# Revisiting the thorny issue of missing values in single-cell proteomics

**Authors:** Christophe Vanderaa, Laurent Gatto  
**arXiv:** [2304.06654](https://arxiv.org/abs/2304.06654)  
**Date:** 2023-04-13  
**Summary page:** https://www.summarizepaper.com/en/paper/2304.06654v1/

## Key points

- Mass spectrometry-based proteomics data analysis faces challenges due to missing values.
- Single-cell proteomics has led to a significant increase in missing values.
- Imputation is a popular approach for managing missing values, but it has drawbacks.
- Vanderaa et al. discuss the advantages and drawbacks of imputation and highlight five main challenges linked to missing value management in single-cell proteomics.
- The accuracy of imputed values may not reflect the true underlying biological signal, leading to biased downstream analyses.
- Different imputation methods may produce different results depending on the dataset's characteristics.
- Missingness patterns and proportions should be reported explicitly, and standardized codes should be used for encoding missing values.
- Imputations should be incorporated into downstream analyses with caution as they can impact results.
- Transparency and reproducibility are crucial when reporting methods used for managing missing values in single-cell proteomics data analysis.

## Summary

The analysis of mass spectrometry-based proteomics data poses a significant challenge due to the presence of missing values. This challenge has become even more pronounced with the emergence of mass spectrometry-based single-cell proteomics, which has led to a dramatic increase in missing values. While the field is still actively debating on the best practices for managing missing values, imputation remains a popular approach. However, imputation has several drawbacks that need to be considered when dealing with single-cell proteomics data. In their paper titled "Revisiting the Thorny Issue of Missing Values in Single-Cell Proteomics," Vanderaa et al. (2023) discuss the advantages and drawbacks of imputation and highlight five main challenges linked to missing value management in single-cell proteomics. The authors suggest that future developments should aim to solve these challenges, whether it is through imputation or data modelling. The first challenge highlighted by Vanderaa et al. is related to the accuracy of imputed values. Imputed values may not accurately reflect the true underlying biological signal, leading to biased downstream analyses. The second challenge is related to the choice of imputation method, as different methods may produce different results depending on the characteristics of the dataset. The third challenge relates to how missing values are reported and encoded in datasets. The authors recommend that researchers report both missingness patterns and proportions explicitly and encode missing values using a standardized code. The fourth challenge concerns how imputations are incorporated into downstream analyses such as clustering or differential expression analysis. The authors caution against blindly incorporating imputed values into such analyses without considering their potential impact on downstream results. Finally, Vanderaa et al. highlight the need for transparency and reproducibility in reporting methods used for managing missing values in single-cell proteomics data analysis. Overall, while imputation remains a practical solution widely adopted in single-cell proteomics data analysis, researchers should carefully consider its limitations and explore alternative methods for managing missing values. The authors provide recommendations for reporting missing values and proper encoding of missing values to ensure transparency and reproducibility in single-cell proteomics data analysis.