peptide-imputation-inference

This repository contains codes associated with the paper [Augmented Doubly Robust Post-Imputation Inference for Proteomic Data] by Moon et al. (2024).

General pipeline

A full pipeline for least square inference for proteomic data with missingness is provided in [pipeline_peptide_post_imputation_inference.Rmd].

This pipeline can be applied to a custom dataset, provided that it conforms to the following format:

• raw.pep: high-dimensional peptide data outcome with missing values. (A matrix #observations x #peptides)
• covariate: A low-dimensional covariate without missing values. (A data frame #observations x #covariates)
• missing_pattern: A missing pattern of raw.pep. Either MAR or MCAR.

scVAEIT

The method involves regressing each column of raw.pep on both the covariate and the other columns of raw.pep. Columns of raw.pep has many missing entries, even when used as a covariate in the regression problem. We use an algorithm called `scVAEIT', a variant of variational auto-encoder, which is a deep neural network tool that allows for flexible input and simultaneous estimation of the multi-response regression (Du et al. (2022)).

Here, we provide version 0.2.0 of scVAEIT, which is the version used for the analysis in the paper by Moon et al. (2024). For general use, we recommend downloading the newest version of the code from the repository jaydu1/scVAEIT. We also provide an R wrapper function, R_wrapper_VAE.R, to compile scVAEIT in R. Both the folder scVAEIT and the file R_wrapper_VAE.R should be located in the same directory with a pipeline code.

The scVAEIT requires setting up the Python package dependencies. Below are the versions that are used for the analysis in the paper.

python                    3.9.18
scanpy                    1.1.10 
scikit-learn              1.3.2
tensorflow                2.14.0
tensorflow-probability    0.22.1

The dependencies can be installed via the following commands:

mamba create --name tf python=3.9 -y
conda activate tf
mamba install -c conda-forge "tensorflow>=2.12" "tensorflow-probability>=0.12" pandas jupyter -y
mamba install -c conda-forge "scanpy>=1.9.2" matplotlib scikit-learn -y

If you are using conda, simply replace mamba above by conda.

Reproducibility materials

We provide codes for reproducing the results presented in the paper.

The scpdata folder contains codes for reproducing the result in Section 4. The data used for this analysis is the single-cell proteomic data measured by Leduc et al. (2022), and can be downloaded from a Bioconductor package scpdata. A file scpdata_reproduce_figures.Rmd reproduces figures in the main text and the supplementary material. A file scpdata_reproduce_main_results.Rmd reproduces the peptide discovery results. A file scpdata_reproduce_realistic_simulation1.Rmd and scpdata_reproduce_realistic_simulation2.Rmd reproduce the realistic simulation result presented in Section 4.1.

The ADdata folder contains codes and data for reproducing the result in Section 5. The data used for this analysis is the bulk-cell brain data related to Alzheimer's Disease. The file meta.csv was downloaded from url. Four other files for peptide data on each brain region can be downloaded from url with a selection of Level3A.

For running the codes, both the folder scVAEIT and the file R_wrapper_VAE.R should be located in the same directory.

References

• Moon, Haeun, Du, Jin-Hong, Lei, Jing, and Roeder, Kathryn. 2024. "Augmented doubly robust post-imputation inference for proteomic data" Arxiv
• Du, Jin-Hong, Cai, Zhanrui, and Roeder, Kathryn. 2022. "Robust probabilistic modeling for single-cell multimodal mosaic integration and imputation via scVAEIT" Proceedings of the National Academy of Sciences, 119(49)
• Leduc, Andrew and Huffman, R Gray and Cantlon, Joshua and Khan, Saad and Slavov, Nikolai. 2022. "Exploring functional protein covariation across single cells using nPOP", Genome Biology, 23(1)