Maximum bits in ggseqlogo is less then 2?

Question

[Mattan]

Hi all, I'm using this code as a toy example to see if I understand how ggseqlogo works

# Create a data frame with the sequences
sequences <- c("ACA","ATA","AGA","AAA","ATA","AGA")
df <- data.frame(Sequence = sequences)

# Plot the sequence logo
ggseqlogo(df$Sequence)

I fail to understand the results I'm getting.

Why isn't the maximuum 2 bits for positions 0,2? Also, why is the middle poistion (1) isn't populated? We do know that G and T are more common there.

Answer 1

[Mensur Dlakic]

Most sequence logo implementations have a small sample correction by default. This is easy to test by using 15-20 sequences, in which case the height of terminal residues should be closer to 2 bits and the middle stack might show up. This is what I get after increasing the number of sequences to ~100.

You may need to turn the small sample correction off.

Answer 2

[Alex Reynolds]

The original paper describes the information at a position as:

Information(l) = 2 - (H(l) + e(n))

Here, Information(l) is the value along the y-axis in your plots. H(l) is the uncertainty at position l (which decreases when you have more of the same bases at that position in the sequence), while e(n) is a correction factor required when the number n of sample sequences is small. That correction factor is apparently described in this paper but it is also defined in the source code for ggseqlogo.

Answer 3

[bk11]

Did you check with method="prob" too?

ggseqlogo(df$Sequence, method="prob")