Hybrid Approach: Overcoming the Left-Hand Missing Values Barrier in Proteomics

Background: Quantitative mass spectrometry-based proteomics is a crucial driver for neurodegenerative biomarker discovery. In Parkinson's disease research, identifying low-abundance proteins in plasma or cerebrospinal fluid such as Alpha-synuclein and Apolipoprotein is crucial for early diagnosis. However, these important target biomarkers are highly susceptible to left-censoring (Missing Not At Random, MNAR) when their expression falls below the instrument's Limit of Detection (LoD). Conventional imputation methods, including k-Nearest Neighbors (kNN) and Bayesian Principal Component Analysis, assume random missingness, thus introducing severe bias and overestimating the levels of low-abundance proteins.