1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
|
Note on IPD classifier training conventions
===========================================
Getting the strand and context correct is notoriously confusing.
In the output of the basemods package (kineticsTools), strand refers the
_template_ strand, because that's where the modification that we're detecting
are. The PacBio basecaller and dye/base mapping always work in the product strand.
The IPD model itself is trained to take sequences in the template strand and make predictions about
the IPD that will observed when synthesizing the product strand. In KEC, we will be working entirely
with the product strand sequence. We set up the Kinetic Model code to accept product strand sequences and give
product strand predictions.
Here's a reference for the context windows and strands:
Case 1: the + strand is the product strand:
- The input sequence to pass to the classfier is the - strand sequence, indexed according to the top numbers
- The IPD GBM model prediction is for the G incorporation in the product strand synthesis
Case 2: the - strand is the product strand:
- The input sequence to pass to the classfier is the + strand sequence, indexed according to the bottom numbers
- The IPD GBM model prediction for the C incorporation in the product strand synthesis
strand sequence pol motion
14 4 0
| | |
- 3'-xxxxxxxxxNNNNNNNNNNCNNNNxxxxx-5' <-
+ 5'-xxxxxxxxxNNNNNNNNNNGNNNNxxxxx-3' ->
| | |
0 10 14
|
