Monday notes

Over Thanksgiving break, completed running Trinity on digitally-normalized reads. Script here.

Comparison of Trinity transcriptome assemblies

Assembly	Total trinity transcripts	Total trinity components	Contig N50
Trinity-full	126172	96093	1631
Trinity-diginorm	143078	88080	2047

The diginorm assembly has more total transcripts, but fewer components. Interesting…

Trying to evaluate similiary among assemblies using BLAST.

Damn - ran into memory limits with R!

Error: C stack usage is too close to the limit

Increase limit using

 ulimit -s 16384 # enlarge stack limit to 16 megs
 # check R memory limit
 R --slave -e 'Cstack_info()["size"]'

Better solutions:

simply read as table…no memory problems. wonder what read.blast() does that causes these problems
run blast with flags to keep only matches with significant evals and only keep top hit

blastn … -max_target_seqs 1 -evalue 1e-20

This reduces blast results table from 681k lines to 115k

Even better

Use Trinity provided script to compare assemblies

# Make BLAST database for diginorm assembly
makeblastdb -dbtype nucl -in ../results/trinity/Trinity.fasta

# BLAST full assembly against diginorm assembly
blastn -query ../results/trinity-full/Trinity.fasta -db ../results/trinity/Trinity.fasta -evalue 1e-20 -max_target_seqs 1 -outfmt 6 -out blast-full-vs-diginorm.txt

# Examine percent of diginorm target length aligned to best matching transcript in full assembly
/opt/software/trinityrnaseq_r2013_08_14/util/analyze_blastPlus_topHit_coverage.pl blast-full-vs-diginorm.txt ../results/trinity/Trinity.fasta ../results/trinity/Trinity.fasta

Table of percent length of diginorm assembly transcripts aligned by full assembly transcripts

hit_pct_cov_bin	count_in_bin	>bin_below
100	73880	73880
90	5925	79805
80	5382	85187
70	3800	88987
60	2989	91976
50	1607	93583
40	1157	94740
30	1060	95800
20	1053	96853
10	0	96853
0	0	96853

Of 143,078 transcripts in diginorm assembly, 96,853 (68%) have significant matches to transcripts in full assembly
73,880 (76%) are 100 percent matches
85,187 (88%) have greater than 80% length alignment

Table of percent length of full assembly transcripts aligned by diginorm assembly transcripts

hit_pct_cov_bin	count_in_bin	>bin_below
100	80308	80308
90	4381	84689
80	3277	87966
70	2664	90630
60	2189	92819
50	1218	94037
40	828	94865
30	554	95419
20	390	95809
10	0	95809
0	0	98509

Of 126,172 transcripts in full assembly, 98509 (78%) have matches in diginorm assembly
80,308 (82%) are 100 percent matches
87,966 (89%) have greater than 80% length alignment

NOTE the diginorm assembly includes 100 in silico spike-in sequences that do not exist in full assembly…plan to use these to computationally check assembly. need to remove for fair comparison