...

As reads can be aligned to more than one location, the number of alignments may be greater than the number of reads; since some reads may be unaligned, the number of alignments may be less than the number of reads in the bam/sam file (Figure 1).

PGS shows the number of alignments per sample in the “parent” spreadsheet of a RNA-seq project, while the alignments and reads per sample are reported in the mapping_summary spreadsheet. The alignment_counts spreadsheet contains the number of alignments per read in each sample and it can be invoked through the QA/QC section of the RNA-seq workflow.

...

• Exonic: A read is labeled exonic if any one of its alignments is completely contained within the respective exon as defined by the database (i.e., even if there is a single base shift of the read relative to the exon, the read will not be called exonic but will fall in the category ‘partially overlaps exon’). If the alignments are strandspecific, then the strand of the alignment must also agree with the strand of the transcript.
• Partially overlaps exon: A read is assigned to partially overlap an exon if any of its alignments overlaps an exon, but at least partly (one base-pair or more) maps out of the exon.
• Intronic: A read is labeled intronic if any one of its alignments maps completely within an intron, but none of the alignments are exonic (either fully or in part). If the alignments are strand-specific, then the strand of the alignment must also agree with the strand of the transcript.
• Reads between genes: A read is labeled ‘between genes’ if none of its alignments overlap a gene.

RPKM Scaling 

Standard output of mapping performed by the quantification step includes raw read counts and scaled read counts for every gene and transcript for each sample. The scaling method currently applied is reads per kilo-base of exon model per million mapped reads (RPKM) (Mortazavi et al. Nat Methods 2008). It scales the abundance estimates using exon length and millions of mapped reads and is calculated according to the formula below.

...

Compatible reads with junction are reported in a separate block of columns in the transcripts spreadsheet (“junction RPKM”).

Paired-End Data Scenarios 

...

Considering genes with multiple transcripts, a read can be both counted as compatible for some transcripts, as well as counted incompatible for other transcripts of the same gene (Figure 4). Please note that this concept holds for single-end reads as well.

Furthermore, all the reads that have at least one alignment contribute to the denominator of the RPKM (millions of _mapped reads_ per sample). Similar to that, the denominator of the RPKM contains all the mapped reads, regardless of whether or not a read is compatible with any transcript.

With respect to the gene-level summarization, user can decide whether or not to consider intronic reads (both single- and paired-end) as compatible with the gene (Figure 5). In the latter case, the entire gene is basically treated as one giant exon. In the case that one end of a paired-end read falls within a gene, and the other end does not, the read will not be included in the gene-level summarization.

Unexplained regions 

The unexplained regions portion of quantification considers any read that is considered “not compatible” with all transcripts. It is basically a 3 step process:

...

When interpreting the unexplained reads, you should have in mind that these are actually the reads not compatible with the applied transcript model. By changing the model, some reads will become compatible and thus will not be labeled “unexplained” any more. Figure 6 shows such an example. The depicted regions map just downstream of the human LONP2 gene as defined by RefSeq and are hence flagged as unexplained. However, by overlaying the AceView transcripts, it is apparent that mapping to AceView would yield a different result.