Using your example you would need to only process read 2 since based on Takarabio directions that is the read that contains UMI.

$ more test.fq
@NB551656:25:H3N2YBGXK:1:11101:17925:1178 2:N:0:4
GTCATGAACGAGTCAGGCCAAGGGCATCAATTGCCCGTCACCGGAAGGCGCATTCTACGTCTACCCGTCCTGCGCC
+
AAAAAEEE////A/EEEEEEEEEEEEEEEEEEEE<EEEEAEEEEEEEEEEEEEEEEEEEEEEEE<EEEEEEEAEAE
@NB551656:25:H3N2YBGXK:1:11101:6227:1179 2:N:0:4
GTGGGTTCCTTTGGTCTTGTTGCGTACCTGGAGAACGGAAGAGCGTCGTGTAGGGAAATAGTGTAAGTCCAAGTGT
+
AAAAAEEA/A///EE//EE/EE/AEEE<E//E/E/AE/EE//E6/EE/E/EE/E<E<E/E/E/EAEE</6AE///A

$ umi_tools extract --stdin=test.fq --bc-pattern=NNNNNNNN --log=processed.log --stdout=processed.fastq

$ more processed.fastq 
@NB551656:25:H3N2YBGXK:1:11101:17925:1178_GTCATGAA 2:N:0:4
CGAGTCAGGCCAAGGGCATCAATTGCCCGTCACCGGAAGGCGCATTCTACGTCTACCCGTCCTGCGCC
+
////A/EEEEEEEEEEEEEEEEEEEE<EEEEAEEEEEEEEEEEEEEEEEEEEEEEE<EEEEEEEAEAE
@NB551656:25:H3N2YBGXK:1:11101:6227:1179_GTGGGTTC 2:N:0:4
CTTTGGTCTTGTTGCGTACCTGGAGAACGGAAGAGCGTCGTGTAGGGAAATAGTGTAAGTCCAAGTGT
+
/A///EE//EE/EE/AEEE<E//E/E/AE/EE//E6/EE/E/EE/E<E<E/E/E/EAEE</6AE///A

After this you can hard trim 6 bp from front of processed reads to remove the linker+UMI adapter using reformat.sh from BBMap suite.

$ reformat.sh -Xmx2g in=processed.fastq out=out.fq forcetrimleft=6

$ more out.fq

@NB551656:25:H3N2YBGXK:1:11101:17925:1178_GTCATGAA 2:N:0:4
AGGCCAAGGGCATCAATTGCCCGTCACCGGAAGGCGCATTCTACGTCTACCCGTCCTGCGCC
+
EEEEEEEEEEEEEEEEEEEE<EEEEAEEEEEEEEEEEEEEEEEEEEEEEE<EEEEEEEAEAE
@NB551656:25:H3N2YBGXK:1:11101:6227:1179_GTGGGTTC 2:N:0:4
TCTTGTTGCGTACCTGGAGAACGGAAGAGCGTCGTGTAGGGAAATAGTGTAAGTCCAAGTGT
+
E//EE/EE/AEEE<E//E/E/AE/EE//E6/EE/E/EE/E<E<E/E/E/EAEE</6AE///A

I am not familiar with that particular kit, but you are almost there. However, the UMI is in your second read, so using --bc-pattern would mean that the UMI is being looked for in the wrong read!

As the UMI-tools manual describes, you can use the letters N C and X to denote the composition of the barcode in string mode. Bases with Ns and Cs will be extracted and added to the read name. The corresponding sequence qualities will be removed from the read. So using --extract-method=string and --bc-pattern2=NNNNNNNN should work.

Even more clever might be the --bc-pattern2=NNNNNNNNCCCCCC, which would treat the UMI-Linker and 3 base adapter sequence as cell barcode and move it to the header, too. Hence, it trims the linker away such that it doesn't interfere with the mapping later, since denoting the bases with an X would mean that they are reattached to the read and thus not removed. Treating the linker as cell barcode would also help to corroborate if you are correctly trimming the UMI, because that sequence should always be the same. So by visually inspecting a few read headers later, you will know if everything worked out as expected.

If first base of read2 is not always the first UMI base, which might happen if there is still a part of the adapter sequence in the read, use the Regex mode instead. In that case the 3'-sequence of the i7 adapter as well as the linker sequence can be used to locate your UMI.

I couldn't find the sequence of the linker (but you should be able to see that in your reads easily), but the i7 adapter should be GATCGGAAGAGCACACGTCTGAACTCCAGTCAC-[index i7]-ATCTCGTATGCCGTCTTCTGCTTG. Therefore, a possible Regex to locate your UMI could look like --bc-pattern2="(?P<discard_1>CGTCTTCTGCTTG){s<=1}(?P<umi_1>.{8})(?P<discard_2>REPLACE WITH SEQUENCE OF THE LINKER)".

You might need to adapt that depending on the amount of adapter still to be found. It might also be possible to only use the sequence of the linker as anchor and skip the discard_1 group entirely. Good luck with your project!

PS: I would appreciate if you could post the six bases here that constitute the NNNXXX part in the figure above as a future reference in case we start working with the same kit. Thanks!

You can also refer to Nextflow documentation here: https://nf-co.re/rnaseq/3.18.0/docs/usage/#unique-molecular-identifiers-umi