By looking at summ_for_scaffolds.txt, we see that some collectContigs
assemblies don't have any reads map to them, in expectation, according to RSEM.
This is confusing. Let's check whether blat aligns any reads to them.

Specifically, we take a collectContigs assembly, ws30, which has 12
transcripts with no reads mapping to them in expectation, according to RSEM
(see ./sum_for_scaffolds.txt). For every transcript named in
  trinity-gridsearch/rsem_oracleset.ws30.fa
we look in 
  trinity-gridsearch/rsem_oracleset-summary.ws30.rsem-eval-workingdir/expression.temp/expression.sam.gz
to see if there is at least one pair of reads that maps to it:

  sam=trinity-gridsearch/rsem_oracleset-summary.ws30.rsem-eval-workingdir/expression.temp/expression.sam.gz
  for id in `dl-fasta-list <trinity-gridsearch/rsem_oracleset.ws30.fa`; do
    echo -n "for $id: "
    zcat $sam|grep "$id"$'\t'|grep -v ^'@'|wc -l
  done

Indeed, from teh output of this script, we see that all the transcripts do
have at least one pair of reads mapping to them. However, some transcripts
only have 2 reads mapping to them. How many are like this?

  sam=trinity-gridsearch/rsem_oracleset-summary.ws30.rsem-eval-workingdir/expression.temp/expression.sam.gz
  for id in `dl-fasta-list <trinity-gridsearch/rsem_oracleset.ws30.fa`; do
    echo -n "for $id: "
    zcat $sam|grep "$id"$'\t'|grep -v ^'@'|wc -l
  done | grep ": 2$" | wc -l

There are 10 such transcripts. This doesn't exactly match the 12 expected to
have no matches by RSEM, but it is close. (and there are other transcripts
that have only 4 or 6 reads map to them.)

Example:
$ zcat $sam|grep "scaffold_11"$'\t'|grep -v ^'@'
SRR203276.21852708      99      scaffold_11     1       255     76M     =       95      170     CTTTTTTTTTTTTTTTTATTATTTATTTTTTTTTTTTTTTCTTATTATTTTTATTATTATTTTATTATTATAAAAA        ############################################################################        XA:i:2  MD:Z:16A3T4A6A5A1T2T0A1T2G0G1T1G0T0G0G1A0A2C2T0A0G0G1T0T0T2 NM:i:26
SRR203276.21852708      147     scaffold_11     95      255     76M     =       1       -170    AAAAAAAAAAAATAAAAAAAAAAAAAAGAAAAAGAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAT        ############################################################################        XA:i:2  MD:Z:0G0G0G0G0G0G0G0G0G0G0G0G1G0T0T0G0T1G1C1C0T0G0G1G0G0C0T0C0T2T4G0T0G1C0T0C1C0C2C21A0     NM:i:40

If we look at the actual sequences in question:
  from sample_1.fq:
  @SRR203276.21852708 length=76
  CTTTTTTTTTTTTTTTTATTATTTATTTTTTTTTTTTTTTCTTATTATTTTTATTATTATTTTATTATTATAAAAA
  from sample_2.fq:
  @SRR203276.21852708 length=76
  ATTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTCTTTTTCTTTTTTTTTTTTTTATTTTTTTTTTTT
  from trinity-gridsearch/rsem_oracleset.ws30.fa:
  >scaffold_11 tid=2348 pos=1943 length=182 number_of_contigs=2
  CTTTTTTTTTTTTTTTAATTTTTTAATTTTTTATTTTTATTTTTATTTTGGTTTGTGGAAATTCTTTAGGTTTTAAnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnTGTTGTAGACACTGGGGGCTCTAATAAAAGTGACTCACCAACAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
We see that it's plausible for it to be counted as a match, but it is
certainly not a great match.