Fast way to select rows within table in R?

Question

I am looking for a fast way to extract a large number of rows from an even larger table. The top of my table is as follows:

> head(dbsnp)

      snp      gene distance
rs5   rs5     KRIT1        1
rs6   rs6   CYP51A1        1
rs7   rs7 LOC401387        1
rs8   rs8      CDK6        1
rs9   rs9      CDK6        1
rs10 rs10      CDK6        1

And the dimensions:

> dim(dbsnp)
[1] 11934948        3

I want to select the rows that have the rownames contained in a list:

> head(features)
[1] "rs1367830" "rs5915027" "rs2060113" "rs1594503" "rs1116848" "rs1835693"

> length(features)
[1] 915635

Not surprisingly, the straightforward way of doing this temptable = dbsnp[features,] takes quite a long time.

I've been looking into ways to do this through the sqldf package in R. I thought that that might be faster. Unfortunately, I can't figure out how to select rows with certain rownames in SQL.

Thanks.

Answer 1

The data.table solution:

library(data.table)
dbsnp <- structure(list(snp = c("rs5", "rs6", "rs7", "rs8", "rs9", "rs10"
), gene = c("KRIT1", "CYP51A1", "LOC401387", "CDK6", "CDK6", 
"CDK6"), distance = c(1L, 1L, 1L, 1L, 1L, 1L)), .Names = c("snp", 
"gene", "distance"), class = "data.frame", row.names = c("rs5", 
"rs6", "rs7", "rs8", "rs9", "rs10"))

DT <- data.table(dbsnp, key='snp')
features <- c('rs5', 'rs7', 'rs9')
DT[features]

   snp      gene distance
1: rs5     KRIT1        1
2: rs7 LOC401387        1
3: rs9      CDK6        1

Answer 2

Using sqldf you will need rownames = TRUE then you can query on the rownames using row_names:

library(sqldf)

## input

test<-read.table(header=T,text="      snp      gene distance
rs5   rs5     KRIT1        1
rs6   rs6   CYP51A1        1
rs7   rs7 LOC401387        1
rs8   rs8      CDK6        1
rs9   rs9      CDK6        1
rs10 rs10      CDK6        1
")
features<-c("rs5","rs7","rs10")

## calculate

inVar <- toString(shQuote(features, type = "csh")) # 'rs5','rs7','rs10'

fn$sqldf("SELECT * FROM test t
          WHERE t.row_names IN ($inVar)"
           , row.names = TRUE)

## result
#      snp      gene distance
#rs5   rs5     KRIT1        1
#rs7   rs7 LOC401387        1
#rs10 rs10      CDK6        1

UPDATE: Alternately if fet is a data frame whose features column contains the required items to find:

fet <- data.frame(features)
sqldf("SELECT t.* FROM test t
          WHERE t.row_names IN (SELECT features FROM fet)"
           , row.names = TRUE)

Also if the data were sufficiently large we could speed it up using indexes. See the sqldf home page for this and other details.

Answer 3

The way most people would initially try it would be:

dbsnp[ rownames(dbsnp) %in% features, ]  # which is probably slower than your code

Because you say this is taking a long time, I suspect you have exceeded your RAM capacity and have started using virtual memory. You ought to shutdown your system and then restart with just R as a running application and see if you can avoid "going virtual".