Skip to content

Latest commit

 

History

History
118 lines (82 loc) · 3.23 KB

README.md

File metadata and controls

118 lines (82 loc) · 3.23 KB
Raw

RBPNet: Sequence-to-Signal Learning for CLIP-Seq Data

Install:

pip install git+https://github.com/mhorlacher/rbpnet.git

Usage:

Training

Training is performed in three steps. First, a dataspec.yml file is created which defines tasks and assigns signal bigWig and input regions (e.g. called peaks). Using the dataspec.yml, a training dataset in form of TFRecords is generated. Finally, training is performed with the rbpnet train command.

1. Creating a dataspec.yml

For an example, check out examples/QKI.dataspec.yml.

2. Creating a TFRecord file

Use the rbpnet tfrecord command to create a train.tfrecord file.

Usage: rbpnet tfrecord [OPTIONS] DATASPEC

Options:
  -o, --output TEXT
  -w, --window-size INTEGER
  -s, --shuffle INTEGER
  -o, --output TEXT
  --help                     Show this message and exit.

3. Training a RBPNet model

Use the TFRecord file and the rbpnet train command to train a RBPNet model. While the default parameters are optimized to work well on many eCLIP and iCLIP datasets, individual model and training parameters can be overwritten via a config.gin file. See examples/config.gin for an example.

After training, a model.h5 file can be found in the output directory.

Usage: rbpnet train [OPTIONS] [TRAIN_DATA]...

  Train an RBPNet model.

Options:
  --validation-data TEXT
  -c, --config TEXT
  -d, --dataspec TEXT
  -o, --output TEXT
  --help                  Show this message and exit.

Prediction

Takes as input a FASTA file, model and optional target layer. Returns a FASTA file with additional, nucleotide-wise prediction tracks for each sequence and task.

Usage: rbpnet predict [OPTIONS] FASTA

Options:
  -m, --model TEXT
  --layer TEXT      Target layer, i.e. 'profile', 'profile_target' or 'profile_control'.

Feature Importance / Attribution

Takes as input a FASTA file, model and optional target layer. Returns a FASTA file with additional, nucleotide-wise attribution scores for each sequence and task.

Usage: rbpnet explain [OPTIONS] FASTA

Options:
  -m, --model TEXT
  -o, --output TEXT
  --layer TEXT       Target layer, i.e. 'profile', 'profile_target' or 'profile_control'.

Evaluate

Takes as input a TFRecord file, dataspec, model and optional target layer. Prints TSV-formatted SCC scores between predictions and true counts.

Usage: rbpnet evaluate [OPTIONS] TFRECORD

Options:
  -m, --model TEXT
  -d, --dataspec TEXT
  --layer TEXT         Target layer, i.e. 'profile' or 'profile_control'.

Variant Impact Prediction

Usage: rbpnet impact [OPTIONS] VARIANTS

Options:
  -f, --fasta TEXT
  -m, --model TEXT
  --assert-a-is-ref      Assert that allele A is the reference allele.
  --layer TEXT           Target layer, i.e. 'profile', 'profile_target' or
                         'profile_control'.
  --window-size INTEGER

rbpnet impact takes as positional argument a variant.tsv file, with one variant specification per line of the form <chrom> <position> <strand> <allele_A> <allele_B> <*OPTIONALS>, e.g.

chr8	38262758	+	C	A
chr9	83969990	-	A	C