A bit of a historical question on a number, 30 times coverage, that's become so familiar in the field: why do we sequence the human genome at 30x coverage?

My question has two parts:

  • Who came up with the 30x value and why?
  • Does the value need to be updated to reflect today's state-of-the-art?

In summary, if the 30x value is a number that was based on the old Solexa GAIIx 2x35bp reads and error rates, and the current standard Illumina sequencing is 2x150bp, does the 30x value need updating?

enter image description here


The earliest mention of the 30x paradigm I could find is in the original Illumina whole-genome sequencing paper: Bentley, 2008. Specifically, in Figure 5, they show that most SNPs have been found, and that there are few uncovered/uncalled bases by the time you reach 30x: 30xSequencingDepth

These days, 30x is still a common standard, but large-scale germline sequencing projects are often pushing down closer to 25x and finding it adequate. Every group doing this seriously has done power calculations based on specifics of their machines and prep (things like error rates and read lengths matter!).

Cancer genomics is going in the other direction. When you have to contend with purity, ploidy, and subclonal populations, much more coverage than 30x is needed. Our group showed in this 2015 paper that even 300x whole-genome coverage of a tumor was likely missing real rare variants in a tumor.

On the whole, the sequence coverage you need really depends on what questions you're asking, and I'd recommend that anyone designing a sequencing experiment consult with both a sequencing expert and a statistician beforehand (and it's even better if those are the same person!)

| improve this answer | |
  • $\begingroup$ This is as far as I could trace back, but would love to know if anyone can find an earlier mention of 30x as a standard depth! $\endgroup$ – chrisamiller Aug 4 '17 at 17:50
  • $\begingroup$ That is likely the earliest. Previous technologies could not reach that depth at a reasonable cost. $\endgroup$ – Bioathlete Aug 4 '17 at 18:03

Solexa Inc. sequenced NA12878 chrX to ~30x in early 2007, which later became part of Bentley (2008). This, I believe, was the first time that 30x showed up. I don't recall they had a particular reason for that. Figure 5 in the published paper was just aftermath. It does not really explain why not 25x or 35x, given that the curves between 25x and 35x in that figure are about linear.

In the abstract of Ajay et al (2011), the authors argued "the current recommendation of ~30x coverage is not adequate". Nonetheless, the discussion section seems to suggest 50–60x would be necessary with GAIIx, but 35x was adequate with HiSeq2000 plus better recent chemistry. Overall, this paper provides a more thorough analysis. The data quality at that time is also closer to data we produce today.

The required coverage is largely determined by two factors: read placement bias (e.g. GC bias) and base/mapping error rate. While GC bias has been reduced with the PCR-free protocol, base error rate has been going the downward since HiSeq2500. I guess 30x coverage would be necessary if you want to achieve the sensitivity with the older 30x data. Illumina, as a sequencing service provider, and our sequencing facility still insist on the 30x threshold.

| improve this answer | |
  • 2
    $\begingroup$ Interestingly enough we also found in Sudbery et al, 2009 (genomebiology.biomedcentral.com/articles/10.1186/…), that the number of continuous regions without a break in coverage was more or less linear in the number of sequenced reads between 25x and 60x when sequencing the mouse genome. $\endgroup$ – Ian Sudbery Aug 7 '17 at 8:46

30 times coverage is not unique to this problem, but number 30 has its empirical role in statistics:

In statistical analysis, the rule of three states that if a certain event did not occur in a sample with n subjects, the interval from 0 to 3/n is a 95% confidence interval for the rate of occurrences in the population. When n is greater than 30, this is a good approximation to results from more sensitive tests.

source: Wikipedia: Rule of three (statistics)

Similarly, you can search for related questions like this one:

In line with this, I have seen data processing in other disciplines which required n ≥ 30 for sufficient reliability of results.

| improve this answer | |
  • $\begingroup$ In this type of community, I would expect some opponenture under the answer but I am getting only downvotes so far (curent score: +6/-3). $\endgroup$ – miroxlav Apr 18 '18 at 16:27

The point I always miss in the discussion about coverage is, that no one tells how it was calculated. Were duplicates removed? How do you count overlapping paired-end reads? As 2 or 1? Just to point out two things that have influence on the coverage.

| improve this answer | |
  • 1
    $\begingroup$ Anyone who counts overlapping paired-end reads as 2 is doing it wrong. There's never an excuse for that. $\endgroup$ – Devon Ryan Aug 10 '17 at 18:13
  • $\begingroup$ @DevonRyan why? Overlapping parts of pair-end reads are two independent technical replicates of the same region. What is the difference to overlap of two reads that are not paired, should they also be counted as one? Of course if you merge overlapping reads into longer sequences before an assembly, then you got to count it once, but otherwise I do not see a reason why they should be counted as one... $\endgroup$ – Kamil S Jaron Aug 10 '17 at 19:51
  • $\begingroup$ @KamilSJaron They represent one sequenced fragment, counting them otherwise is lying. $\endgroup$ – Devon Ryan Aug 10 '17 at 19:53
  • $\begingroup$ @DevonRyan Same fragment, but different sequencing and it is sequencing coverage, not a fragment coverage. I do not get your point. $\endgroup$ – Kamil S Jaron Aug 10 '17 at 19:56
  • $\begingroup$ @KamilSJaron You're being overly literal. The point of the metric is to assess how much data you have per position. Overlapping PE reads don't represent different data points, that's why they're treated as a single unit in variant and peak calling. $\endgroup$ – Devon Ryan Aug 11 '17 at 6:56

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.