The University of Arizona
TCW Summary
  AGCoL | TCW Home | Doc Index | singleTCW Guide | DE Guide | multiTCW Guide | Tour
TCW provides two sets of programs: singleTCW for annotating and viewing single species transcriptions, and multiTCW for annotating and viewing comparative transcripts.

Contents:

Programs

bullet singleTCW contains four major programs:
1. runAS -- annotation setup for input to runSingleTCW
  • Downloads the UniProt taxonomic and/or full SwissProt and TrEMBL .dat files
  • Create fasta files from .dat files for searching against TCW sequences
  • Download the GO mysql database and augment it with UniProt information about GO, KEGG, Pfam, and InterPro
Note: runSingleTCW can take as input other databases such as Genbank nr, but these results will not have associated GO annotations.
2. runSingleTCW -- builds a single-species TCW database (sTCWdb)
  • Input: sequences and optional counts, where any of the following are valid ( the first three are NT-sTCWdbs and the last is a AA-sTCWdb):
    1. Load RNA-seq transcripts and count data with optional replicates.
    2. Load sequences with location information (e.g. predicted genes).
    3. Assemble up to ~1M sequences, such as: transcript sets, paired-end Sanger ESTs, or a mix of transcripts and ESTs.
    4. Load protein sequences and spectra (count) data with optional replicates.
  • Annotation:
    1. Annotate sequences with one or more nucleotide or protein databases (called annoDBs). UniProt should be downloaded with the runAS program. The searching may be done with the super-fast DIAMOND or the standard BLAST.
    2. If UniProt is used, GO annotations along with EC, KEGG and Pfam identifiers are extracted from the GO database and entered into the sTCWdb (The GO database is set up with the runAS).
    3. Compute ORFS and GC content.
  • All data and results are stored in a MySQL database.
3. runDE -- run Differential Expression analysis
  • An interface to several R packages (EdgeR, DESeq) for calculating differential of sequences. Additionally, it can execute a user-provided R script for DE calculation.
  • If the sTCWdb contains GOs, the GOseq R program can be used to compute differential GO terms.
4. viewSingleTCW -- view a single-species TCW database
  • Query and view the results. There are various filters, for example, filters are provided specific to taxonomic databases, trimmed GOs, filter by annotation, etc. The initial view is the Overview, which summarizes the results.
bullet multiTCW contains two major programs:
1. runMultiTCW -- builds a multi-species comparison TCW database (mTCWdb)
  • Builds a database from multiple sTCWdbs, which can be NT-sTCWdbs, AA-sTCWdbs or a mix. This has been tested with up to 5 input sTCWdbs.
  • Runs DIAMOND or BLAST to compare the sequences from the input TCWs. Add the resulting pairs to the database.
  • Clusters the pairs into ortholog groups. They can be clustered with TCW Closure, TCW BBH (best bi-directional), TCW shared hit, OrthoMCL, and/or user-supplied clusters can be uploaded. Multiple ortholog clustering can be in the database for query.
  • If the input is from two or more NT-sTCWdbs, coding statistics are calculated. Additionally, alignment files are output for input to KaKs_calculator, and the results of running the KaKs_calculator are input to runMultiTCW.
  • MSA (multiple sequence alignments) are computed for each cluster and scored with Trident.
2. viewMultiTCW -- view a multi-species TCW database
  • Query and view the results. The results can be filtered on various attributes. A cluster can be viewed graphically with the results of MAFFT or MUSCLE MSA or pairwise alignment (e.g. see graphical views MSA and Pair, respectively).

Machines tested on

Go to top
OSArchitecturePurchasedDatabaseJavaTest
Linux x86.64 (Centos 7) 3.2 Ghz AMD 24-Core, 128Gb2011MariaDB v10.4.12v1.8Build and View
Linux x86.64 (Centos 7) 2.0 Ghz AMD 4-Core, 20Gb2008MariaDB v5.5.60v1.7View
MacOS (Catalina 10.15.4)3.2 GHz Intel 6-Core i7, 64Gb2020MySQL v8.0.17v14.0Build and View
MacOS (Maverick 10.9.5) 2.4 Ghz Intel 2-Core i5, 16Gb2012MySQL v5.6.21v1.7View

Timings

Go to top

singleTCW

The timings were run on the MacOS (row #3) and Linux (row #1) of the above machine table. The "Mb" numbers are maximum memory (when available), and are approximate.
TaskCountMac MiniLinux
Load64,930 seqs 1m:26s 20Mb 2m:17s 19Mb
Instantiate64,930 seqs 47s 4Mb 1m:16s 1Mb
Diamond15G-6G file size 12m:09s --- 11m:54s ---
Annotate2461,118 unique hits 18m:14s 2374Mb 56m:7s 2384Mb
ORF64,930 ORFs 3m:54s 150Mb 9m:36s 246Mb
GO10,724 GOs 28m:42s 1030Mb 52m:04s 1350Mb
1TR-plants; The May2020 file (5G) was used on Mac, and the Oct2020 file (6G) was used on Linux, which influenced the increased times on Linux.
2SP-plants, TR-plants, SP-full; times do not include Diamond. Linux had 462,883 unique hits.

multiTCW

Using the same two computers as above, the mTCW database was built from a dataset of 28,392 and another of 26,685 sequences. Both datasets were annotated with SP-plants, TR-plants, and SP-full, where the Mac versions were downloaded May2020 and the Linux on Oct2020.
TaskCountMac MiniLinux
Build Database55,077 seq and 595,698 unique hits1  6m:28s 150Mb 9m:31s 152Mb
GO21,693 GOs 16m:48s 11Mb 30m:26s 506Mb
Add Pairs with Hits392,919 AA and 43,485 NT pairs 5m:03s 61Mb 15m:54s 68Mb
Add 3 cluster sets222,376 total clusters 2m:56s 33Mb 8m:10s 31Mb
Run Stats
PCC410,401 Pairs 1m:32s 68Mb 2m:16s 63Mb
Pair Stats95,628 Pairs to align 44m:33s 624Mb 2h:5m:19s 661Mb
MSA3 Stats22,376 clusters 1h:20m:11s --- 4h:34m:23s ---
1The Linux database had 606,845 unique hits.
2The methods added were BBH, Closure and BestHit. Closure used the most memory.
3This step uses the external program MAFFT.

The Linux times are much slower than the Mac, even though they comparable architectures; the Linux may not be optimized well and is a much older machine. It is important with MariaDB to make sure the variables are set right; execute ./runSingleTCW -v to print out some variables that seem to effect performance with MariaDB on Linux.

Linux only for TCW v3.0

The following are from the supplement 1 of Soderlund (2019). The singleTCW was 26,856 sequences. The UniProts were five SwissProt, the full SwissProt and five TrEMBL databases (but not the full). The multiTCW was created from three sTCWdbs for a total of 198,702 sequences. There is not much speed difference between TCW v3 and later versions.
ProgramStepTimeNote
runAS Download and create UniProt FASTA files 5h:33m:24s a
  Download SwissProt and create subset FASTA file 4m:34s a
  Download and build GO database 8h:59m:55s b
runSingleTCW Build sTCW_rhi_NnR of 26,685 transcripts 3m:13s
  Annotate sequences 2h:15m:22s c
  ORF-finding 6m:09s
  Add GO annotations 50m:26s
runDE edgeR for 12 condition pairs 0m:20s d
  GOseq for each of the 12 DE results 10m:56s d
runMultiTCW Build mTCW_rhi of 198,702 transcripts 20m:56s
  Add GO annotation for 23,516 unique GOs 56m:02s
  Add 1,099,986 pairs after searching 28m:20s e
  Build Clusters (total) 22m:00s f
        BBH (OlR-Os) :04s
        Closure :08s
        OrthoMCL 8m:04s
  Compute PCC for 1,099,986 pairs 3m:37s
  Pair-align and analyze 354,280 pairs 2h:27m:08s
  Multi-align and analyze 71,339 clusters 12h:14m:43s g
  Add KaKs values 1m:1s h

a There can be considerable variation in download times.
b The bulk of the time was from loading uniprot_trembl_bacterial.dat.gz, which took 6h:56m:58s; the compressed file was 57G and contained 96,592,456 entries.
c This did not include the DIAMOND formatting but did include the time for searching; the longest time for searching was against TrEMBL bacterial, which took 35m:20s.
d The time for executing the R code and loading the results.
e The time does not include searching all sequences against themselves. The protein self-comparison took DIAMOND 3m22s and nucleotides self-search took Blastn 5m:13s.
f The time for computing and adding all clusters; the majority of the time was for loading the clusters into the database. The following three individual cluster times are only for computing the clusters.
g This includes the time to run MAFFT in parallel on each cluster.
h This does not include the time to execute the KaKs_Calculator, which is run by the user before the add.

References

Go to top

C. Soderlund (2019) Transcriptome computational workbench (TCW): analysis of single and comparative transcriptomes.
BioRxiv
Describes the TCW v3 package.

C. Soderlund, W. Nelson and S. Goff (2014) Allele Workbench: transcriptome pipeline and interactive graphics for allele-specific expression.
PLOS ONE
Describes a pipeline that can be used with TCW, plus a new GO trim algorithm.

C. Soderlund, W. Nelson, M. Willer and D. Gang (2013) TCW: Transcriptome Computational Workbench.
PLOS ONE
Describes the TCW v1 package.

C. Soderlund, E. Johnson, M. Bomhoff, and A. Descour (2009) PAVE: Program for Assembling and Viewing ESTs.
BMC Genomics
Describes the assembly algorithm.


Email Comments To: tcw@agcol.arizona.edu