This is not a text on biology, genomics or bioinformatics. But we will be using a population genomics dataset, hence we will introduce some necessary concepts. The text will be kept to a bare minimum with links to further explanations (as a rule the English version of Wikipedia is a great port of call).
We will not teach you genetics and genomics here, furthermore the level of explanation that you will find is below what you can find on Wikipedia. But we do need to explain some basic concepts in order to make sure you know all that is needed to understand the data.
When you were born, you received genetic material (DNA) from your parents. In our species that means 22 pairs of autosomes , 1 pair of allosomes and mitochondrial DNA. While the terminolgy might seem strange, it is actually quite easy:
one X chromosome from your mother and either an X (if you are a female) or Y (if you are a male) from your father. Females have 2 X chromosomes, males are XY.
your father. They have mostly the same form and size and encode similar genes.
of DNA in circular form.
In this course we will only be using autosomes, that is the 22 pairs of chromosomes where you got similar forms from both of your parents (actually we will be using mosquito autosomes, but the concept is the same).
Because this is a big-data course, the sizes of the chromosomes are an important topic. Here, for reference is the size of human ones (take from Wikipedia):
So, we are dealing with around 3GB of information per individual. In theory 2 bits ber base is enough, remenber that we only have ACTG, but most encodings that you see are text based (ASCII), so it is probably a byte per base (actually two bytes as we will see).
Note the small size of the mitochondrial DNA and the difference in size between X and Y.
Think, for example on chromosome 3 that you received from your mother. Where does it come from? From your grandmother or your grandfather? It turns out that the answer is not as obvious at it seems. Autosomes do recombine inside our cells. This means that that chromosome 3 from your mother might be:
Roughly you might think that there is around one recombination event per chromosome, say between 0 and 2.
Think a bit about the consequences here: while you have roughly half of the genetic material from your parents, your grandparents are not equally represented.
Recombination is important in terms of statistical properties of the dataset, take a bit of time to reflect on the data analysis consequences of recombination.
Finally, you might remember from school that the most common unit of chromosome size is not the number of base pairs, but Morgans, well these reflect exactly the recombination rate!
The DNA that you receive from you mother and father is different. There are many kinds of differences, but here we will only concentrate on arguably the simplest one: Single-Nucleotide Polymorphisms (SNPs). A SNP is the variation of a single base pair on the same position across the genome. For example, look the following piece of genome in 3 different individuals:
Remeber that with autosomes individuals have 2 copies of the same genetic material, hence two entries per individual.
Positions 2 and 5 are SNPs, that is, there is a mutation across the individuals sequenced at those positions.
Individual 1 is heterozyguous at position 2, i.e. it has a different nucleotide for the same position.
For humans, there is very little variation across the genome, roughly 1 SNP every 2000 base pairs (note to self: check the accuracy of the number).
For most of our examples, we will use not human data but a dataset from the mosquito that transmits malaria.
Rigorously, the Anopheles mosquito does not transmit malaria, but transmits Plasmodium, the parasite that causes malaria.
Now that you know a bit about human genomics we can discuss mosquito genomics. Fortunately they are very similar (if you are not a geneticist, you would be shocked at the variation in genomic structure that can be found in nature).
Sex in anopheles mosquitoes is similar (genomically) to humans: A X and a Y chromosome. There is also a mitochondria. Mosquitoes have only two pairs of autosomes, that purely for ease of convention are split in left and right arms. For some weird reason they are numbered 2 and 3 (2L, 2R, 3L and 3R with the arms) - no 1. The sizes are:
We have around 270 Mbp, one order of magnitude lower than humans. Notice that, as with most species, we do not have a very good genome assembly for Anopheles. No Y assembled (females - XX - are more important, because they are the ones that transmit malaria) and quite a lot of unknown bits.
Now, the interesting part is that while humans have little genomic variation, Anopheles mosquitoes site on the other extreme. We are probably dealing with a SNP every 4 base pairs. When we are reduced to SNPs, these mosquitoes have at least 2 orders of magnitude more information than humans. Genome size does not have to be a good proxy for SNP density.
There are plenty of sequencing technologies around, this text will be based on the most common one in use. Obviously we will keep this very simple.
Now that you have your DNA available it is time to sequence it. Unfortunately sequencing technology is very redumentary. Do you think you can get a chromosome from start to end? We are very far away from that. What we normally get is reads of around 100 base pairs.
So the first problem that we have to solve is a mapping problem: given our 100 bp read where does it fit on the genome (That is a search space of 3Gbp for humans or 270). If you think about it, this is a massive puzzle to solve. Years of research and millions of dollars have been put into this.
Now the biggest problem is that the read from a sequencer can have errors. So algorithms will have to deal with that. The sequencer gives you a measure of trust for each base read (A PHRED score) - i.e. the probability of being a correct read. So you get not only the base, but a level of trust. So, for a 3 Gbp genome, you now can expect to deal with 6 Gbp of data (50% DNA reads, 50% trust levels).
How do the algorithms deal with errors? They do that by requiring you to sequence a lot of data. Ideally you should cover each position around 30 times. You need 30 times coverage. So, a 3 Gbp human genome will generate 6Gbp of data times 30. We are now at 180 Gigabytes (uncompressed) per human sample.
So after solving the puzzle with mapping, where you assign each 100 bp read to a position in a reference genome you can now think in SNP calling where you call your SNPs per sample. This is a fairly complex process as it will have to look at error rates per position and available coverage.
So, at the end you will have a file with SNP calls, because calling SNPs is not trivial, you get the calls and a lot of other metrics that you can use to filter the data yourself.
The data that we will be using is made available in VCF format, lets check that now.
Links to wikipedia