genDL

Genomic experiments in Deep Learning. Our goal is to improve upon classical methods (PWM, WAM, MM) using neural networks.

Directory Structure

• arch - archive of old stuff, don't modify
• data - data files and scripts for managing/generating data files
• docs - tutorial, API
• gendl - shared libraries (include in PYTHONPATH)
• nn - neural network experiments
• pwm - position weight matrix experiments
• test - unit and functional testing
• xtra - uncategorized experiments that may find an eventual home elsewhere

Each of the directories above has its own REAMDE.md file that describes its contnents and intents. An overview is given below.

The arch Directory

This is the previous incarnation of genDL in all its former glory. This entire directory will be deleted eventually. Please don't make changes to this directory ever.

The data Directory

If data is used by more than one experiment, place it in the data directory. Scripts used to generate data may also be placed in data. Some data generation procedures are in datacore/project_gendl.

• splice42 contains 42 bp windows centered on the donor or acceptor site
• ime50 contains 50 bp windows of A. thaliana introns

The gendl Directory

All shared libraries go in the gendl directory. Don't put shared libraries inside experiment directories. All imports should come from the parent gendl namespace followed by the library name. Here are some examples of how to import gendl libraries.

import gendl.pwm
import gendl.pwm as pwm
from gendl.pwm import make_pwm

If you're editing gendl libraries, don't import obscure libraries or libraries with large dependancies.

Experiment Directories

Experiment directories contain a mixture of code, data, and maybe figures or other files. Each type of experiment may have several flavors. For example, you will find pwm/strawman and pwm/wam. Each sub-experiment should have its own README.md to describe its contents and intents.