Targeting AAV vectors to the CNS via de novo engineered capsid-receptor interactions

Code and documentation supporting "Targeting AAV vectors to the CNS via de novo engineered capsid-receptor interactions", including data, SVAE-based variant generation method, and figure-generation code.

Contents

Installation

Code is provided as a collection of Jupyter Notebooks and requires python3.8 or python3.9.

Install dependencies:

# Dependencies are pinned for python3.8/3.9

pip3 install -r requirements.txt

SVAE model and training data

1. Training data processing - AAV_capsid_receptor/notebooks/pulldown_assay_data_processing.ipynb

   • Starting from read counts, compute reads per million (RPM) and $\log_2$ enrichment for LY6A-Fc and LY6C1-Fc; export a CSV of mean_RPM, cv_RPM (coefficient of variation), and log2enr values for each of LY6A-Fc and LY6C1-Fc.

2. SVAE model and variant generation (for LY6C1-Fc) - AAV_capsid_receptor/notebooks/SVAE_variant_generation.ipynb

   1. Starting from the CSV exported by pulldown_assay_data_processing.ipynb, format LY6C1-Fc data into TensorFlow-compatible training batches.

   2. Initialize and train an SVAE model.

   3. Cluster and sample the trained SVAE model's latent space to generate novel variants.

Paper figures

Note: all figure-generation notebooks assume figure data is contained in AAV_capsid_receptor/data (see Data for more details).

1. Figure 1 and 1S (supplemental) panels - AAV_capsid_receptor/figures/fig1.ipynb
2. Figure 2 and 2S (supplemental) panels - AAV_capsid_receptor/figures/fig2.ipynb
3. Figure 3 panels - AAV_capsid_receptor/figures/fig3.ipynb
4. Figure 4 and 4S (supplemental) panels - AAV_capsid_receptor/figures/fig4.ipynb

Data

SVAE training data

Sequencing read count data for LY6A-Fc and LY6C1-Fc are in AAV_capsid_receptor/data.

Figures data

All relevant data is stored on Zenodo at DOI 10.5281/zenodo.7689794. Once downloaded, data files should be put into AAV_capsid_receptor/data - by default, figure-generation notebooks will search for data there.