DreamAI2: an imputation tool for data-independent acquisition proteomics data
Year: 2026
Type: article
Abstract: Background: Missing values are pervasive in high-throughput proteomics data from mass spectrometry-based experiments. They can compromise statistical power and mislead downstream inference, necessitating proper imputation as a key preprocessing step. While previous efforts, such as the 2017 National Cancer Institute-Clinical Proteomic Tumor Analysis Consortium (NCI-CPTAC) DREAM Proteogenomics Challenge, established benchmarks for imputation on isobaric labeling-based data-dependent acquisition (DDA) profiles, the applicability of these strategies to the increasingly popular label-free data-independent acquisition (DIA) experiment data remains unverified. This is a critical gap, given the distinct differences in instrument settings and acquisition modes between the two experiments. Methods: To address this gap, we systematically benchmarked seven individual algorithms and four ensemble strategies on label-free DIA data, using a unique dataset comprising paired label-free DIA and tandem mass tag (TMT)-DDA global proteomics profiles from 162 melanoma tumors. By leveraging the matched TMT measurements as an independent external reference, we evaluated DIA imputation accuracy under realistic conditions. Results & Conclusion: Building on these benchmarking results, we introduce DreamAI2, a DIA-optimized ensemble imputation integrating complementary strategies to achieve superior performance. This study clarifies the transferability of existing imputation methods and delivers DreamAI2 as an open-source tool (https://github.com/WangLab-MSSM/DreamAI2) alongside a rigorous validation framework for future research.
Source: Bioinformatics Advances
Language: English
Subfield: Spectroscopy
Field: Chemistry
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,726
OpenAlex ID: W7235054324
