Code accompanying the Kipoi manuscript

1. Install miniconda or anaconda.
2. Install git-lfs: conda install -c conda-forge git-lfs && git lfs install
3. Clone this repository: git clone https://github.com/kipoi/manuscript.git && cd manuscript
4. Run: git lfs pull '-I data/**'
5. Run: conda env create -f env.yaml. This will install a new conda environment kipoi-manuscript
6. Activate the environment: source activate kipoi-manuscript
7. Install the m_kipoi python package for this repository : pip install -e .

• m_kipoi - python package (contains python functions/classes common across multiple notebooks)
• src - scripts for running the experiments and producing the figures
  • tf-binding - Figure 2
  • transfer_learnining - Figure 3
  • mutationmaps - Figure 4
  • splicing - Figure 5
• data - data files
• models/Divergent421 - (optional) Pre-trained DNA accessibility model (Kipoi format)
• slurm - (optional) SLURM profile for snakemake https://github.com/Snakemake-Profiles/slurm

• Fig 2 - Making model predictions using a generic command
  • With Snakemake, it's possible to write a generic command and make predictions for all the models
    • Each model is executed in a separate conda environment
      • Installation rule
      • Prediction rule
      • Function to get the kipoi binary given the conda environment name: get_env_executable
• Fig 3 - Transfer learning:
  • Script for performing transfer-learning using a Keras-based model from Kipoi: tlearn.py
  • Pre-computing the activations of the frozen part of the network and then training the model pre-computed-tlearn.ipynb
• Fig 4 - Plugins (variant effect prediction and interpretation)
  • Generic rule for variant effect prediction (CLI) used for building KipoiSplice/4 (relevant for Fig. 4 and Fig. 5):
    • create_env
    • annotate_vcf
  • Producing a mutation-map (Figure 4): src/mutationmaps/plot.ipynb
• Fig 5 - Composite models
  • Tutorial: http://kipoi.org/docs/tutorials/composing_models/
  • Concrete example (KipoiSplice/4): KipoiSplice/4/dataloader.py