license: openrail
DEEPBIND v0.11
The deepbind command-line executable can be used to score DNA/RNA sequences according to any RBP/TF model listed in the DeepBind web repository:
http://tools.genes.toronto.edu/deepbind
For each input sequence, the deepbind executable scores each subsequence of a pre-determined length (e.g. 20) and returns only the maximum or the average over these per-position scores.
Larger scores indicated stronger binding. The scores themselves are on an arbitrary scale, and vary from model to model due to variation in the quality of training data for different proteins.
EXAMPLE
To generate predictions with DeepBind, you need two things:
- a list of model IDs, and
- a list of DNA/RNA sequences.
The file example.ids contains 4 example model IDs, one on each line, reproduced here:
- D00210.001 # RBFOX1 (RNAcompete)
- D00120.001 # MBNL1 (RNAcompete)
- D00410.003 # GATA3 (SELEX)
- D00328.003 # CTCF (SELEX)
The file example.seq contains 4 example sequences, which were chosen such that the nth sequence scores highly for the nth model. The file example.seq is reproduced here:
- AGGUAAUAAUUUGCAUGAAAUAACUUGGAGAGGAUAGC
- AGACAGAGCUUCCAUCAGCGCUAGCAGCAGAGACCAUU
- GAGGTTACGCGGCAAGATAA
- TACCACTAGGGGGCGCCACC
To generate 16 predictions (4 models, 4 sequences), run the deepbind executable as follows:
% deepbind example.ids < example.seq
| D00210.001 | D00120.001 | D00410.003 | D00328.003 |
|---|---|---|---|
| 7.451420 | -0.166146 | -0.408751 | -0.026180 |
| -0.155398 | 4.113817 | 0.516956 | -0.248167 |
| -0.140683 | 0.181295 | 5.885349 | -0.026180 |
| -0.174985 | -0.152521 | -0.379695 | 17.682623 |
To see details of each ID, use the --dump-info flag:
% deepbind --dump-info example.ids
| ID | Protein | Type | Species | Family | Class Experiment |
|---|---|---|---|---|---|
| D00210.001 | RBFOX1 | RBP | Homo sapiens | RRM | RNAcompete |
| D00120.001 | MBNL1 | RBP | Homo sapiens | Znf | RNAcompete |
| D00410.003 | GATA3 | TF | Homo sapiens | GATA | SELEX |
| D00328.003 | CTCF | TF | Homo sapiens | C2H2 ZF | SELEX |
CHANGES v0.1 -> v0.11
Fixed bug where last position in input sequence was not evaluated for a score; suggested by Irene Kaplow.
Added --window-size and --average flags based on feedback.