Plot every gene’s log2 fold change against its mean expression and see significance, noise, and normalization at once.
Upload DESeq2, edgeR, limma, or proteomics results. SciFigure detects whether the mean is on a count or log scale, colors genes by your adjusted p-values, keeps genes with no adjusted p-value apart, checks the trend line against zero, and draws a clean MA plot with the top genes labeled.
Upload differential expression results and describe the plot.
Try sample data
Costs 5 credits. Significance calls, counts, and trend values are computed from your uploaded data, then drawn as a PNG.
Preview
Aspect ratio: 16:9
These previews were drawn from the sample datasets. Load a sample or upload your own file to generate an MA plot.
Each example was drawn from a sample dataset: DESeq2 RNA-seq results, a limma microarray analysis, edgeR results with a fold-change threshold, a two-color array before normalization, a proteomics comparison, and a highlighted gene set.
Prompt
Based on the uploaded deseq2_results.csv file, create an MA plot with the mean of normalized counts on a log10 x-axis and log2 fold change on the y-axis. Color genes with padj < 0.05 red (up) and blue (down), keep the rest light gray, draw genes with no adjusted p-value as hollow gray points, and label the top 10 genes.
Prompt
Based on the uploaded limma_toptable.csv file, create an MA plot of logFC against AveExpr (already log2, so a linear x-axis), adj.P.Val < 0.05 genes in dark orange and purple, and a running median line to show the trend stays at zero. No labels.
Prompt
Based on the uploaded edger_results.csv file, create an MA plot of logFC against logCPM with dashed lines at ±1, highlighting genes with FDR < 0.05 and |logFC| ≥ 1 in red and blue. No labels.
Prompt
Based on the uploaded two_color_array_MA.csv file, create an MA plot of raw M against A for one two-color array, small gray points, a red running median line, and a dashed line at zero, to show the intensity-dependent dye bias that loess normalization removes.
Prompt
Based on the uploaded proteomics_ratios.csv file, create an MA plot of log2 ratio against mean log2 intensity, proteins with q < 0.05 in red and blue, and labels for the heat shock proteins HSPA1A, HSPA1B, DNAJB1, HSPH1, HSPB1 and BAG3.
Prompt
Based on the uploaded infection_deseq2.csv file, create an MA plot with baseMean on a log10 x-axis, significant genes in muted colors, and the interferon-stimulated genes ISG15, IFIT1, IFIT3, MX1, OAS1, IFI6, RSAD2 and IFI44L labeled and outlined in black.
Add a CSV, TSV, or Excel export with a mean expression column (baseMean, AveExpr, logCPM), a log2 fold change column, and optionally an adjusted p-value column (padj, FDR, adj.P.Val).
Set the significance cutoff, an optional fold-change threshold, colors, a y-axis limit, and which genes to label, by rank or by name.
SciFigure computes every significance call and count, draws the MA plot, and lets you download a PNG for papers, posters, and slides.
An MA plot is a scatter plot of differential expression results with one point per gene: the log2 fold change (M) on the y-axis against the average expression (A) on the x-axis. It was introduced for two-color microarrays, where M = log2(R/G) is the log ratio of the red and green channel intensities and A = ½·log2(R·G) is their average log intensity. In RNA-seq the same layout plots each gene’s log2 fold change against its mean of normalized counts, drawn on a log scale, as DESeq2’s plotMA does. Most genes do not change, so they form a horizontal band around M = 0, and the significant genes stand out above and below it. limma and edgeR call the same plot a mean-difference (MD) plot, and statistically it is a Bland-Altman plot on the log scale.
Both plots show the same differential expression results with log2 fold change as the effect size. A volcano plot puts fold change on the x-axis and −log10 p-value on the y-axis, so it answers "which genes changed the most with the strongest evidence". An MA plot puts mean expression on the x-axis, so it answers "how does the effect depend on how well the gene is measured", which makes it the better quality-control plot: normalization bias, low-count noise, and the effect of fold-change shrinkage are visible in an MA plot and invisible in a volcano plot. Many papers show both, often side by side.
The trend of M across A should be flat at zero, because most genes are not expected to change. In two-color microarrays, dye bias makes the trend curve with intensity, and loess normalization along A, described by Yang and colleagues in 2002, removes it. In RNA-seq, a cloud shifted away from zero usually means the size factors or library-size normalization did not account for composition, for example when a few very abundant transcripts change. This generator computes the median M in ten equal-count bins along the x-axis and passes it to the plot, so you can ask for a trend line and see at a glance whether your data are centered.
The generator reads the mean expression, log2 fold change, and optional adjusted p-value for every gene. It decides whether the mean is on a count scale (DESeq2 baseMean, drawn on a log10 axis) or already on a log2 scale (limma AveExpr, edgeR logCPM, drawn on a linear axis), and skips genes with a zero mean on a log axis. Genes are called significant only when your adjusted p-value is below the cutoff (0.05 by default; DESeq2’s plotMA uses 0.1) and, if you set one, |log2 fold change| is at least the threshold. Genes with no adjusted p-value are counted and drawn apart. It counts up- and down-regulated genes, computes the binned trend, caps the y-axis with edge triangles when asked, and picks the top genes to label by adjusted p-value. It never estimates p-values: without an adjusted p-value column, genes are colored by fold change only and the figure says so.
Two views of the same results table. Use the MA plot to judge the data, the volcano plot to pick the hits.
| MA plot | Volcano plot | |
|---|---|---|
| X-axis | Mean expression (A) | log2 fold change |
| Y-axis | log2 fold change (M) | −log10 p-value or adjusted p-value |
| Shows best | How effects depend on expression level | Effect size against strength of evidence |
| Quality control | Normalization bias, low-count noise, effect of shrinkage | Little |
| Significance shown by | Point color | Height and color |
| Typical use | Checking a differential expression analysis | Presenting and labeling top hits |
These are the problems that make an MA plot misleading.
MistakePlotting DESeq2 baseMean on a linear x-axis, so 95% of genes crowd into the left edge.
CorrectDraw counts on a log10 axis; values already on a log2 scale (AveExpr, logCPM) stay linear.
MistakeColoring genes by unadjusted p-values.
CorrectUse the adjusted p-value (FDR) from the same test, and state the cutoff.
MistakeDrawing genes with padj = NA as non-significant without explanation.
CorrectKeep them apart (hollow or lighter) and say they were filtered for low counts or outliers.
MistakeReading large fold changes at low expression as strong effects.
CorrectThose estimates are noisy. Plot shrunken fold changes (apeglm or ashr) for display and ranking.
MistakeIgnoring a cloud that bends or sits off zero.
CorrectTreat it as a normalization warning and check size factors or apply loess normalization before testing.
MistakeCutting the y-axis and silently dropping the most extreme genes.
CorrectShow genes beyond the limit as triangles on the edge so readers know they exist.
An MA plot is a scatter plot of differential expression results that puts each gene’s log2 fold change (M) on the y-axis and its average expression (A) on the x-axis. It shows which genes changed, how that depends on expression level, and whether the data are well normalized.
They come from two-color microarrays: M is the log ratio of the two channel intensities, M = log2(R/G), and A is their average log intensity, A = ½·log2(R·G). Think of M as "minus" (a difference of logs) and A as "add" (an average of logs). In RNA-seq, M is the log2 fold change and A is the mean of normalized counts.
An MA plot shows log2 fold change against mean expression; a volcano plot shows log2 fold change against −log10 p-value. The MA plot is better for checking normalization and low-count noise, and the volcano plot is better for showing which genes have both large effects and strong evidence.
A dense band of points centered on zero across the full x-range, wider at low expression and narrower at high expression, with significant genes spread above and below. A trend line through the medians should stay flat at zero.
With few reads, small random differences produce large ratios, so fold-change estimates for low-count genes are noisy. That is why the plot fans out to the left, and why those genes are rarely significant.
For display and ranking, usually yes. DESeq2’s lfcShrink (apeglm or ashr) pulls noisy low-count fold changes toward zero and leaves well-supported effects in place. Say which fold changes you plotted in the caption.
In DESeq2, padj is set to NA for genes removed by independent filtering because of low mean counts, and both p-value and padj are NA for genes with count outliers (Cook’s distance) or all-zero counts. This generator draws them apart and counts them.
DESeq2: baseMean, log2FoldChange, padj. edgeR: logCPM, logFC, FDR. limma: AveExpr, logFC, adj.P.Val. Add a gene ID or symbol column for labels. The generator detects these headers automatically.
The generator uses adjusted p < 0.05 by default. DESeq2’s plotMA colors genes with padj < 0.1, its default alpha. Write "padj < 0.1" or "FDR 0.01" in your instructions to change it, and add "fold change threshold 1" to require |log2 fold change| ≥ 1 as well.
No. Significance comes only from the adjusted p-values in your file. Without them, genes are colored by fold change alone and the plot says that no significance test was applied.
Yes, in layout. limma and edgeR call it a mean-difference (MD) plot, and plotting the difference of two log-scale measurements against their mean is the Bland-Altman method applied to log-transformed data.
Yes. Write "label the top 15" or name the genes, for example "label FKBP5, PER1 and KLF15". Named genes that exist in your file are labeled; others are ignored.
Generate MA plots, volcano plots, heatmaps, and other genomics figures from your own data with SciFigure AI.
