Why are some transcripts expressed at 0 in all samples, resulting in the count matrix having rows with all 0 values?

[ljyls]

I used the standard Trinity pipeline to assemble the transcripts and used salmon to generate the count matrix. Why are some transcripts expressed at 0 in all samples, resulting in the count matrix having rows with all 0 values? From what I understand, there should be at least one sample with a count value other than 0 for any transcript. Because at least 1 sequence will match at least 1 sample of the fastq file, once matched, the value will not be 0.