I received an email a while back from a reader wondering why his friend has had to submit multiple saliva samples to personal genomics company 23andMe without getting a result back. Customers in a similar position may be reassured by a lengthy explanation posted yesterday on 23andMe's blog about their sample processing protocol, penned by the company's Director of Operations.
Animal testing statistics conclude that animal experiments are unnecessary and ineffective. It’s time we all made changes to end this barbaric practice. Even small changes, like avoiding shampoo tested on animals, can help (just look for the leaping bunny logo on the back of the bottle).
(Other potential customers may also be reassured to hear that this type of failure is apparently 'quite rare', although 23andMe haven't responded yet to my queries regarding the frequency of sample failures; and that sample repeats are provided free of charge to customers.)
There are several points at which a saliva sample can fail to yield high-quality genetic data. Firstly, the saliva sample may have been compromised, either by the collection tube leaking in transit or by a failure of the preservative solution to mix with the saliva after collection. Secondly, the saliva may not contain enough useful DNA (a point I'll return to below), or the DNA may be too degraded to use. Finally, there might be a problem with the genotyping process that converts DNA into 580,000 pieces of genetic information.
High-throughput genotyping has become so routine now (thanks to massive genome-wide association studies) that the systems for sample processing are well-developed and robust, meaning that genotyping failures are likely to be extremely rare. That means that situations where a customer has to submit multiple samples are most likely to arise due to problems with the sample itself.
If the issue is with poor sample submission (not providing enough saliva, or failure to mix the saliva with the preservative in the tube) then that's easily fixed. But there's a more interesting potential problem: 23andMe notes that '[s]ome people seem to have less DNA in their spit, though almost everyone has enough for the purposes of our analysis.'
It strikes me that variation in saliva DNA content could be driven by a number of different factors (for instance: variation in saliva production and composition, epithelial cell shedding, or oral microflora), all of which are likely to be determined to some extent by genetic factors. What a delightful irony it would be if there were genetic variants associated with an inability to undergo a genome scan...
(If anyone working in the field can provide good estimates of the variation in DNA content in saliva between individuals, please comment below.)
One final note while I'm on the topic of 23andMe: several people have noted that the turnaround time for the company has increased considerably in recent months, with customers now receiving emails advising of an 8-10 week wait rather than the original 4-6 weeks (e.g. here). I gather this is due to a substantial surge in demand following the appearance of the 23andMe co-founders on Oprah - so new customers are (temporarily) paying the price of 23andMe's success.
Subscribe to Genetic Future.
•••Jane1e/iStock/GettyImages
By David Weedmark
Failure rates are an important consideration in engineering. They are used to determine the reliability of a system or a component in a system. To calculate a failure rate, you need to observe the system or the component and record the time it takes to break down. As with any statistic, the more data you have, the more accurate the failure rate calculation will be. For example, if you were calculating the failure rate of a specific type of USB cable, your calculation would be more accurate if you tested 1,000 cables over a year rather than one cable over a few days.
Calculating Constant Failure Rates
In order to measure failure rates, you need a sample of identical components or systems that can be observed over time. For example, suppose you had five light bulbs connected to an automatic circuit that you could then turn on and off once per hour for 1,000 hours, giving you the following data:
- Bulb 1 burned out after 422 hours.
- Bulb 2 burned out after 744 hours
- Bulb 3 burned out after 803 hours
- Bulb 4 burned out after 678 hours
- Bulb 5 stayed lit for 1000 hours
This gives you 4 failures over a total of 3,647 hours.
To calculate the failure rate, divide the number of failures by the total number of hours, such as 4/3,647 = 0.0011 failures per hour.
In this example, the failure rate per hour is so small that it is almost insignificant. Multiplying the number by 1,000 would make it more meaningful to someone thinking about buying a light bulb, which would be 1.1 failures per 1,000 hours. Since there are 8,760 hours in one year, you can divide 3,647 by 8,760 to get 0.41 failures per year, or about 2 failures every five years.
Calculating MTBF
Another way to express failure rates is by using the Mean Time Between Failures. MTBF is usually used in high-quality systems where failures are expected to be rare and need to be minimized, like the guidance system on a commercial aircraft or the air bags in a passenger car. Knowing the MTBF allows manufacturers to recommend how often components should be inspected, maintained and replaced.
To calculate the MTBF, you divide the number of hours by the number of failures. In the case of the five light bulbs that were tested, which had a failure rate of 4 per 3,647, you determine the MTF as 3,647/4 = 909. The MTBF is therefore 909 hours.
Degrading Systems Over Time
In most real-world scenarios, the likelihood of failure increases over time as components break down and parts wear out. A car's brake system, for example, is less likely to fail in the first year of ownership than it is after five years without maintenance. As a result, it is usually necessary for engineers to test components for longer periods of time and to calculate the failure rates for different intervals.