate proteins are found in the upper right and left sector of the plot. ```{r setupVolcano, echo = FALSE} datax <- as.data.frame(dea$annotated_contrasts) is_saint <- identical(dea$default_model, "saint") contrast_col <- if (is_saint) "Bait" else "contrast" effect_col <- if (is_saint) "log2_EFCs" else "diff" score_col <- if (is_saint) "BFDR" else "FDR" table_cols <- if (is_saint) { c("ID", "description", "Bait", "modelName", "log2_EFCs", "BFDR") } else { c("ID", "description", "contrast", "modelName", "estimate_type", "diff", "FDR") } volcano_cap <- paste0( "Volcano plot showing $-\\log_{10}$ transformed FDR as function of the difference ", "between groups. The red line indicates the $-log_{10}(FDR)$ of FDR = ", cfg$processing_options$FDR_threshold, ", while the green lines represent the difference of minus and plus ", cfg$processing_options$diff_threshold, ". Points are coloured by `estimate_type`, so rows estimated by missing-value ", "imputation or group-mean fallback are visually distinguished from observed fits." ) ``` ```{r} #| label: fig-table-all #| fig-cap: "Interactive, filterable table of per-protein, per-contrast differential expression results: protein ID, description, contrast, model, estimate type, log2 fold-change and FDR." datax <- datax |> tidyr::unite("ID" , tidyselect::all_of(hkey), sep = "~") bb <- datax |> dplyr::select( tidyselect::all_of(table_cols)) bb <- crosstalk::SharedData$new(bb, as.formula("~ ID"), group = "BB") DT::datatable(bb, filter = "bottom", extensions = "Scroller", style = "auto", class = "compact", options = list(deferRender = TRUE, scrollY = 300, scrollX = 400, scroller = TRUE) ) |> DT::formatRound(columns = c(score_col, effect_col), digits = 3) ``` ```{r} #| label: fig-volcano #| fig-dpi: 96 #| fig-cap: !expr volcano_cap #| fig-alt: "" #| fig-width: 9 #| fig-height: 7 #| eval: true plotter <- if (is_saint) { dea$contrast_results[[dea$default_model]]$get_Plotter( fc_threshold = cfg$processing_options$diff_threshold, bfdr_threshold = cfg$processing_options$FDR_threshold ) } else { dea$contrast_results[[dea$default_model]]$get_Plotter( fc_threshold = cfg$processing_options$diff_threshold, fdr_threshold = cfg$processing_options$FDR_threshold ) } volcano_widget <- plotter$volcano_plotly()$FDR volcano_widget$width <- 600 volcano_widget$height <- 500 sized_widget(volcano_widget) ``` ## Differentially Expressed Proteins Here we use the FDR threshold of `r cfg$processing_options$FDR_threshold` and a difference threshold of `r cfg$processing_options$diff_threshold` to select differentially expressed proteins. @tbl-nr-significant summarizes the number of significant calls. ```{r} #| label: tbl-nr-significant #| tbl-cap: "Number of proteins called significant versus not significant per contrast, applying the fold-change and FDR thresholds (total, significant, not significant)." #| eval: true library(dplyr) mx <- datax |> dplyr::mutate( passes = if (is_saint) { log2_EFCs > cfg$processing_options$diff_threshold & BFDR < cfg$processing_options$FDR_threshold } else { abs(diff) > cfg$processing_options$diff_threshold & FDR < cfg$processing_options$FDR_threshold }) x <- mx |> dplyr::group_by(.data[[contrast_col]]) |> summarize( n = n(), Significant = sum(passes, na.rm = TRUE), "Not Significant" = n() - sum(passes, na.rm = TRUE)) knitr::kable(x) ``` ```{r areThereSig} datax_signif <- dea$annotated_contrasts_signif showSignificant <- TRUE if (nrow(datax_signif) == 0) { showSignificant <- FALSE } ``` ```{r} #| results: asis #| eval: !expr get0("showSignificant", ifnotfound = FALSE) cat("The table shown in @fig-sig-prey lists all the significant proteins.") ``` ```{r} #| label: fig-sig-prey #| fig-cap: "Significant proteins obtained by applying the difference and FDR thresholds." #| fig-alt: "" #| eval: !expr get0("showSignificant", ifnotfound = FALSE) ctdata <- datax_signif |> tidyr::unite("ID", tidyselect::all_of(hkey), sep = "~") ctdata <- ctdata |> dplyr::select(tidyselect::all_of(table_cols)) |> as.data.frame() sig <- crosstalk::SharedData$new(ctdata, as.formula( "~ ID"), group = "BB") DT::datatable(sig, filter = "bottom", extensions = "Scroller", style = "auto", class = "compact", options = list(deferRender = TRUE, scrollY = 300, scrollX = 400, scroller = TRUE) ) |> DT::formatRound(columns = c(score_col, effect_col), digits = 3) ``` ```{r prepareHeatmap} #| eval: !expr get0("showSignificant", ifnotfound = FALSE) signif <- dea$lfq_data$get_copy() signif <- signif$get_subset(datax_signif) if (signif$hierarchy_counts()[2] > 30) { rownames = FALSE } else { rownames = TRUE } sigheat <- signif$get_Plotter()$heatmap(rownames = rownames) ``` ```{r makeText} #| eval: !expr get0("showSignificant", ifnotfound = FALSE) #| results: asis cat(paste( "Furthermore, @fig-sig-proteins shows a co-clustered heatmap", "of the row z-scored protein abundances of all significant calls.", "Proteins and samples with similar abundance profiles are grouped by", "unsupervised hierarchical clustering.")) ``` ```{r} #| label: fig-sig-proteins #| fig-cap: "Heatmap of the proteins passing the FDR and difference thresholds. Cell colour encodes the row z-score of the $\\log_2$ protein abundance (green: low, black: average, red: high; missing values in light gray). Proteins and samples are co-clustered by unsupervised hierarchical clustering (complete linkage, euclidean distance)." #| fig-alt: "" #| eval: !expr get0("showSignificant", ifnotfound = FALSE) sigheat ``` ```{r} #| label: fig-venn-sig #| fig-cap: "UpSet plot of significant proteins across contrasts. Bar heights give the number of proteins passing the FDR and fold-change thresholds in each contrast and in the contrast intersections." #| eval: !expr get0("showSignificant", ifnotfound = FALSE) xx <- split(ctdata[["ID"]], ctdata[[contrast_col]]) if (length(xx) > 1) { UpSetR::upset(UpSetR::fromList(xx)) } ``` # Additional Analysis The zip file contains an Excel file __DE_Groups_vs_Controls.xlsx__. All the figures can be recreated using the data in the excel file. The Excel file contains the following spreadsheets: - __annotation__ - the annotation of the samples in the experiment - __raw_abundances__ table with empirical protein abundances. - __normalized_abundances__ table with normalized protein abundances. - __raw_abundances_matrix__ A table where each column represents a sample and each row represents a protein and the cells store the empirical protein abundances. - __normalized_abundances_matrix__ A table where each column represents a sample and each row represents a protein and the cells store the empirical protein abundances. - __diff_exp_analysis__ A table with the results of the differential expression analysis. For each protein there is a row containing the estimated difference between the groups, the false discovery rate FDR, the 95% confidence interval, the posterior degrees of freedom. - __missing_information__ - spreadsheet containing information if a protein is present (1) or absent in a group (0). - __protein_variances__ - spreadsheet which for each protein shows the variance (var) or standard deviation (sd) within a group, the number of samples (n) and the number of observations (not_na) as well as the group average intensity (mean). The data can be used to perform functional enrichment analysis [@monti2019proteomics]. To compare the obtained results with known protein interactions we recommend the [string-db.org](https://string-db.org/) [@Szklarczyk2017], which is a curated database of protein-protein interaction networks for a large variety of organisms. To simplify the data upload to string-db we include text files containing the uniprot ids: - `ORA_background.txt` all proteins. - `ORA_<contrast_name>.txt` proteins accepted with the FDR and difference threshold. Other web applications allowing to run over representation analysis (ORA) [@monti2019proteomics] are: - [DAVID Bioinformatics Resource](https://david.ncifcrf.gov/home.jsp) - [WEB-based GEne SeT AnaLysis Toolkit](http://www.webgestalt.org) [@Wang2017] Furthermore, protein IDs sorted by t-statistic can then be subjected to gene set enrichment analysis (GSEA) [@subramanian2005gene]. To simplify running GSEA, we provide the file: - `GSEA_<contrast_name>.rnk` This file can be used with the webgestalt web application or used with the GSEA application from [gsea-msigdb](https://www.gsea-msigdb.org/) For questions and improvement suggestions, with respect to this report, please contact <protinf@fgcz.uzh.ch>. # Session Info ::: {.panel-tabset} ## Report provenance ```{r} #| label: tbl-report-provenance #| tbl-cap: "Compact report provenance, including the source input-data reference." knitr::kable(prolfquapp:::.report_provenance_table(report_provenance)) ``` ## R session info ```{r sessionInfo} pander::pander(sessionInfo()) ``` ::: # References {.unnumbered} ::: {#refs} ::: # Glossary - groups - different treatments, genotypes etc. - diff (difference) - it is the difference of the protein abundance estimate of two groups - FDR - false discovery rate --- *This report was generated from the Quarto template `Grp2Analysis_V2_R6.qmd` included in the `prolfquapp` R package (version `r packageVersion("prolfquapp")`).*