More Recent Comments

Monday, February 08, 2021

The 20th anniversary of the human genome sequence: 3. How many genes?

This week marks the 20th anniversary of the publication of the first drafts of the human genome sequence. Science choose to celebrate the achievement with a series of articles that had little to say about the scientific discoveries arising out of the sequencing project; one of the articles praised the opennesss of sequence data without mentioning that the journal had violated its own policy on openness by publishing the Celera sequence [The 20th anniversary of the human genome sequence: 1. Access to the data and the complicity of Science].

I've decided to post a few articles about the human genome beginning with one on finishing the sequence. In this post I'll summarize the latest data on the number of genes in the human genome.

The first drafts of the genome sequence predicted somewhere between 30,000 and 35,000 genes based largely on software predictions. In spite of what you might have read elsewhere, this number was very close to the predictions made by knowledgeable scientists dating back to the 1970s [False history and the number of genes].

We know more about protein-coding genes than noncoding genes. The number of predicted protein-coding genes dropped steadily for about 15 years following publication of the draft sequences but it is beginning to stabilize around 19,500 genes [How many protein-coding genes in the human genome?]. That number is likely to fall to about 19,000 in the future because there are hundreds of predicted protein-coding genes that are missing a protein There will never be an exact number because some people have more genes than others.

The number of noncoding genes is still up in the air. Here's a brief summary of the most likely numbers ...

  • unique small RNAs: Humans have genes for a number of unique small RNAs such as the RNA component of RNAse P and the 7SL RNA of signal recognition particle. There are about 30 of these genes.
  • ribosomal RNA genes: There are mutliple copies of the large ribosomal RNA operon and multiple copies of 5S RNA genes. A good estimate of the average number in a typical human genome is about 300.
  • transfer RNA (tRNA) genes: There are at least several hundred tRNA genes in a typical genome. There are also hundreds of pseudogenes.
  • small nuclear RNAs (snRNAs) genes: Some of the main spliceosomal RNAs (U1, U2, U4, U5 and U6) are produced from a single gene but in other cases there are mutliple copies. There are a number of extra spliceosomal RNA genes such as U4atac and U7. The total number of snRNA genes is about 20.
  • small nucleolar RNA (snoRNA) genes: These are genes involved in modifying ribosomal RNAs. There are more than 100 snoRNA genes.
  • microRNA (miRNA) genes: Nobody knows exactly how many miRNA genes there are in the human genome. The predictions range from 100-1000 but the algorithms for detecting these genes aren't very good. It's likely that most of the predicted genes are pseudogenes. A good estimate is 100 miRNA genes.
  • short interfering RNA (siRNA) genes: This is the same situation as with miRNA genes. The best guess is 100 siRNA genes.
  • PIWI-interacting RNA (piRNA) genes: There are several thousand predicted piRNA genes in our genome but it's almost certain that most of them are nonfunctional. It's safe to assume there are about 100 functional genes in this category.
  • long noncoding RNA (lncRNA) genes: This is an extemely heterogeneous category consisting of RNAs that are at least 1000 nucleotides in length with, in most cases, no substantial open reading frame. Some lncRNAs have a function but these are rare. It's likely that there are no more than 1000 lncRNA genes and the remaing transcripts are junk RNA.

The total number of noncoding RNA genes comes to less than 2000 but I usually feel quite generous in estimating this number so let's say that there are about 5,000. If we round up the total number of protein-coding genes to 20,000 then I'm estimating that there are no more than 25,000 genes in our genome.

Most of the sequence databases list more genes and the "extra" genes are mostly noncoding genes; for example, Ensembl estimates 23,997 noncding genes in the latest build (GRCh38.p13) [Human assembly and gene annotation]. About 17,000 of these genes are lncRNA genes but there's no evidence that these are functional genes. Until that evidence become available (I'm not holding my breath) we should stick with the best estimate of functional lncRNA genes [How many lncRNAs are functional?] [Functional RNAs?].

Image credit: The figure is from Palazzo and Lee (2015) - a must-read paper for those who are interested in following up on the number of noncoding genes.

Palazzo, A.F., and Lee, E.S. (2015) Non-coding RNA: what is functional and what is junk? Frontiers in Genetics, 6:2(1-11). [doi: 10.3389/fgene.2015.00002]


  1. You seem to have omitted tRNA genes, though tRNA's themselves are on your pie charts.

    1. I think there are 300-700 tRNA genes, the wrinkle being that many are actually pseudogenes ... so maybe closer to 300.

    2. Thanks. I forgot to add the tRNA genes to the list. Fixed it.

      Nobody seems to know how many functional tRNA genes there are in the human genome. The most frequently referenced papers are more than 20 years old and I couldn't find anything more up to date. I'm skeptical about numbers in the high hundreds since these genes are class III genes and processed pseudogenes are common.

    3. Hello John Harshman I would like to ask you a question since you know about Dinosaur phylogenetics. The baraminologist Todd Wood wrote an article in 2013 an analysis that "refutes" the article by Phil Senter on which the baraminology supports the evolution here is the article could you give me your opinion on that study sa = t & source = web & rct = j & url = https: //

    4. Incidentally, your link, if that's what it's intended to be, doesn't work at all.

    5. Hi Darwin!

      Since evolution has the status of a theory, so that “baraminology” is a disproved hypothesis, this is basically a pseudo-problem.

      Phil Senter has now shown by [i]reductio ad absurdum[i] using the example of the dinosaurs that “baraminology” of course does not apply here either. Todd Wood responded in a misguided way by “moving the goalposts”: He claims that Phil Senter should not have used Classical multidimensional scaling (CMDS) as statistical method, but distance correlation (DC), then the result would be correct.

      Remarkable is first of all the admission of Phil Senter that CMDS delivers the disadvantageous result for his position. CD increases the statistical resolution by means of more extensive data sets - thus, in principle, discontinuities can be constructed more easily. But if discontinuities are to be found, it means nothing else that more data are necessary to resolve these discontinuities. [b]Because for the obtained similiarity measure phylogeny is axiomatically presupposed![/b]

      So that you do not have to remain a creationist, here are the papers mentioned:



    6. John Harshman the article is by Todd Wood, the link you gave is to get the download

  2. You'll find a list of everything the HUGO Gene Nomeclature Committee has a gene name assigned to under / which of course doesn't make all of that real genes.