Daniel:Notebook/GenomeMiner/2013-9-18
Mock HL155 (Started 9/9/2013)[edit]
Talked over with Matt today about the goals of the Mock HL155 project. The major goal is to test irregularities seen in the bowtie2 alignment data and check them using mock sequencing data, which we can control better. These irregularities include:
- Specific sites where errors more frequently occur
- General questions about bowtie's ability to accurately count the errors (is it off by a little bit?)
- The dependence of bowtie alignment on the orientation of the reference strand (See 9/3/13)
- The dependence of bowtie alignment on the length of the reference strand (See 9/3/13)
We've checked up, to an extent, on the site locations and the general questions about error count (although there are more tests I'd like to run). However, we should also check on the dependence of bowtie alignment on the reference strand length/orientation. To test this we will use four reference strands:
1. Original strand, full length (98 bp) 2. Reverse complement, full length (98 bp) 3. Original strand, read length (50 bp) 4. Reverse complement, read length (50 bp)
Additional testing (goal #2):
- Use perfect base quality scores on mimic data
Mimic Data, Perfect Base Quality[edit]
Matt has discovered that the first and last bases of a read generally have higher error rates than the middle 40. To see if this is base-quality derived, I'm generating new reads with mimic style error (1.01% sub, 0.26% ins, 0.05% deletions)
Matlab Error Counts
628149 substitutions (0.42 pct) 161459 insertions (0.11 pct) 31202 deletions (0.02 pct)
Alignment Results[edit]
2959000 reads; of these: 2959000 (100.00%) were unpaired; of these: 11781 (0.40%) aligned 0 times 2941663 (99.41%) aligned exactly 1 time 5556 (0.19%) aligned >1 times 99.60% overall alignment rate
Error Counting Results[edit]
Error Rate: 0.526% Error Rate of Insertions: 0.091% Error Rate of Deletions: 1.82e-04% Error Rate of Substitutions: 0.435%
Testing the Reference File[edit]
Using the fastq file generated for the mimic data. The reference files are as described above. The mimic data results are the data for reverse complement, read length (4).
Workflow[edit]
1. Copy hl155bash.sh, fasta file, 2. bowtie2-build -f fasta file fastaoutputname 3. ./hl155bash.sh 4. perl imp_count_mismatch.plx
Alignment Results[edit]
Original Strand, Full (1)
2959000 reads; of these: 2959000 (100.00%) were unpaired; of these: 106386 (3.60%) aligned 0 times 2843081 (96.08%) aligned exactly 1 time 9533 (0.32%) aligned >1 times 96.40% overall alignment rate
Reverse Complement Strand, Full (2)
2959000 reads; of these: 2959000 (100.00%) were unpaired; of these: 106404 (3.60%) aligned 0 times 2843063 (96.08%) aligned exactly 1 time 9533 (0.32%) aligned >1 times 96.40% overall alignment rate
Original Strand, 50bp (3)
2959000 reads; of these: 2959000 (100.00%) were unpaired; of these: 132674 (4.48%) aligned 0 times 2821187 (95.34%) aligned exactly 1 time 5139 (0.17%) aligned >1 times 95.52% overall alignment rate
Error Counting Results[edit]
Original Strand, Full (1)
Error Rate: 1.126% Error Rate of Insertions: 0.188% Error Rate of Deletions: 0.028% Error Rate of Substitutions: 0.910%
Reverse Complement Strand, Full (2)
Error Rate: 1.126% Error Rate of Insertions: 0.188% Error Rate of Deletions: 0.028% Error Rate of Substitutions: 0.910%
- Makes any difference? error:
Line 1748162, sequence length 43 vs 50 from CIGAR Parse error at line 1748162: CIGAR and sequence length are inconsistent
Original Strand, 50bp (3)
Error Rate: 1.082% Error Rate of Insertions: 0.195% Error Rate of Deletions: 5.48e-04% Error Rate of Substitutions: 0.887%
So interestingly, the error rates for the full strand original and reverse complement are the same (what we would expect), but they differ slightly from the results from the trimmed references (which are also the same as each other). Deletions especially seem to be underestimated when the reference strand is only 50bp, although insertions were called slightly better.