Contents:

1. ORF Finder
   1. Algorithm
   2. Executing the ORF finder
   3. Output files
   4. Summary of results
2. viewSingleTCW - Viewing the ORFs
3. TCW ORF finder and Trinity TransDecoder
4. References

ORF Finder

runSingleTCW has an Option menu that allows the user to set the ORF options, which are shown on the right. This ORF finder algorithm was designed to find the 'best' ORF for input into runMultiTCW. However, a file of all the 'good' ORFs can be output (see Output files).

Figure 1: The runSingleTCW ORF finder options.

1. For each of the 6 frames, find the best ORF as follows:
   If there is a good hit with e-value < 1E-10 and similarity>20, use the hit frame.
      If the hit terminates with a Start and Stop codon, use the corresponding coordinates.
      Else compute all ORFs surrounding the hit and select the best based on the weight score.
   If there is not a good hit, find all possible ORFs and select the best using a weighted score.

2. To find the assigned ORF from the 6 ORFs found in Step 1, find the best ORF using the weighted score.

3. Weighted score
   • The weighted score uses: (a) Hit bitscore, similarity, and coverage, (b) length of ORF, (c) whether it has ATG and/or Stop, and (4) Markov score.
   • The weights are set so that a long ORF is selected over an ORF with a poor hit, and an ORF with a good hit is selected over a longer ORF. The Markov score and ATG/Stop do not have as much weight, but contribute to nudge it one way or another on close calls.
   • A full-length ORF with ATG/Stop takes precedence, regardless of hit and Markov score.

Markov training set

The Markov model is trained using either of following:

• Train with Best Hits. Minimum Set [default 200]
  1. The algorithm makes a list of sequences with a unique Uniprot ID and the top bitscore.
  2. It sorts the list on bitscore and take the top 2000 (or maximum sequences).
  3. It uses the sequence from hit-start to hit-end for the training set.
   
• Train with CDS file [default none]
  The file name of a FASTA file of nucleotide coding sequences.

   ./execAnno <project> -r -t 5000

Additional heuristics

• Multi-frame: The best hit frame will be used if it has e-value < 1E-10 and similarity>20, else the sequence is not assigned a hit for ORF finding.
   
• Stops in Hit: All possible ORFs are computed, and the selected one will be from those that cover the largest region of the hit.
   
• Nucleotide Hit is Best Bits: If the 'Best Anno' is a protein hit, it will be used.

Executing the ORF finder

   ./execAnno <project> -r

Table 2. Output files of the TCW ORF finder.

Example of AllGoodORFs.txt, where the first one listed is the selected ORF.

  ### Name    ntLen aaLen    Type   Frame Start..Stop     Markov  Codon
  tra_002_a    2160  720  5p-partial    3     3..2162     139.08  67.16   Hit: 3E-215  52%  84% (6..2150)
  tra_002_b     141   47  5p-partial    1  2269..2409       5.01   5.50

Example of bestORFs.scores.txt of the 6-frame scores for the best ORF:

  >tra_002   type:5p-partial seqLen:3111   orfLen:2160   ORF:3-2162(3)   Hit:2.94E-215,52%,84%
  Markov  139.08  -68.93  -26.47  -51.14  -25.33  -29.83

ORF Stats:   Average length 1272
  Has Hit            52,854 (99.0%)    Both Ends    40,855 (76.5%)    Multi-frame    12,988 (24.3%)
  Is Longest ORF     41,639 (78.0%)    ORF>=300     48,891 (91.5%)    Stops in Hit    7,849 (14.7%)
  Markov Best Score  46,046 (86.2%)    ORF=Hit      32,253 (60.4%)    >=9 Ns in ORF       0    (0%)
  All of the above   38,628 (72.3%)      with Ends  28,922 (54.2%)

Additional ORF info                   For seqs with hit   52,859 (99.0%) ORF=Hit with Ends    28,922 (54.2%)
  One End             52,328 (98.0%)   Both Ends          40,717 (77.0%)   ORF>=300           28,012 (96.9%)
  Markov Good Frame   47,700 (89.3%)   Markov Good Frame  47,398 (89.7%)   Markov Good Frame  27,792 (96.1%)
  ORF=Hit             32,253 (60.4%)   Markov Best Score  45,724 (86.5%)   Markov Best Score  27,353 (94.6%)
  ORF~Hit              7,488 (14.0%)   Is Longest ORF     41,329 (78.2%)   Is Longest ORF     23,328 (80.7%)
  ORF>Hit             11,827 (22.1%)   Longest & Markov   38,628 (73.1%)   Longest & Markov   22,776 (78.7%)
    with Ends          4,878  (9.1%)   Not hit frame                   5   Sim>=90            28,399 (98.2%)

  Frame: 3(30.5%)  2(29.9%)  1(34.3%)  -1( 1.9%)  -2( 1.6%)  -3( 1.9%)

  Both Ends:           Has Start and Stop codon
  ORF=Hit with ends:   ORF coordinates=Hit coordinates with ends
  Markov Best Score:   Best score from best ORF for each of 6 frames
  Markov Good Frame:   Score>0 and best score from 6 RFs of selected ORF

A. Basic Sequence Query for Remark B. Columns of the Sequence table

C. Sequence Details Frame view D. 6-frame scores for the displayed ORF

A. Basic Sequence Filter for TCW Remark: In this example, the remark "!Lg" was searched on, which shows the sequences with ORFs that are not the longest. All rows were selected and Seq Table shows them in the main Sequence table as shown in Figure 3B.

B. Columns of the Sequence table: This figure shows most of the ORF columns. A row can be selected followed by Seq Detail to view the frame as shown in Figure 3C.

C. Sequence Details Frame view: This example shows that the selected ORF is not the longest ORF, but has a hit with an E-value of 3.3E-98 and good Markov and Codon scores.

D. By toggling the ORFs/NT to Scores/AA, the upper region will show the 6-frame scores of the selected ORF.

TransDecoder: longest_orfs.cds.scores
 #acc    Markov_order    seq_length      score_1 score_2 score_3 score_4 score_5 score_6
 Osj_00011.p1    5       1047    83.19   0.06    15.85   27.04   26.83   5.91

TCW: bestORF.scores.txt
 >Osj_00011     type:complete   seqLen:1509   orfLen:1047   ORF:286-1332(1)   Hit:2.72E-79,45%,93%
  Markov   82.69    0.06   15.86   27.04   26.84    5.92

1. GeneID: Parra G, Blanco E, Guigo R (2000) GeneID in Drosophilia. Genome Research 10:511-515.
2. TransDecoder: Hass BJ, Papanicolaou A, Yassour M. et al. (2013) De novo transcript sequence reconstruction from RNA-seq using the Trinity platform for reference generation and analysis. Nature Protocols 8:1494-1512
3. Search programs used in TCW:
   1. BLAST: Altschul SF, Madden TL, Schaffer AA, Zhang J, Zhang Z, et al. (1997) Gapped BLAST and PSI-BLAST: a new generation of protein database search programs. Nucleic Acids Res 25: 3389-3402.
   2. DIAMOND: Benjamin Buchfink, Chao Xie & Daniel H. Huson, Fast and Sensitive Protein Alignment using DIAMOND, Nature Methods, 12, 59-60 (2015) doi:10.1038/nmeth.3176.