(Burak: I did not keep detailed minutes; Here are a bunch of notes I scribbled. Feel free to edit. See Slides for figures.)

Bottomline:

• Different callers give very different results.
• Callers vary by sizes of domain called and number vs. sequencing depth.

Jennifer Cremins:

• Different classes of callers may need to be considered separately
• Different tools have parameters that allow studying different length scales. A fair comparison needs to sweep parameters.

Erez:\__

• With 10M or so reads, I don't think you have power to call TADs. So, the calls (reproducible as they may be) may not be accurate calls.
• Using the 3Billion deep biological replicate from Rao et al would be more appropriate.
• We should study consistency vs. read depth (compared to the deepest sample)
• A visual inspection would be valuable (we could use juicer 2D format for different TADs)

Different enzymes or different papers also give very different results. Jakkard index of 0-0.2 depending on the tool.

• (even mboI vs hindIII)
• also IMR90 Rao vs Jin