I also worked with Dr. Ke "Kurt" Zhang as an Undergraduate Research Assitant on Deep Gaussian Mixture Models for scRNA-seq data. We developed a Gaussian Mixture Variational Autoencoder that utilized Maximum Mean Discrepancy to merge batches. In addition, raw scRNA-seq data is represented as counts of gene expressions in cells which results in a large number of zero count data. Our model used a Decoder that was adapted to handle Zero Inflated Negative Binomial distributed data. To tune the hyperparameters of this method, I used Bayesian Optimized Hyperband. I was a co-presenter at the Single Cell Omics Symposium this year where we presented a poster of our results and methods from this project. We are continuing to explore the applications of the method to overcome batch effect...
An amazing website about me.