Summary of feedback from BOG, SOBP and SAGES

[lcolladotor]

Here's a summary of the feedback I remember from those 3 conferences. I know that I'm missing a few more comments, but I'm not recalling them right now.

• Explain more slowly what are the samples
• Highlight when we are working only with non-psychiatric controls
• Highlight that the SCZD vs controls analysis is with age > 17
• Maybe show a plot of PMI by diagnosis status in the supp material
• Some confusion between brain region and ERs for the degradation part
• The scatter plot for t-stats between the DLPFC results and the BSP1 DLPFC results has ok correlation (0.8) but there are still a few that don't seem DE in BSP1 (near 0 on the Y-axis). Maybe there's another plot we can make to show that it's looking ok between the 2 phases. Maybe a concordance at the top plot?
• Are any of the 25 and 11 genes common in the venn diagram of special interest?
• Split the venn diagram for diagnosis DE by region into 2 or use upSetR.
• Maybe plot one of the region dependent eQTLs, to show how the relationship changes between brain regions. Maybe there are some that have different trends (positive vs negative) instead of different angles in the slope.
• Conclusions on the development model feel a bit weak to me.

Some questions:

• Why are we not seeing a lot of overlap between gene/exon/exon-exon junctions? I mentioned that each has different cutoff filters and also resolutions (like number of reads that could overlap them). Should we focus on those that are present in at least 2 feature levels? Though well, a gene could be DE without its exons being DE (if the gene is not highly expressed and some exons don't have enough data to pass the expr filter). Maybe I could make a venn diagram with all features > expr cutoff, and then see if the proportions in the final diagrams are similar or not.
• What if a prenatal/child sample ended up developing schizophrenia later in life? Here I'm not sure if there's a part of the process that Amy and Joel do that could help resolve this.
• Do we expect degradation to affect the same regions of the genome consistently? What are the properties of those regions? From Andrew: check some of the original qSVA paper. Plus the lack of degradation * region interaction degradation ERs.
• What about differential eQTLs by diagnosis status for each brain region?
• Volcano plots for the region-specific model by BrainSpan replication status: someone was confused as to why there were so many points in the genes (more than in exons), but it was the other way around. In any case, probably I need to highlight better which panel is which.

[lcolladotor]

Other recent todos:

• random intercept

• compare p-values for some SNPs that are and aren't eQTLs

• scale interaction p-values?

• gene level duplication correlation age > 13 (match samples)

• take a random 20k SNP subset (keep all p-values)

• KEGG GO for paper

[lcolladotor]

We used some of this feedback in recent months.