Evaluation of differential abundance analysis strategies for phosphoproteomics data with missing values.
We evaluated different testing strategies (i.e. ignoring missing values, imputation, compound hypothesis tests) under 3,234 data scenarios of various levels of sample size, fold change, missing ratio and MNAR ratio. Each scenario was simulated for 500 rounds and a total of 18 methods were compared.
- invgamma
- Bioconductor
- limma
- SDAMS
- pcaMethods
- trust
- ggplot2
- gridExtra
- data.table
- NumPy (1.19.2+)
- Pandas (1.2.1+)
- Scikit-Learn (0.23.2+)
- SciPy (1.5.2+)
The codes of simulation study are in simulation_HPC.R
The whole simulation work is very computationally intensive in R, so a simple implementation can be carried out by setting 'simple_implement=TRUE'.
Note:
- The method of SDA was modified by cancelling the data-cleaning step of removing features with low numbers of non-zeros.
- bPCA imputation was modified by lower the tol parameter in solve function to prevent the error of computationally singular.
The codes of summarizing the results (adjusted p-values）and evaluating the different methods for the 500-round simulation are in results_HPC.py
The codes of plotting figures are in results_plot.R
Only the results data for generating the figures is in this repository due to the limit of file size. The results data of all methods under all scenarios in the 500-round simulation is available upon request. If you have problems, please contact jing.li@sjtu.edu.cn