It is usually thought that we can confidently say that if our genotyping results say that we carry a certain genetic variant, that we actually do carry that variant. However, why does this not mean that we can be confident about the prediction about disease risks?

There are many risks and benefits associated with population screening, but it turns out that the results of screening tests, whether they be genetic or otherwise, might not actually provide us with a definitive diagnosis. 

Usually, an individual’s test results will categorize them as being either high or low risk for getting a disease. Here, we will explain the ways in which we actually assess the predictive ability of a diagnostic test and why the results that we receive from these tests are simply estimations and not guarantees. 

In order to get an idea of the validity of a screening test, it is important to look at the problem from the opposite perspective. Before you look at how probable that it might be for someone to have a particular genotype, you should look at the probability of someone developing it if they have the specific genotype. 

If there were a population of 100 people, we know from epidemiological studies that the prevalence of a condition, for this case we will use Madeupitis as an example, in this population is 5%. This means that 5 of the 100 people will develop Maudeupitis, and the remaining 95 people will not.

We can summarise the screening results in the diagram pictured below, and the people colored in blue will develop the disease, and the people colored in green will not develop the disease. The red dot will indicate a positive test result.

Through this test, we can identify that 42 people are likely to develop Madeupitis, and from these people, 4 of them will be true positives, and the rest of the people will be false positives. The positive Prediction value is the number of people with a positive test result that actually go on to develop the condition.

In this case, it is 0.095 of the test population. The test will also categorize 58 people as being unlikely to develop Madeupitis, and one of these will be a false positive. The Negative Prediction value is the number of people with a negative test result that do not develop the condition, which, in this case, is 0.983.

This tells us that the test isn’t accurately identifying cases, and if you are given a positive result, there is no confidence that you will go on to develop it. Essentially, it is not an accurate prediction.

However, if you are given a negative result, you are more than likely not going to develop the condition, and this can be said with confidence. This is problematic for those that are given a positive result, which may, in fact, be false.

This does not mean that a screening test cannot be useful, but it does mean that the accuracy cannot be guaranteed with more complex conditions that have not yet been thoroughly explored. It is highly likely that the predictive accuracy of these tests will continue over time through research and additional testing and screening.

I agreed to make my 23andMe genotyping results available publicly as part of GNZ because I knew that the results were slightly dull, and I’m not majorly at high or low risk for any diseases.

I was also very unsurprised to find out that I have blue eyes and that I was identified to be most likely descended from Northern European Ancestry.

A few hours after releasing the data, I was directed to Dienekes Pontikos’ post, where they wrote about the results of running all our data through his ancestry prediction program.

Most people were predicted to be of Northwestern European descent, and I was given an estimate of 20% Ashkenazi Jewish ancestry. When asked about this, I had no answers to give and instead decided to do my own research. 

What Does Dienekes’ Program Do?

This program is based on a paper that explores the differences in allele frequencies in European Americans.

In the study, the authors identified two main components of variation in ancestry that corresponded to three groups: Mostly northwest European descent, southeast European descent, and Ashkenazi Jewish descent. 

They created a list of 300 genetic markers that gave information about the ancestry in the sample, which was made publicly available.

The program uses those allele frequencies at the level of markers in the 23andMe platform to infer the proportional membership of an individual from each group.

If someone has the genotype CC at an SNP, and the C allele has 20% frequency in northwestern Europe and 60% frequency in Ashkenazi, the person is more likely to be Ashkenazi. This method put me at 20% Ashkenazi ancestry, but with not enough data to make this a confident statement.

The question left to be answered is ‘what does this mean?’ It doesn’t mean that I have 20% Ashkenazi ancestry; it means that I carry the alleles that are rare in Europe and more common in the Ashkenazi Jewish population.

However, this is based on the assumption that my ancestry only goes back to these three locations.

Knowing that I had a grandparent of Italian descent, I tested for southern European ancestry. It is also commonly known that southern European populations are similar to the Ashkenazi populations in relation to genetics.

Visualizing Population Structure with Genomes Unzipped Data

In order to explore how the GNZ participants relate to European populations, I was able to combine multiple data sources. These were from the Human Genome Diversity Panel, the 12 GNZ individuals, and a set of Ashkenazi Jewish individuals.

They were all genotyped on Illumina arrays. To begin to look at the relationships between these individuals, I used the smartpca program to look at the average genetic relationships between the populations.

In the below plot, each point is an individual on the most important axis of genetic variation in this sample. The Ashkenazi population is blue, and the GNZ individuals are black.

Dan is with the Ashkenazi population, Vincent with the French, and me with the French on component 1 and the Italians on component 2.

Next are the additional components of variation with the second versus third components, followed by the third versus the fourth.

The fourth axis of variation separates the Ashkenazi population from the rest of Europe.

Conclusion

The analysis suggests that I do not have any Ashkenazi Jewish ancestry, and the results were actually hidden south European ancestry.

I have satisfied my curiosity surrounding my knowledge of family history, and it seems I have a small amount of south European ancestry mixed with a large amount of northwest European ancestry.

Much of the hype surrounded the focal point of the above image, a MinION. No, we don’t mean those banana loving yellow fools, but in fact a minuscule, disposable USB-key sequencer, capable of sequencing a whopping one billion base pairs of DNA from the comfort of your own lap.

Though it was initially a pricey purchase at between $500 and $900, its applications were seemingly endless. Being able to take some biological matter, combine it with a couple of chemicals, and then read its DNA was considered revolutionary. By no means affordable, it was the first tentative step into DNA experimentation at home. 

A second, slightly less exciting but equally as intriguing announcement, was that of the GridION sequencing machine.

Built to read a great deal of DNA to scale up at humongous sequencing centers, it was able to read long strings of data that had previously been unthinkable (at least in a way as reliable as the human brain!) 

What implications did these products have on those with a personal interest in genomics?

Simplified Sequencing

According to their estimations, one GridION machine, also referred to as a node, could read sequence as fast as 600 million base pairs an hour, the equivalent of 14GPB per day, and read high coverage (30x) human genome in less than six days.

Stacking up four nodes, you could therefore produce sequences working at a similar rate to Illumina’s high-end HiSeq 2500 machine.

At around $2200-$3600 for each genome, the machine was considered a new competitor in terms of throughput, data quality, and expenses, which at the time was a pretty sizeable achievement in comparison to the major sequencing companies from a smaller producer.

What was particularly clever about Oxford Nanopore’s inventions was that they avoided some of the difficulties traditionally found in DNA sequencing; each sequencer was tiny and stacked efficiently, with 20 units equating to the size of a traditional HiSeq at only a fifth of the footprint per genome.

Power consumption was also considerably reduced, and preparing DNA for sequencing takes moments as opposed to days of processing and much more expensive equipment. When people examine this company retrospectively, they will be able to clearly recognize their contribution to the simplification of genome sequencing.

This wasn’t just about reducing the cost - though they did, by a staggering 20% - but making at-home, outside the lab sequencing a possibility altogether. Nobody thinks about the price decrease in recent years, but everyone certainly remembers when Oxford Nanopore used a laptop computer to sequence rabbit DNA.

Alzheimer’s disease is a form of dementia that impacts the brain and results in memory loss. In recent years we have been able to study it and its risk factors to calculate the chances of you being affected by the disease. Today we are going to look at how that risk is calculated. 

Most people who develop Alzheimer’s are elderly and near the end of their lives. However, this is not always the case. Looking into calculating your risk can be interesting and done using APOE environmental factors and other risks. 

It is important to remember that this number is not absolute and can be subject to change. If you are concerned, make sure to consult your doctor. 

A little while back, Razib Khan used data from 23andMe to explore his family’s genetic history. He previously published his findings and summarized them. Today, I’m going to fill you in on what he had to say!

Khan was interested in genetics, anthropology, and history, mainly how we have changed the way lineages are marked thanks to the increase of technology. This developed a curiosity for Khan to learn more about his genome. 

Khan’s family are from the Comilla district in Bengal; all four of his grandparents were ethnic Bengali’s, and appearance-wise, he looks typically South Asian. He had some skepticism about the test and his genome.

Khan comes from a large family, where both his parents have many siblings. The large family has allowed Khan to have already a sense of his risks for certain diseases, along with the knowledge of his family’s medical history. 

Skeptical but curious, Khan tried out 23andMe during a sale offer. Expecting a generic result, Khan was surprised by the results. 23andMe’s ancestry painting algorithm.

This calculated Khan as 57% European, 43% Asain, and less than 1% African. Initially, Khan thought the Asian percentage was a little high, so he asked to see whether other South Asians had received such a high value. The answer was none had. 

From his ‘gene sharing,’ he found that generally, the Asian range from 23andMe’s algorithm was 10-35%. The lower portion of this percentage was from people in the northwest of the Indian subcontinent.

Those higher values were coming from the east and south. Intrigued by this, Khan consulted the paper Reconstructing Indian History, which explained this wide percentage range. 

The paper explained that South Asians are a combination of the European-like population and a marginally Asian-like population, made up of Ancestral Northern Indians (ANI) and Ancestral South Indians (ASI).

The former’s mixture went from 75% in the northwest to 45% in the far south. The paper cites the class system as a contributing factor to this. Khan assumed that his high Asian percentage was due to having more ASI than the norm. 

There was skepticism towards this due to the Comilla district borders, which would indicate that he has Tibeto-Burman ancestry that he may not have been aware of.

Neither of his parents was aware of a potential link, and 23andMe does not offer a more detailed response, leaving Khan with some uncertainty.

What next?

Khan decided to use 23andMe to genotype his parents and gain more insight into his family's heritage. During the waiting time, he spoke on his blog about his predicted outcomes of these tests. 

Khan initially thought that his father’s heritage might have been different from what they believed and wrote some extensive blog posts on this issue. The results from 23andMe stated that his father was less Asian than his mother. 

Khan ran their date through ADMIXTURE and EIGENSOFT using various parameters, and overwhelmingly the results said that his mother was more Asian than his father, leaving him with conflicting advice. 

Khan returned to the drawing board, looking at oral histories from both sides of the family, consulting origins of Tibeto-Burman ancestry and the possibility of eastern Bengali Muslims being in his lineage.

From his research, he concluded that he had no real change in terms of his self-conception, but rather Khan was now intrigued by historical population genetics instead of scientific genealogy. 

Final Word 

Khan concluded that although not much was revealed to him about his family, it is a hobby he has decided to keep hoping that more knowledge will become available in the future. 

In May 2019, it was reported in an article that over ten thousand sequence mismatches were observed between messenger RNA and DNA from the same individuals.

More recently, three technical comments were published by Science surrounding this article. It was concluded that at least 90% of the Li et al. RDD sites are technical artifacts. Here, we are going to explain how this conclusion was drawn.

For each RDD site that had at least five reads mismatching the genome, it was calculated that the fraction of reads with the mismatch, or the match, at each position in the alignment of the RNA-seq read to the genome on the + DNA strand.

Over 10,000 exonic RDD sites were found, which included thousands of RDD sites that were predicted to change protein sequence. These results actually implied the existence of a minimum of one, if not more, novel mechanisms of gene regulation.

This questioned some of the basic assumptions that are used daily in genetics. 

It turns out that it is not the existence of RDD sites that was so surprising; it was the significant biological impact of the sites and the subsequent implication that there are new regulatory pathways that were not previously known to us. 

The reason that some think that all of the RDD sites in Li et al. are false positives is that two groups have raised issues regarding the reported RDD sites. Both of these sources claimed that the majority of the sites presented in the findings were false positives. 

Actually, mismatches to the genome at RDD sites are almost always occurring at the ends of sequencing reads. All three of the technical comments that were made after the paper was published had included this observation. 

In response to the comments being made, a plausible explanation was proposed for the observations. To generate the cDNA, they added random short DNA sequences to each sample that acted as a primer for a DNA synthesis reaction.

At some sites, the random primers were not perfect matches to the mRNA, but they were still able to bind. During synthesis, the mismatches from the primers were incorporated into the cDNA, and this, in turn, led to a false signal of RNA editing. This explains the previous results. 

Another exercise that aimed to validate the finding had involved identifying peptide sequences that correspond to ‘edited’ RDD sites. It was pointed out that many of these sequences are actually good matches to multiple genes.

It was concluded that these RDD sites are false positives due to mismatched reads from paralogous genes. 

In conclusion, when looking for regions of the genome that look strange during an analysis, an individual will find strange and unexpected results. This applies even when a systematic error only affects 0.001% of the bases in the genome.

It turns out that the most interesting findings are actually less likely to be real. This was discovered through the process of finding false positives and mismatched results.

However, it does remain a possibility that forms of RNA editing that have not always been known are active in humans, and RNA sequencing technology can be helpful in finding out if new forms do exist. 

The common notion running through molecular biology is that the information present in DNA is transferred to RNA and then to protein.

Back in 2010, researchers made a potentially ground-breaking observation.

They found that within any given individual, there are tens of thousands of places where transcribed RNA does not match the template DNA from which it is derived.

This Phenomenon is Known as RNA Editing

In humans, it is generally thought to be limited to conversions of the base adenosine to the base inosine (which is read as guanine by DNA sequencers), and occasionally from cytosine to uracil.

However, these authors reported something new. They found that any type of base can be converted to any other type of base. If their observations are correct, these findings represent a fundamental change in how we view the process of gene regulation.

The Study 

The authors of this study sequenced the mRNA expressed by an individual (or rather, cDNA from a cell line derived from the individual). They then obtained DNA sequences from the same individual and compared the two.

Any difference between the RNA and DNA was taken as an indication of RNA editing. However, because it is impossible to sequence an entire mRNA or genome in a single pass, the researchers used short reads (of 50 bases) from the mRNA of the individual and reads of various lengths from the DNA of the individual.

They then matched these sequencing reads to the genome (or transcriptome) to see where they came from.

Complications 

However, there are several complications with this method: choosing the best spot to take reads from involves several assumptions, including how you weight insertions and deletions and possible sequencing errors.

These sorts of mapping issues are well understood and have been widely discussed in the literature on SNP calling from sequencing data, which is another situation where the researcher is looking for a difference between a sequencing read and the genome.

A naïve SNP caller that just looks for differences between aligned reads and a genome will output tens (or probably hundreds) of thousands of false-positive SNPs which must be filtered out by various criteria.

Therefore, mismapping of reads in paralogous regions can lead to false signals of RNA editing, and these false signals can even be replicated in follow-up experiments like those done by Li et al. (2011).

This is because the two forms of RNA and protein are indeed present in the cell, giving the illusion of RNA editing.

However, the two forms of RNA and protein do not come from the same DNA sequence, and thus are not evidence of RNA editing.

Besides, mapping biases around splice sites (and other sorts of insertions/deletions in the genome) will cause mismapping and false inference of RNA editing.

So in conclusion, while RNA editing is a potentially important phenomenon in humans, there is significant scope for further research, and skepticism of studies carried out so far certainly seems warranted.