variant allele fraction

[zhaopeihua]

hi:
What are "variant allele fraction" when talking about somatic mutations?

[lethalfang]

hi:
What are "variant allele fraction" when talking about somatic mutations?

The fraction of alleles in the sample that are mutations.
For instance, if all cells contains the same heterogeneous mutation (e.g., 100% tumor cell purity if you're looking at tumor cells), the variant allele fraction would be 50% (half-half from each chromosome).

[zhaopeihua]

The fraction of alleles in the sample that are mutations.
For instance, if all cells contains the same heterogeneous mutation (e.g., 100% tumor cell purity if you're looking at tumor cells), the variant allele fraction would be 50% (half-half from each chromosome).

thanks for you reply，but I didn't fully understand this concept. Could you show me a detail example with computational process? thanks again

[lethalfang]

This is an example of an output in my somatic calling pipeline:

Variant Annotation	Gene	Exon	Codon	Chr	Position	Variant	N(A)	N(C)	N(G)	N(T)	T(A)	T(C)	T(G)	T(T)	QSS	Frequency
nonsynonymous SNV	KRAS	exon2	G12A	chr12	25398284	G35C	0	83	0	0	1	37	12	0	57	0.24

N(A), N(C), N(G), and N(T) stand for the number of calls for A, C, G, and T in the normal sample. T(A), T(C), T(G), and T(T) stand for the number of calls for A, C, G, and T in the tumor sample.

So for the normal sample, there are 83 sequences where the base is called C, and 0 for everything else. So it's pretty clear that in the reference and the normal sample, the base is C.
However, in the tumor sample, there are 37 calls for C (reference) and 12 calls for G (variant), so the variant frequency is 12/(12+37) = 0.245.
Notice the variant is labeled G35C. That's because the coding strand is the minus strand, where the DNA reads use the plus strand.

Anyway, how do you interpret 24.5%?
Well, if I assume that
1) the KRAS mutation is present in all tumor cells (just use this assumption as an example), and
2) KRAS mutation is a heterozygous mutation, then
The tumor sample must contain 1/2 non-tumor cells, because those cells give me half of the reference reads. Out of the remaining 1/2 are tumor cells, half of those chromosomes give me reference reads, and the other 1/2 chromosomes give me mutant reads.

[zhaopeihua]

Awesome, thanks!

[vd4mindia]

@lethalfang
`Would you like to share the pipeline for finding somatic mutations. I have 2 samples, one low and one high grade tumor and its normal. I want to detect the point somatic mutations. I have already employed GATK, VarScan and Mutect, but I have some descrepancy when am looking at my Low grade tumor since it is 50% pure. So I would like to try something more and have a read statistics like you have shown to validate the hits. If its ok for you

[lethalfang]

@lethalfang
`Would you like to share the pipeline for finding somatic mutations. I have 2 samples, one low and one high grade tumor and its normal. I want to detect the point somatic mutations. I have already employed GATK, VarScan and Mutect, but I have some descrepancy when am looking at my Low grade tumor since it is 50% pure. So I would like to try something more and have a read statistics like you have shown to validate the hits. If its ok for you

That's something I used to do a couple of years ago.
I just had a script to get those information out of the Strelka output vcf file. I had all those Strelka out directories in a single location, and ran this script to "summarize" all mutation calls from all the samples:

http://kimlab.surgery.ucsf.edu/media/5654495/summarize_strelka_findings.txt

It also calls annovar, which has a hard-coded path within the script.

[virpericulosus]

Just wondering: The T(A) isn't counted because the frequency is just for T(G) variants?

[vd4mindia]

This is an example of an output in my somatic calling pipeline:

Variant Annotation	Gene	Exon	Codon	Chr	Position	Variant	N(A)	N(C)	N(G)	N(T)	T(A)	T(C)	T(G)	T(T)	QSS	Frequency
nonsynonymous SNV	KRAS	exon2	G12A	chr12	25398284	G35C	0	83	0	0	1	37	12	0	57	0.24

N(A), N(C), N(G), and N(T) stand for the number of calls for A, C, G, and T in the normal sample. T(A), T(C), T(G), and T(T) stand for the number of calls for A, C, G, and T in the tumor sample.

So for the normal sample, there are 83 sequences where the base is called C, and 0 for everything else. So it's pretty clear that in the reference and the normal sample, the base is C.
However, in the tumor sample, there are 37 calls for C (reference) and 12 calls for G (variant), so the variant frequency is 12/(12+37) = 0.245.
Notice the variant is labeled G35C. That's because the coding strand is the minus strand, where the DNA reads use the plus strand.

I would just want to ask you in case you have have not other bases as 0 in normal in that case do you still calculate the variant allele frequency as the above? Variant allele frequency should be always considered for the reads having variant allele for the tumor sample divided by the total reads at that variant site? Right? Irrespective of the reads falling for other bases in normal samples as well right?

[lethalfang]

I would just want to ask you in case you have have not other bases as 0 in normal in that case do you still calculate the variant allele frequency as the above? Variant allele frequency should be always considered for the reads having variant allele for the tumor sample divided by the total reads at that variant site? Right? Irrespective of the reads falling for other bases in normal samples as well right?

I considered "variant allele frequency" as the frequency of variant alleles in a given sample.
So the tumor's variant allele frequency does not care what's in the normal tissue, and vice versa.
That way, you can monitor the change in variant allele frequency between the tumor and the normal.

[vd4mindia]

Thank you very much for the reply. That makes sense.