log2FC ↔ Fold Change Converter

Differential expression results report log2FC, but when presenting findings people typically say “upregulated X-fold.” The relationship is:

FC = 2 ^ log2FC          log2FC = log2(FC)

Why log2? Raw fold changes are asymmetric: 2-fold up is 2, 2-fold down is 0.5. All downregulated genes are squeezed between 0 and 1 while all upregulated genes spread from 1 to infinity — awkward for both plotting and statistics. After log2 transformation, 2-fold up is +1 and 2-fold down is −1, symmetric around 0.

Common threshold reference:

One important point: The log2FC threshold and statistical significance (p / q value) are two independent filters that must both be applied. Filtering on fold change alone lets random noise in low-expression genes produce many spurious high-fold-change hits; filtering on q value alone lets tiny changes in highly expressed genes pass as significant without any biological relevance.

High fold changes in low-expression genes are artifacts

log2FC measures the ratio, not the absolute magnitude. The following scenarios give identical numbers:

Both rows give the same log2FC, but their reliability is worlds apart. The first differs by only 6 reads — a small sequencing depth fluctuation could shift it to 3 → 6 or 1 → 9. The second differs by 600 reads and is far more stable.

This is why ranking genes by |log2FC| alone almost always leads to trouble: the top hits are typically low-expression noise. Statistical significance must also be applied; significance tests (DESeq2, edgeR) explicitly model count dispersion, so p-values for low-expression genes naturally stay high.

Why software log2FC differs from manual calculation

Tools like DESeq2 apply shrinkage estimation by default: log2FC values for low-expression, high-variance genes are pulled toward zero; values for highly expressed, well-supported genes are barely affected. This prevents noisy genes from dominating ranked lists.

As a result, fold changes back-calculated from shrunken log2FC will be more conservative than those computed directly from raw counts — this is by design, not an error. This tool performs pure mathematical conversion: whatever you input is what gets converted.

How to choose thresholds

There is no universal answer, but two practical principles apply:

Define biological relevance first, then set the threshold. A 1.5-fold change in a transcription factor may already have substantial downstream effects, while a 2-fold change in a structural protein may mean nothing.

Fix the threshold before looking at the data. Trying several thresholds and selecting the one with the best-looking result is effectively an uncorrected multiple comparison.