================================
NIAID Large Internal Delete (LID) HIV Genomes

- GOAL: PacBio sequence HIV genome mixtures that have been isolated by
  SGA that have shorter lengths due to large internal deletions (LIDs)
  and compare results to MiSeq results so that algorithms can be
  perfected. We look forward to seeing the unblinded MiSeq estimates
  for the runs we present here.

- Here is the sample list of the shorter genomes sent to PacBio:


Name                    Samples                         AvgEstimatedSize
----                    -------                         ----------------
Pt3-95-3mix-A-short     1995: s2, s19, s20,             2200
Pt3-95-5mix-B-short     1995: s1, s3, s8, s10, s11      3300    
Pt3-95-6mix-C-short     1995: s4,s12,s13,s14,s16,s18    3400       
Pt3-95-6mix-D-short     1995: s5, s6, s7, s9, s15, s17  4700    
Pt3-01-3mix-E-short     2001: s29, s30, s31             2400
Pt3-01-4mix-F-short     2001: s21, s23, s27, s32        3300    
Pt3-01-5mix-G-short     2001: s22, s24, s25, s26, s28   3400    


- For this set of results, we give consensus genonomes for the
  Pt3-95-3mix-A-short (3 expected genomes) and Pt3-95-3mix-D-short (6
  expected genomes) runs. Below we give 4 download links applying two
  algorithms to these two runs.

- By comparing the results against the unblinded MiSeq estimates, we
  will be able to iterate the algorithms to get them perfect. We will
  start with two runs and then be able to use the other runs as new
  "test sets" to make sure our algorithms don't overfit. After getting
  everything perfect, the goal is to apply the final algorithms to the
  20-mix of full length HIV genomes as the final validation test set.



================================
Consensus Methods

- We try two classes of algorithms for these results: Long Amplicon
  Analysis (LAA) and ClusteringConsensus (CluCon).

-- Long Amplicon Analysis (LAA) is a de-novo method that works to
  cluster out more widely divergent genes and then finds smaller scale
  differences like SNPs within the genes. It was originally developed
  for multi-gene HLA sequencing.

- LAA was run using the standard analysis protocol in SMRTportal.

-- ClusteringConsensus (CluCon) is a reference based approach that
  clusters out different genomes and then mitigates noise using
  consensus within each clustering in a recursive fashion.

- CluCon was run using a development version of the code available at
  https://github.com/mpsbpbi/clusteringConsensus

- The development version works to handle PacBio error modes to
  achieve better performance.

- An HIV HXB2 reference was used with the middle deleted. This was
  necessary because the deletions were so large, the alignment was
  broken into two separate parts.



================================
LAA Consensus Genomes

------------
- LAA results for the Pt3-95-3mix-A-short run:


Sequence Cluster	Sequence Phase	Length (bp)	Estimated Accuracy	Subreads coverage
Cluster0	Phase0	2,238	99.96	298 1
Cluster0	Phase1	2,234	99.76	202 2
Cluster1	Phase0	2,818	100.00	500 3
Cluster2	Phase0	1,628	100.00	328 4
Cluster3	Phase0	831	99.71	54
Cluster4	Phase0	1	100.00	37
Cluster5	Phase0	475	99.54	23
Cluster6	Phase0	1,387	96.07	22


  Remove any results with less than 200 coverage leaving four.

- RESULT:Here are four consensus geonomes from the
  Pt3-95-3mix-A-short run: laa-A-amplicon_analysis.clean.fasta

- From the runsheet, we expected three genomes but the algorithm found
  four with high coverage.

------------
- LAA results for the Pt3-95-3mix-D-short run:


Sequence Cluster	Sequence Phase	Length (bp)	Estimated Accuracy	Subreads coverage
Cluster0	Phase0	4,234	100.00	500 1
Cluster10	Phase0	3,185	99.87	109
Cluster11	Phase0	121	99.25	89
Cluster12	Phase0	2,081	99.74	87
Cluster13	Phase0	2,999	100.00	81
Cluster14	Phase0	2,819	99.90	36
Cluster1	Phase0	2,297	95.40	53
Cluster1	Phase1	2,260	0.05	4
Cluster1	Phase2	2,296	99.57	76
Cluster1	Phase3	2,297	99.45	63
Cluster1	Phase4	2,298	0.05	4
Cluster1	Phase5	2,282	99.60	62
Cluster1	Phase6	2,282	98.66	72
Cluster1	Phase7	2,190	9.22	3
Cluster1	Phase8	2,278	99.98	162
Cluster2	Phase0	4,695	100.00	500 2
Cluster3	Phase0	5,201	100.00	422 3
Cluster4	Phase0	4,698	100.00	341 4
Cluster5	Phase0	5,019	100.00	285 5
Cluster6	Phase0	5,012	100.00	234 6
Cluster7	Phase0	2,705	100.00	142
Cluster8	Phase0	2,472	100.00	136
Cluster9	Phase0	2,933	100.00	115


  Remove any results with less than 200 coverage leaving six.

- RESULT:Here are six consensus geonomes from the
  Pt3-95-3mix-D-short run: laa-D-amplicon_analysis.clean.fasta

- For completeness, here are all consensus geonomes from the
  Pt3-95-3mix-D-short run: laa-D-amplicon_analysis.fasta (superset of above)

- From the runsheet, we expected six genomes and the algorithm found
  six with high coverage.

================================
CluCon Consensus Genomes

------------

- RESULT:CluCon consensus genomes for the Pt3-95-3mix-A-short
run: short-mixes-A-clucons.fasta

There are 3 estimated genomes with expected 3.

len  name                       VarCols
--- -----                       -------
2767 short-mixes-A-clucons-num0 84
2771 short-mixes-A-clucons-num1 128
2781 short-mixes-A-clucons-num2 142

(VarCols is the number of columns that are estimated to be
variant. Should be low for pure subsets.)

------------

- RESULT:CluCon consensus genomes for the Pt3-95-3mix-D-short
run: short-mixes-D-clucons.fasta

There are 7 estimated genomes with expected 6.

len  name                       VarCols
--- -----                       -------
5064 short-mixes-D-clucons-num0 0
4667 short-mixes-D-clucons-num1 1
4668 short-mixes-D-clucons-num2 2
4204 short-mixes-D-clucons-num3 33
4994 short-mixes-D-clucons-num4 36
4880 short-mixes-D-clucons-num5 60
4612 short-mixes-D-clucons-num6 265

(VarCols is the number of columns that are estimated to be
variant. Should be low for pure subsets.)