For ws3, we have

$ zcat ws3.expression.sam.gz |grep scaffold_25_contig_4
@SQ     SN:scaffold_25_contig_4 LN:186

i.e. no reads align to this scaffold according to rsem-compute-expression.

If we just look for lines containing scaffold_25, we find

@SQ	SN:scaffold_25_contig_0	LN:76
@SQ	SN:scaffold_25_contig_1	LN:187
@SQ	SN:scaffold_25_contig_10	LN:76
@SQ	SN:scaffold_25_contig_2	LN:201
@SQ	SN:scaffold_25_contig_3	LN:76
@SQ	SN:scaffold_25_contig_4	LN:186
@SQ	SN:scaffold_25_contig_5	LN:86
@SQ	SN:scaffold_25_contig_6	LN:220
@SQ	SN:scaffold_25_contig_7	LN:121
@SQ	SN:scaffold_25_contig_8	LN:325
@SQ	SN:scaffold_25_contig_9	LN:97
SRR203276.5781444	163	scaffold_25_contig_6	1	255	76M	=	145	220	GGACTCTTCATTTTTAATTGTTATTCAATGTTTTATTTTCCAGAACTCATTTTTCACATTCAGGGGTGGGAGCTGA	CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCBBBBCCCCCCCCCCCCCCCACCCCCACCCCCC@@;CCCC<	XA:i:0	MD:Z:76	NM:i:0
SRR203276.5781444	83	scaffold_25_contig_6	145	255	76M	=	1	-220	CGGGGCAATAGCTCCTGAGCAGACTTTCAGAAGGGAGAGGTCTTCTCAGAAGCCTGTTGGACCTGGCTTGCAGAGG	BBDBB@BDBDBCDDACCDDCDCDCCDDCCCCCCCCCCCCCCDCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC	XA:i:0	MD:Z:76	NM:i:0
SRR203276.6596098	163	scaffold_25_contig_8	1	255	76M	=	250	325	GTGGGCTGAAAACAGTTTCTTCAGGGGATTACTAAACTCAAGAATTATTAACCTTTCTACTTTTTGAAACAATGAT	CCCCCCCCCCCCCCCBCCCCCCCCCCCC?CCCCC@CDCCDBCADCCACCCCCDBCDCCACDCCCCCBBCCA9CCB@	XA:i:0	MD:Z:76	NM:i:0
SRR203276.6596098	83	scaffold_25_contig_8	250	255	76M	=	1	-325	GAAAGACAACTGTTTATATAGACCCCATTGGAAAAATCCCAGTTCTGTACTGAGATCAGAGACCCCAAACTCTTAC	#B9C@@AB@CCCABDCCAACBBCABCC@?DDCCCCABCCCCBCCCCCCCDCCCCCCCCCCCADCCCCCCCCCCCCC	XA:i:0	MD:Z:76	NM:i:0
SRR203276.50638906	163	scaffold_25_contig_6	1	255	76M	=	145	220	GGACTCTTCATTTTTAATTGTTATTCAATGTTTTATTTTCCAGAACTCATTTTTCACATTCAGGGGTGGGAGCTGA	CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCBBBBCCCCCCCCCCCCCCCCCCCCC?CCCCBC@@>CDCC@	XA:i:0	MD:Z:76	NM:i:0
SRR203276.50638906	83	scaffold_25_contig_6	145	255	76M	=	1	-220	CGGGGCAATAGCTCCTGAGCAGACTTTCAGAAGGGAGAGGTCTTCTCAGAAGCCTGTTGGACCTGGCTTGCAGAGA	@DDBCADDDDBCDCDCCBCCDCBCCCDCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC	XA:i:1	MD:Z:75G0	NM:i:1

i.e. only contigs 6 and 8 are said to be aligned.

If we run

$ bowtie -l 13 -q --phred33-quals -n 3 -e 99999999 -I 1 -X 1000 -p 8 -a -m 200 -S /tier2/deweylab/nathanae/rnaseq-eval/data-jul26//trinity-gridsearch/rsem_oracleset-summary.ws3.rsem-eval-workingdir/reference -1 /tier2/deweylab/nathanae/rnaseq-eval/data-jul26/sample_1.fq -2 /tier2/deweylab/nathanae/rnaseq-eval/data-jul26/sample_2.fq|grep scaffold_25_contig_4>>differences2.txt

(the same bowtie command that rsem-compute-expression runs, except with -l 13
and -n 3 instead of -l 25 and -n 2) we find the matches

@SQ	SN:scaffold_25_contig_4	LN:186
SRR203276.32744748	99	scaffold_25_contig_4	61	255	76M	=	108	123	AGGGGAAACTGTGGCGCCCAGGGGTTGAACTGACCTACCCTCAAAAATAGAGACTTTTGTAGGAGGCTGACCTTAG	############################################################################	XA:i:3	MD:Z:1T1A7A1C2T0G0G0C0C0T0T0C0G0C0T1C0T0A0T0T0A0T0G2A0A0A0A1T1T0C0A0T0C0C0A0C0A0A0G0A0C1G0C2G0C0A0A0G0C0G1T0A0C0G0T0	NM:i:55
SRR203276.32744748	147	scaffold_25_contig_4	108	255	76M	=	61	-123	TTCCCCCAACTGAACCCACGGTCAACCCGACTTTAAAAAACCAAGGCCCCAAACAAGTTGGCGGGATACACCCGAT	############################################################################	XA:i:3	MD:Z:0A3A1A1G0A0C1G0C0G0G0G0C0A0A1C0G0C0T0A1G0T1T0A0C0C0G0C0T0T0T0G0T0G0T0G0C1A2T0G0C1G1G0C0C2C0T11C0T0G0	NM:i:49

Note that these two matches involve the two pairs of the single read
SRR203276.32744748. The first match comes from sample_1.fq and the second
match comes from sample_2.fq (reverse complemented).

>scaffold_25_contig_4 scaffold_25 tid=9388 pos=672 length=2123 number_of_contigs=12
CTGACCCAGATGAGGTGGCCAGGAGATGGGGAAAGAGGAAAAACAAACCTAAGATGAATTATGAGAAACTGAGCCGTGGCCTTCGCTACTATTATGACAAAAATATCATCCACAAGACGGCGGGCAAGCGCTACGTATACCGCTTTGTGTGCGACCTGCAGAGCCTGCTGGGATACACCCCTGAAG

which has reverse complement

CTTCAGGGGTGTATCCCAGCAGGCTCTGCAGGTCGCACACAAAGCGGTATACGTAGCGCTTGCCCGCCGTCTTGTGGATGATATTTTTGTCATAATAGTAGCGAAGGCCACGGCTCAGTTTCTCATAATTCATCTTAGGTTTGTTTTTCCTCTTTCCCCATCTCCTGGCCACCTCATCTGGGTCAG

The paired reads SRR203276.32744748 are in some sense the closest pair of
reads to scaffold_25_contig_4, so one would think that they should have a good
alignment to this scaffold. (This is because if the scaffold was included by
collectContigs one would think that at least one read should really cover the
scaffold well.)

However, if we align the reads to this contig or its reverse complement using water,
we find that the alignment is not very good at all. E.g. try
water -asequence blah.a.fa -bsequence blah.b.fa -gapopen 10 -gapextend 0.5 -outfile /dev/stdout |less
or
water -asequence blah.a.fa -bsequence blah.b.fa -gapopen 100 -gapextend 0.5 -outfile /dev/stdout |less

For the left read (from sample_1.fq) it is better in terms of score to align to
the un-reverse complemented contig. With -gapopen 10 one gets

SRR203276.327     10 TGTGGCGCCCAGGGGTTG-------------AACTGACCTACCCTCAAAA     46
                     ||.||.|.|||||.|.||             |||..|||||...|.||..
scaffold_25_c     11 TGAGGTGGCCAGGAGATGGGGAAAGAGGAAAAACAAACCTAAGATGAATT     60

SRR203276.327     47 AT-AGAGACTTTTGTAGGAGGC-TGACCTTAG     76
                     || |||.|||       |||.| ||.||||.|
scaffold_25_c     61 ATGAGAAACT-------GAGCCGTGGCCTTCG     85

and with -gapopen 100 one gets

SRR203276.327     10 TGTGGCGCCCAGGGGTTG     27
                     ||.||.|.|||||.|.||
scaffold_25_c     11 TGAGGTGGCCAGGAGATG     28

It is similar for the right read.

----

We can also look at the BAM file produced by RSEM's expression computation of
each true isoform wrt the reads. This is the BAM file that extract alignment
information from needed by collectContigs. We do this using

samtools view ../expression.rsem-expr-workingdir/expression.sorted.bam|less

Note that the read SRR203276.32744748 doesn't appear anywhere in this output.

How could this happen? In a sense, it is not that surprising, because the
alignment of SRR203276.32744748 to scaffold_25_contig_4 is really bad.