Hi,

  I have human total RNA-seq data (PE, Next-Seq) from a pilot study with two samples and am getting results from featureCounts that do not make any sense to me. I process the data as follows: (1) I perform adapter trimming using ea-utils mcf and mild quality trimming using btrim. I use STAR for alignment to the genome. STAR tells me that the first sample has 99,018,190 input reads (these are read pairs) and the 2nd sample has 126,164,150 input reads.  For the first sample, 86,571,963 reads were aligned (70,579,007 uniquely), and for the 2nd sample, 112,525,928 reads were aligned (89,730,382 uniquely).  Now when I run featureCounts on these data, it prints out this information:

First sample:    Total fragments: 123128857,   Successfully assigned fragments : 63971619 (52.0%)

2nd sample:    Total fragments: 168328570,    Successfully assigned fragments : 83223103 (49.4%)

and this warning: 

WARNING: reads from the same pair were found not adjacent to each other in the input (due to read sorting by location or reporting of multi-mapping read pairs).    

It seems that the total fragments from featureCounts do not match up at all with the read counts from STAR.

Thanks, Ina

FeatureCounts reports number of alignments whereas STAR reports number of reads (number of reads pairs in this case since the data is paired end). The reporting of a mapped read may include one or more alignments. A uniquely mapped read will lead to the reporting of one single alignment, but a multi-mapping read will result in more than one alignment being reported. So the difference you observed was caused by the reporting and counting of multi-mapping reads.

If you instruct STAR to output uniquely mapped reads only, then featureCounts will report the same total count. When STAR is allowed to output multi-mapping reads, the total count from featureCounts is always higher because it reports the number of alignments rather than number of reads.