Nonlinear methods for dimensionality reduction and clustering of bacterial single-cell sequencing data - intermediate data and figures (MSc thesis)

Data, intermediate results and figures for analyses of my master's thesis in biostatistics at LMU Munich. I took a look on how to use Nonlinear Matrix Decomposition (NMD) (Saul, L., 2022) in the context of bacterial scRNA-seq analysis (Heumos, L., et. al. 2023), replacing Principal Component Analysis in the optimized workflow, as outlined in Ostner, J. (2024).

My thesis was structured along the following objectives:

implement the algorithms from Seraghiti, G., et. al. (2023) in the Python module nomad in cooperation with Flatiron Institute

code for the simulation study of the algorithms in Seraghiti, G., et. al. (2023) with varying sparsity can be found in /simulation

apply NMD in the context of the BacSC workflow (Ostner, J., et. al. (2024)) on raw and normalized counts (found in /application/analysis), also for manually set number of latent dimensions

explore NMD's potential for imputation of sampling zeros (check /application/NMD_zero_imputation /)

potential of Poisson-Hurdle model-based clustering (Qiao, Z., et. al. (2023)) for scRNA-seq (/application/poisson_hurdle).