Open issues

[yhoogstrate]

• Refactor extract_subnetworks
• Figure out random result glitch (somewhere there is an unsorted yield or loop)

Algorithmic challenges:

• Fix alignment
• Data structure / inserting the Graph
• Pruning
  • Is double insert size currently possible? - if so, reduce it to the sum - fixed in 9679d29
• Re-joining splice junctions
  • Currently they are first inserted into the main Graph, extracted after pruning and re-inserted in a second system. It makes sense to put them in a Graph in the first place, because for pruning it is necessary to use a genomic index. Suggested solution: use 2 graphs (one for SJ and one for other edges)
• Extract subnetworks
  • Rename to something that includes by splice junctions
• Merge overlapping subnetworks
  • Disable weird rnodes code and see if this can be taken up by next step(s)
• Filtering / classification
  • Add more parameters, do data mining?
  • Distinguish by pruned-by-SJ and normal pruned: normal pruned and only discordant is a very rich source of information (S027) has 15 disco reads pruned normally
• Optional: try to align intronic and exonic bp
• Output:
  • is_circular
  • sort by score
  • genome dist
  • Number of splice junctions
  • Test if merge_subnets also contributes to n_nodes and n_edges
  • disco / split read ratio
• Use consistent variable for insert size
• Figure out what happen in S055
  • Add offsets to all discordant_paired_ types
  • Add weight for _paired_s
• Make real bi-directional edges (should reduce mem as well)
  • Start with three graphs: 1 positive, 1 negative strand and one for SJs (uni-strand)
• Merge all the HTSeq indextree iterator functions into a few generic ones
• Classification on discordant reads seem to fail if there is a junction from a starting exon (because transcription effectively starts there) - these are often intronic and it is plausible other sub-graphs to merge them with are present