some good advice has been given above. if performance is of any concern, you'll save a little time by compiling the regex only once, by using the -o modifier, such as my $exp = $source =~ s/$group//go; to borrow gmax's solution for illustration.

of course, direct interpolation of the pattern will be even faster, such as my $exp = $source =~ s/BC//g;

my benchmark results look like:

C:\WINDOWS\Desktop>perl test_regex-o.pl
Benchmark: timing 100000 iterations of s_direct, s_witho, s_withouto..
+.
  s_direct:  4 wallclock secs ( 3.02 usr +  0.00 sys =  3.02 CPU) @ 33
+112.58/s (n=100000)
   s_witho:  4 wallclock secs ( 3.41 usr +  0.00 sys =  3.41 CPU) @ 29
+325.51/s (n=100000)
s_withouto:  5 wallclock secs ( 4.28 usr +  0.00 sys =  4.28 CPU) @ 23
+364.49/s (n=100000)
[download]

~Particle