DeepLINCS: A framework for deep learning with LINCS L1000 Data

DeepLINCS offers a variety of utilities for working with L1000 data and training deep learning models.

The class Dataset, built on pandas dataframes, offers methods to load, sample, filter, subset, and explore these data in memory.

DeepLINCS also provides a wrapper around around TensorFlow to efficiently load a Dataset and train a variety of deep learning models on the GPU.

You can find the software on github.

Installation

pip install deep-lincs

Downloading the data

The data for 1.3 million L1000 profiles are availabe on GEO. The script load_files.sh fetches the Level 3 data along with all metadata available. The largest file is quite big (~50Gb) so please be patient.

git clone https://github.com/manzt/deep_lincs.git && cd deep_lincs
source load_files.sh # download raw data from GEO

Getting started

Navigate to the example notebooks and try building a model!

cd notebooks
jupyter lab # easiest for plotting with altair

High-level API reference

• Dataset

• settings.yaml

• Extendable Models
  • AutoEncoder
  • MultiClassifier
  • SingleClassifier