preprocess.md

October 6, 2015 · View on GitHub

Basset

Deep convolutional neural networks for DNA sequence analysis.

Preprocess

#### preprocess_features.py

Merge a set of feature BED files for training into a single BED and activity table.

Arguments	Type	Description
target_beds_file	table listing labels and BED	One line per sample- label then BED path

Options	Variable	Description
-a	db_act_file	Existing database activity table
-b	db_bed	Existing database BED
-c	chrom_lengths_file	Table of chromosome lengths
-m	merge_overlap	Overlap length (after extension to feature_size) above which to merge features [Default: 200]
-n	no_db_activity	Do not pass along the activities of the database sequences [Default: False]
-o	out_prefix	Output file prefix [Default: features]
-s	feature_size	Extend features to this size [Default: 600]
-y	ignore_y	Ignore Y chromsosome features [Default: False]

#### seq_hdf5.py

Construct an HDF5 file, dividng the data into training, validation, and test subsets.

Arguments	Type	Description
fasta_file	FASTA	FASTA file of sequences.
targets_file	Table	Targets activity table.
out_file	HDF5	Output HDF5 file.

Options	Variable	Description
-b	batch_size	Align sizes with batch size
-c	counts	Validation and training percentages are given as raw counts [Default: False]
-r	permute	Permute sequences [Default: False]
-s	random_seed	numpy.random seed [Default: 1]
-t	test_pct	Test % [Default: 0]
-v	valid_pct	Validation % [Default: 0]

#### basset_sample.py

Sample sequences from an existing database.

Arguments	Type	Description
db_bed	BED	Existing database BED.
db_act_file	Table	Existing database activity table.
sample_seqs	int	Number of sequences to sample.
output_prefix	str	Filename prefix for output BED and activity table files.

Options	Variable	Description
-s	seed	Random number generator seed [Default: 1]