lookahead in R to identify a pattern of words in order

user_2387584 0
Viewed 291

I am trying to parse a set of strings. I need to find out whether 'bcl-2' was detected in the sample. For example: "bl-2 was detected in 45% of patients". However there are certain possible variation which are challenging:

1."bcl-2 was detected in 45% bcl-6 was not detected"
2. "bcl-2 was not detected bcl-6 was detected in 45%"
3. "no evidendce of bcl-2 bcl-6 was detected in 45%"

So I am trying to define the regex code that would:

1. Lookahead for 'bcl-2'
2. Then, lookahead from that point for 'detected'
3. Then lookbehind between 'bcl-2' and 'detected' to make sure there is no 'not'.
4. If possible lookbehind 'bcl-2' to make sure there is 'no evidence of' (though I can take care of this condition separately)

I tried the following code which doesn't work. Specifically it doesn't lookbehind, so I am guessing there is something inherent to lookbehind that I am missing.

This regex works for "bcl-2 was not detected" but fails for "bcl-2 was detected in 45% bcl-6 was not detected"

y="bcl-2 was detected in 45% bcl-6 was not detected"
grepl("(?=bcl-?2)(?!.*not)(?=.*detected)",y, ignore.case = T,perl=T)

So I thought this will work but it doesn't:

grepl("(?=bcl-?2)(?=.*detected)(?<!not)",y, ignore.case = T,perl=T)

I am trying to understand the logic of lookbehind. In regards to the last line of code -> I thought (?=bcl-?2) looks forward until the point in the string that begins with 'bcl-2'. Then, I thought the (?=.*detected) looks forward until the position in the string where 'detected' start. Then I thought lookbehind starts looking backwards from that position for 'not'. This is of course wrong ... so what am I missing about the lookaround logic

BTW a great website I have been using in an attempt to figure this out https://www.regular-expressions.info/recurse.html

r pcre regex
1 Answers

Lookarounds are zero-width assertions, which means the regex index is not moving when the patterns are matched (the characters matched are not added to the match value and consecutive lookarounds all start their pattern checks from the same location). So, (?=bcl-?2)(?!.*not)(?=.*detected) matches an empty location (empty string) that is followed with bcl2 or bcl-2, that has no not substring after any 0+ chars other than line break chars, and that is followed with detected after any 0+ chars other than line break chars. This pattern is tried at every location in the input string, because there are no anchors. That pattern is hardly doing what you need.

Here is a possible solution:

\bbcl-2\b(?:(?!\bbcl-\d|\bnot\b).)*?\bdetected\b

See the regex demo:

  • \b - a word boundary
  • bcl-2 - a bcl-2 substring
  • \b - a word boundary
  • (?:(?!\bbcl-\d|\bnot\b).)*? - (a tempered greedy token) any 0+ (but as few as possible) chars other than line break chars that do not start the following two sequences:
    • \bbcl-\d - a wor boundary followed with bcl- and a digit
    • | - or
    • \bnot\b - a whole word not
  • \bdetected\b - a whole word detected

See an R demo below:

x <- c("bcl-2 was detected in 45% bcl-6 was not detected", 
"bcl-2 was not detected bcl-6 was detected in 45%",
"no evidendce of bcl-2 bcl-6 was detected in 45%")
grep("\\bbcl-2\\b(?:(?!\\bbcl-\\d|\\bnot\\b).)*?\\bdetected\\b", x, perl=TRUE, value=TRUE)
## => [1] "bcl-2 was detected in 45% bcl-6 was not detected"
user_38329700
Related