Styjun

I did sequencing on two genotypes and extracted the corresponding SNP chip SNPs to compare our genotypes to publicly available data on the exact same genotypes. I noticed that some SNPs must be of opposite strand polarity and need to be flipped. How can I use R to check if the sample calls match the reference or alternate SNP calls, and if they do not, to replace the sample calls with the complementary nucleotide?

I tried to incorporate ifelse and chartr, but I failed.

Example Data

test <- data.frame("pos" = 1:5, "ref" = c("A", "T", "C", "C","G"), "alt" = c("G","C", "A", "T", "A"), "sample" = c("A", "A", "C", "G", "G"))
View(test)

+---------+------+-------+----------+
| pos | ref | alt | sample1 |
+---------+------+-------+----------+
| 1 | A | G | A |
| 2 | T | C | A |
| 3 | C | A | C |
| 4 | C | T | G |
| 5 | G | A | G |
+---------+------+-------+----------+

Desired Output

+---------+------+-------+----------+
| pos | ref | alt | sample1 |
+---------+------+-------+----------+
| 1 | A | G | A |
| 2 | T | C | T |
| 3 | C | A | C |
| 4 | C | T | C |
| 5 | G | A | G |
+---------+------+-------+----------+

You can use the library dplyr to manipulate your dataset. With mutate() for instance, you can check if the sample call match the reference.
Then, you can use ifelse() to make some changes.

library(dplyr)


test %>% 
 mutate(TEST1 = (ref == sample)) %>%
 mutate(sample2 = ifelse(TEST1 == TRUE, ref, sample))

# pos ref alt sample TEST1
# 1 1 A G A TRUE
# 2 2 T C A FALSE
# 3 3 C A C TRUE
# 4 4 C T G FALSE
# 5 5 G A G TRUE

Using ifelse and chartr:

ifelse(test$sample == test$ref, test$sample, chartr("ATCG", "TAGC", test$sample))
# [1] "A" "T" "C" "C" "G"

While you asked for solution in R, you still may want to consider other tools to flip strands especially if you need to do it on larger scale. You can do it with Plink and many other tools. Also I probably would seek to confirm the strand alignment in public data from their records/Method description. If only some of the SNPs in public datasets are/seems to be flipped, it is suspicious (one would expect that alignment would be uniform within the data from the same source).
Also, if you use positional arguments, you should confirm the genome build/versions in your vs public data, because different versions can also be a source of confusion. Finally, I would probably also want to make sure that the SNPs in questions are not multiallelic, as it can also be a source of divergence..