Deep probabilistic CCA
Medical pathology images are visually evaluated by experts for disease diagnosis, but the connection between image features and the state of the cells in an image is typically unknown. To understand this relationship, we develop a multimodal modeling and inference framework that estimates shared latent structure of joint gene expression levels and medical image features. Our method is built around probabilistic canonical correlation analysis (PCCA), which is fit to image embeddings that are learned using convolutional neural networks and linear embeddings of paired gene expression data. Using a differentiable take on the EM algorithm, we train the model end-to-end so that the PCCA and neural network parameters are estimated simultaneously. We demonstrate the utility of this method in constructing image features that are predictive of gene expression levels on simulated data and the Genotype-Tissue Expression data. We demonstrate that the latent variables are interpretable by disentangling the latent subspace through shared and modality-specific views.
While all the dependencies used for the paper are listed in environment.yml, these are operating system-specific; and some library versions (e.g.
libcxx=4.0.1) will not be available across systems. However, you can build everything you need with
python 3.7 pytorch 1.0.1 torchvision 0.2.2 numpy 1.16.2 scikit-learn 0.20.2 scipy 1.2.1 matplotlib 3.0.2
nose2 to run the unit tests. Create and activate a conda environment,
conda create -n dpcca python=3.7 conda activate dpcca
and then install these dependencies, e.g.
conda install pytorch=1.0.1 -c pytorch.
[Optional] Run the unit tests. Note that these occasionally fail due to numerical tolerances: ```bash bash run_tests.sh
Reproducing multimodal MNIST results
Generate the multimodal MNIST data set.
python -m data.mnist.generate
Create directories for experiments:
mkdir experiments experiments/example
Run the code: