Identify housekeeping genes

1 Methods overview

In brief, housekeeping genes are those with higher expression levels, low variances, and ubiquitous expression profiles across samples and species.

In paper “What are housekeeping genes” by Chintan J. Joshi, housekeeping genes are defined as those with the following four properties:

1. Higher expression stability

2. Cellular maintenance

3. Essentiality

4. Conservation

2 Method 1

From “A Comprehensive Mouse Transcriptomic BodyMap across 17 Tissues by RNA-seq” by Bin Li.

Criteria for identification of housekeeping genes:

1. Highly expressed in all biological samples (\(FPKM > 1\));

2. Low variance across tissues: std(log2(FPKM)) < 1;

3. No logarithmic expression value differed from the averaged log2(FPKM) value by a factor of two (i.e. fourfold) or more.

Criteria for identification of reference genes:

1. \(FPKM > 50\) across all biological samples;

2. std(log2(FPKM)) < 0.5 over tissues;

3. No logarithmic expression value differed from the averaged log2(FPKM) value by a factor of one (i.e. twofold) or more.

3 Method 2

From “Housekeeping protein‑coding genes interrogated with tissue and individual variations” by Kuo‑FengTung.

Gini coefficient of inequality (Gini index):

1. \(TPM > 0.05\);

2. \(\text{Gini index} < 0.2\).

4 Method 3

From “The evolution of gene expression levels in mammalian organs” by David Brawand.

Pipeline used to pick housekeeping genes and normalize expression levels across species:

1. Convert TPM to \(log2(TPM+1)\);

2. Retrieve and only keep one-to-one orthologous genes across all species with confidence equal to \(1\) from Ensembl BioMart;

3. Sort orthologs based on TPMs in descending order and represent each gene by its TPM rank in each sample;

4. Calculate the standard deviation and median of each ortholog based on its ranks across samples;

5. Keep orthologs the medians of which are within \(0.25 \times \text{the number of orthologs} \sim 0.75 \times \text{the number of orthologs}\) (discarding those orthologs with expression levels extremely high or extremely low across samples);

6. Retain the \(1000\) orthologs with the lower variances (standard deviations);

7. Calculate the medians of those \(1000\) orthologs’ TPMs in each sample;

8. Calculate the scaling factor of each sample: the scaling factor of sample A = TPM median of sample A / median(TPM median of all samples);

9. For each sample, scaled TPM = TPM / scaling factor.

Note: be aware of the fact that the expression difference of each same/homologous gene among species and the difference among batches are different.

5 Reference datasets

Human housekeeping genes: https://www.gsea-msigdb.org/gsea/msigdb/cards/HOUNKPE_HOUSEKEEPING_GENES.html.

Mouse housekeeping genes: https://www.gsea-msigdb.org/gsea/msigdb/mouse/geneset/HOUNKPE_HOUSEKEEPING_GENES.html.