Comparative assessment and novel strategy on methods for imputing proteomics data

Abstract

Missing values are a major issue in quantitative proteomics analysis. While many methods have been developed for imputing missing values in high-throughput proteomics data, a comparative assessment of imputation accuracy remains inconclusive, mainly because mechanisms contributing to true missing values are complex and existing evaluation methodologies are imperfect. Moreover, few studies have provided an outlook of future methodological development. We first re-evaluate the performance of eight representative methods targeting three typical missing mechanisms. These methods are compared on both simulated and masked missing values embedded within real proteomics datasets, and performance is evaluated using three quantitative measures. We then introduce fused regularization matrix factorization, a low-rank global matrix factorization framework, capable of integrating local similarity derived from additional data types. We also explore a biologically-inspired latent variable modeling strategy—convex analysis of mixtures—for missing value imputation and present preliminary experimental results. While some winners emerged from our comparative assessment, the evaluation is intrinsically imperfect because performance is evaluated indirectly on artificial missing or masked values not authentic missing values. Nevertheless, we show that our fused regularization matrix factorization provides a novel incorporation of external and local information, and the exploratory implementation of convex analysis of mixtures presents a biologically plausible new approach.

Neither random nor censored: estimating intensity-dependent probabilities for missing values in label-free proteomics

“…There has been much discussion in the literature about the relevance of Donald Rubin’s missing at random (MAR) and missing not at random (MNAR) classification for mass spectrometry data ( Karpievitch et al 2012 ; Webb-Robertson et al 2015 ; Lazar et al 2016 ; Wang et al 2020 ; Gardner and Freitas 2021 ; Liu and Dongre 2021 ; Dekermanjian et al 2022 ; Shen et al 2022 ). Our work shows that missing intensities are MNAR but that the dependence of missing value frequency on intensity is gradual.…”

Dealing with missing values in proteomics data

“…For each missing variable numbered from 1 to M , RMSE or NRMSE is ranked among all imputation methods. The final value of $$SOR$$ is the sum of total ranks for each missing variable [59, 117]. $$$$\begin{equation} {\textit{SOR}}=\sum _{i=1}^{M}{\textit{Rank}}_{i}\left({\textit{RMSE\ or\ NRMSE}}\right) \end{equation}$$$$…”

“…For each missing variable numbered from 1 to M, RMSE or NRMSE is ranked among all imputation methods. The final value of SOR is the sum of total ranks for each missing variable [59,117].…”

The community ecology perspective of omics data

“…Differences in detectability can also be random and result in lower signal-to-noise ratios (e.g., in proteomics data [67]). Mathematical modeling of persistent detection biases has been proposed as a first step to identify where biases arise and to quantify them in metagenomic data [68], while latent variable modeling has been proposed for estimating missing values in proteomics data [69].…”