proposed experiments & analyses

[gfudenberg]

pre-drafting todos

genomic deletions of CTCFs at boundaries:

• model robustness:
  • [x] #74
  • [x] #75
• correlate SCD with:
  • [x] GC content
  • [x] w/ DNA shape done
  • [x] #66
  • [ ] w/motif methylation level
• motif discovery in flanking regions:
  • [x] First pass shows not many informative motifs

virtual insertions (into flat backgrounds):

• technical aspects of virtual insertions

  • [x] background sequence generation: 8mer shuffling is optimal. Lower GC content are easier to flatten.
  • [x] #77
  • [x] #67

• flanking DNA around CTCF

  • [x] vary amount of genomic flanking DNA around CTCF. Result: ~15-20bp is the typical amount of high impact DNA
  • [x] vary flanking sequence around a fixed core motif, related to comment on issue 28 -> 52
  • [x] non-symmetric flanks: vary amount of upstream or downstream genomic flanking DNA around a CTCF. Not important enough to include in first draft.
  • [x] #68
  • [x] #72
  • [x] #76

• multi-CTCF motif grammar

  • [x] vary orientation of CTCFs.
  • [x] #28
  • [x] #70
  • [x] #69
  • [x] #71

• influence of nearby elements on CTCF insertions

  • [ ] insert a promoter in the virtual sequence & see if it modulates boundary strength
  • follow-up: if this has an impact, is there any difference between promoters based on their expression in different cell types
  • [ ] insert heterochromatic-associated sequence & see how much is needed to silence one of our virtual insertions.
  • [ ] influence of nearby motifs on CTCF SCD/insulation/dot-scores

• cell-type and organismal specificity

  • [ ] cell type specificity (aka differences between targets) of virtual insertions vs. genomic deletions.
  • [ ] organismal specificity (aka difference between mouse & human head for ESC) of virtual insertions of CTCF (and maybe something else)

• open Q: how to compare absolute values for insertion scores across models and celltypes?

these are in part from the brainstorm on aug22, 2022 updated 11/28/22

[gfudenberg]

another thing to consider: it may make sense to focus on ESC output for a screen that deletes ESC boundaries (and could use that to also focus on ESC output for insertions). So far I've been averaging across outputs, but the boundaries were called from the Bonev mESC data.

[gfudenberg]

one fun suggestion from Irina Solovei at the conference was to insert various lengths of heterochromatic-associated sequence & see how much is needed to silence one of our virtual insertions

[PSmaruj]

Is there any smart way to get boundaries of the heterochromatic-associated sequences? Are there any successful experiments that we could recreate virtually and compare results?

[gfudenberg]

Yeah, those would both be challenges! Some places I might start:

• trying random ~200kb sequences from the inactive "B" region of the genome
• scanning the genome in 1kb steps for regions that are predicted to be relatively flat and inserting random subsequences from those I think if we saw something this could motivate new experiments! E.g. by the Giorgetti group who recently published this paper https://www.nature.com/articles/s41586-022-04570-y

[gfudenberg]

11/30/22 discussion:

• compatibility between left flank, right flank, core?
  • could be interesting to sort by strength
• pairwise mutagenesis ?

[gfudenberg]

further future experiments

other genomic deletions:

• genomic saturation mutagenesis around CTCFs:
  • [ ] get a directional score for mutation around strong/weak CTCF sites (e.g. like insulation), because SCD wouldn't tell you if the site got stronger. builds on akita fig4a-e.
• genomic mutagenesis of jet/stripe/flames: e.g. jets or stripes from stripenn. first mutate some hand-chosen examples. might need some different scores. Elphege & Nora group members have some example flames that could be a good place to start.
• genomic dot mutagenesis
• genomic bin-wise deletions:
  • [ ] generate combined human+mouse deletion profiles at either 256bp or ~1kb.
  • which cell type predictions are most correlated across species?
  • in variable regions, what are the features? in similar regions what are the features? are they more evolutionarily conserved?

genomic insertions:

• [ ] how correlated are these with genomic deletions/virtual insertions?
• [ ] how variable is this score depending on genomic insertion location?
  • e.g. depend on distance to existing boundaries?
• [ ] how sensitive are different types of inserted boundaries? or motifs with different amount of flanking DNA?
  • for oriented insertions, how different can this score be for forward vs. reverse orientation?
  • how about at specific genomic locations:
    • [ ] sox2 (relates to work in van Steensel lab)
    • [ ] Zuin et al locus
    • [ ] Karissa (from Nora lab) locus