We will soon have a pre-print describing the new 2019 automated data analysis workflow. The code-base for the build can be found here.

Each gene gets its own box. The y-axis is length scaled TPM (log2 transformed). The x axis is samples, colored by tissue. The right panels [9, 10] are tables with external links to gene info [9] and the absolute TPM values and the rank of the gene in the particular tissue (lower is more highly expressed) [10].

To give a rough sense of how highly expressed the gene is in the tissue, the decile of expression is given in [10]; 10 is the highest decile of expression and 1 is the lowest.

This is a very simple differential expression test. When you click this radio button [1] the view changes, where the expression is being shown *relative* to the reference tissues [2]. By default it is all of the tissues; in the screenshot Retina - Organoid and Retina - Stem Cell Line are the baseline expression samples. The data table on the right [3] will then display log2 fold change, average expression, and the p-value (from a t-test) of the differential expression.

This produces a 2D visualization, with each gene as a row and each tissue as a column. More yellow is more expressed. It is a efficient way to display the expression of many genes and tissues.

If you use the Pan-Tissue Boxplot feature a lot, you may find it frustrating to have to input in your favorite genes and tissues. We have added the ability to use a custom url to load in the genes and tissues of your choice. Previously you had to build this link youself - but now there's a handy button [7] you can click that will re-create the parameters. One downside is that the web app is continually using the URL dataset, which makes it impossible for you to change it. You can simply reload the web page with the custom bits.

This is short for differential expression. We have pre-calculated 55+ differential expression tests. All eye tissue - origin pairs were compared to each other. We also have a synthetic human body set, made up of equal numbers of GTEx tissues (see manuscript, above, for more details). The word cloud displayed shows as many as the top 75 terms used in enriched GO terms in the selected comparison. The table data shows the actual GO terms. You can search for the comparison of your choice.

These are the values taken from the limma differential expression topTable() summary table. The following has been taken from the limma manual and edited to match parameters we used (https://www.bioconductor.org/packages/devel/bioc/vignettes/limma/inst/doc/usersguide.pdf):

A number of summary statistics are presented by topTable() for the top genes and the selected contrast. The logFC column gives the value of the contrast. Usually this represents a log2-fold change between two or more experimental conditions although sometimes it represents a log2-expression level. The AveExpr column gives the average log2-expression level for that gene across all the arrays and channels in the experiment. Column t is the moderated t-statistic. Column P.Value is the associated p-value and adj.P.Value is the p-value adjusted for multiple testing (False Discovery Rate corrected).

The B-statistic (lods or B) is the log-odds that the gene is differentially expressed. Suppose for example that B = 1.5. The odds of differential expression is exp(1.5)=4.48, i.e, about four and a half to one. The probability that the gene is differentially expressed is 4.48/(1+4.48)=0.82, i.e., the probability is about 82% that this gene is differentially expressed. A B-statistic of zero corresponds to a 50-50 chance that the gene is differentially expressed. The B-statistic is automatically adjusted for multiple testing by assuming that 1% of the genes, or some other percentage specified by the user in the call to eBayes(), are expected to be differentially expressed. The p-values and B-statistics will normally rank genes in the same order. In fact, if the data contains no missing values or quality weights, then the order will be precisely the same.

The Macosko data is a single-cell (~45,000) retina RNA-seq mouse P14 C57BL/6 dataset from Mackosko and McCarroll's field defining paper. The cluster / cell type assignments are taken from here. The Clark data is a 100,000 cell plus mouse retina RNA-seq time series dataset. Their pre-publication manuscript is on bioRxiv. Data was pulled from here.

To efficiently display a huge amount of information, expression across many individual cells is averaged by cell type, (if available) age, and gene. You can select the Macosko or Clark dataset [1], then one gene [2] to plot. The gene expression is displayed as a heatmap, with each row being a retina cell type (derived by the respective authors) and each column [4] is a time point, arranged from youngest to oldest. More yellow is higher expression [5].

You can add the rank of expression (or rank of percentage of cells with detectable expression of selected gene) with this radio

Each data point is a single cell from the Macosko and McCarroll or the Clark and Blackshaw [2]. Dimensionality reduction with done with the t-SNE (Macosko) or UMAP (Clark) algorithm. Cluster assignments were taken from the respective papers. While we did the t-SNE on the Macosko data, the Clark authors provided the UMAP coordinates. The Clark dataset was generated across multiple time-points during development and thus, you can select time points of interest [4]. Only one gene can selected at a time [3], as it is very computationally expensive to plot many points. Points (cells) expressing the gene of interest are plotted in darker color [arrow]. Hovering over each point [6] will show the metadata for the cell.

We imagine the most common use is to search for your gene of interest (GOI). Simply type your GOI into the search box [1]. If it is not in the network, then the name will not appear. After selecting the GOI, the network will reload to display the module the gene is in, as well as several of the most correlated partners. You can adjust the number of displayed correlated genes by changing the K-nearest genes panel [2]. Hovering over a gene name in the network will display GO terms for the gene [3].

Unfortunately, we have no way of knowing this. Since the network algorithms use correlations, the gene to gene interactions have no directional information.

The count plot [1] simply shows the number of genes in each module. A gene can only be in one module. The pair-wise gene connection strength [2] shows the strongest gene partners for the selected gene. If a module search is selected, then this table shows all gene to gene edge connection strengths (higher is stronger) in the module. The gene connectivity table [3] shows the kWithin metric for each gene in the module, which denotes how connected (and important) the gene is across the module. The GO term table [4] shows the significant GO terms for the genes in the module. This allows you to get a sense of the function of the module.

The edge table allows you to search for a gene and it returns all significant (edge length > 0.01) correlated genes ACROSS the entire network. Using the 'Connections to show' radio button, you can control whether only extramodular or intramodular (or both) connections are included in the table.

This data table set shows the data used to make the Mouse Single Cell Retina Expression plots in Gene Expression.