High sequencing depths provided by Next Generation Sequencing (NGS) technologies indicate an exponential increase in the volume of sequence data in the last decade, but it comes with a cost. Do you think that various errors in NGS-generated sequence reads can be associated with high sequencing depth only? What are the other potential causes?
Well, my personal opinion, is that the polymerase is to blame. You have several thousand polymerases running that each have a certain error probability. So it is clear that the number of errors increased compared to Sanger.
The errors that Rene talked about are often due to heuristics to work around e.g. repeating regions (micro satellites) . Here algorithms use often some sort of randomness, which has been shown to work fine for a lot of other instances. The increased read length of e.g. Pac Bio will help, but there are still some regions left that are problematic. In addition, the longer sequences come with an increase error rate of e.g. 10%. Which I think is due to the synchronization of the laser and the polymerase. Eg. for methylated sites the polymerase seems to take longer for binding a nucleotide. But I think I am loosing the scope...
Dear @Rene Kaden, your point brings another question. Then, which result should we trust in? Depending on the algorithm, thresholds, there is always trade-offs with the results that one can get. How would you select which NGS result to further analyse, in your case (it is not a matter of which platform you are using, but I am curious about the output parameters you are more interested in)?
@Fritz J Sedlazeck, that is an interesting point, and I've never thought of that! However, I am not sure if I understood your point, whereby you mention that the increased read length of certain platforms would help decreasing the error rate. Would you please elaborate on this? How would longer lengths associate with smaller error rate?
Ah, that might be a misunderstanding. The increase read length will help to solve the problems of e.g. de novo assemblies. What Rene is talking about that if you ran it twice you will get different results. This is (I think) because of .e.g micro satellites, which can be overcome by using long reads. So it will just reduce the problems for de novo assembler.
I fully agree with Seweryn on Sanger sequencing being a consensus sequence - yet you can see "background noise variation" in the low peaks you get at the bottom of your electropherograms. Sometimes this even allows you to pursue low frequency variants in your sample by other methods - think of somatic variation in tumours and so on.
About bioinformatic analysis of NGS results: you have to tailor your analysis method to (1) the specific sequencing platform you are using, (2) the kind of experiment you are performing (it is not the same to use PCR-based methods for enrichment than capture methods, for instance, and vene between two capture methods there are differences that will make you choos one analysis method/algorithm over another), (3) the kind of sample you are sequencing... If you do not do this, you will never optimize your NGS. In my lab, we are developing by comparing different sets of analysis parameters what we believe is the "best practice analysis pipeline" for different kinds of experiments that we routinely perform. We intend to share this with he scientific community in the midterm.
Dear Seweryn Bialasiewicz, thanks for the detailed answer. I am wondering about the dataset you obtained, in which you had virus sequences. I assume that those organisms may not necessarily be the ones that you study in the lab, but may as well exist in the source DNA, through various interactions such as cross-individual and cross species. Maybe associations between the organisms can also contribute to the contamination. But the tricky part with the output NGS data is probably that it does not give you the whole story, as in where the so-called ''contaminants'' come from.
Dear Francisco J. Del Castillo, your answer regarding to the detection of somatic variation is interesting, I remember reading papers on this issue (ie Cibulskis et al., 2011). However, the contaminants can affect the results in a way that it causes false positives, so there should be ways of distinguishing real contaminants from low frequency variants.
To my knowledge, there are not many review papers addressing the multiple aspects of NGS data analysis, but rather its individual components, such as mapping and sequence alignments. That is why, I am really looking forward to trying your "best practice analysis pipeline."
Btw, in my lab we are currently developing a tool to detect contaminants in NGS data, and hopefully we will be releasing it soon. I would like to get feedbacks about it afterwards.
http://bioinformatics.oxfordjournals.org/content/27/18/2601.full
The trouble with NGS is that every single mutation you find must be backed up by a confirmation with another kind of technology, whether Sanger sequencing, Sequenom or whatever. At present, real contaminants and low-frequency variants cannot be told apart.
Somatic mutations are to be considered as well, every individual carries somatic mutations, the age of an induvidial is playing a role as well, more somatic mutations are present at higher age. This proces is also responsable for the development of cancers, especially when some risk mutations are already present in the germline.
- Badges
- Science method
- Similar topics
- Biological Science
- Genetics
- Genetic Analysis
- genetic techniques
- Sequencing