What to do with unmatched paired-end reads from kneadata outputs?

Firstly I used kneadata did a QC on paired-end reads from mice stool samples.
Then I got paired_1.fastq, paired_2.fastq, and unmatched_1.fastq, unmatched_2.fastq.

For metaphlan profiling, should I only use paired_1.fastq and paired_2.fastq as input, or should I use all 4 output above? Or say, what to do with unmatched paired-end reads?

Thanks for your help!

MetaPhlAn does not really use paired-end information, so both types of reads (paired and unmatched) can be used as input simultaneously as all reads are treated independently during mapping. You can either provide your fastqs to MetaPhlAn as a comma-separated list or concatenate the 4 files before running it.