Exploring removing path from pandora index and on-demanding loading of PRGs

[leoisl]

Removing path from pandora indexes don't improve much the RAM usage:

• Commit: 6569d90
• RAM usage reduction: from 15.7GB to 13.4GB (15% reduction);
• Results are not identical neither, it is hard to compare, but a quick comparison shows this:
  • DB size: 188k PRGs
  • Nb samples: 114
  • Nb of PRGs in final pandora_multisample.matrix BEFORE optimisation: 13716
  • Nb of PRGs in final pandora_multisample.matrix AFTER optimisation: 13756
  • Nb of common lines in pandora_multisample.matrix (i.e. where both versions found same presence/absence pattern): 9356
  • Nb of diff lines in pandora_multisample.matrix: 4400

I don't think is worth it to try removing paths from pandora indexes right now. Note that 13k genes were mapped to out of 188k. The on-demand PRG loading will be where we will really save RAM. Maybe later we can remove path from the index..