filter out low coverage event

[jphe]

Hi, Yuanhua,

I'm using BRIE to calculate Psi for each event, and then perform different AS detection with △Psi>0.05 and P<0.01 between two different group cells. But many of significant different AS event I found are very low expressed, even though their Psi very significantly different (~0.01 vs ~0.9) between two groups. The Psi value is extracted from fractions.tsv output by brie, but I noticed there are many event that their counts are zero, but the Psi are very high or very low, and their Psi_low and Psi_high are quite different ( the range even nearly 0 to 1). So why their Psi are such different? and how could I filter out the such low coverage AS event? just according to the counts value from fractions.tsv file is OK?

Thanks,

[huangyh09]

Hi,

It might be less reliable if you use events with very limited or even no reads. You can use the number of reads as a filtering, e.g., n_read>=3. Alternatively, I prefer to use the confident interval, e.g., 95% interval < 0.25, which you can obtain from the fractions.tsv file (the 7th and 8th column).

Also, I have some notes on comparing two cell groups: https://github.com/huangyh09/brie/issues/19#issuecomment-424467215

Cheers, Yuanhua