Proteomics Data Imputation With a Deep Model That Learns From Many Datasets.

Area of Science:

• Proteomics
• Bioinformatics
• Computational Biology

Background:

• Missing values are a significant challenge in quantitative mass spectrometry proteomics.
• These missing values impede reproducibility, reduce statistical power for differential abundance analysis, and complicate the study of low-abundance proteins.

Purpose of the Study:

• To introduce Lupine, a novel deep learning-based method for imputing missing values in quantitative proteomics data.
• To demonstrate that joint learning from multiple datasets enhances imputation accuracy.

Main Methods:

• Development of Lupine, a deep learning imputation tool for proteomics.
• Application of Lupine to tandem mass tag (TMT) data from over 1000 cancer patient samples across 10 cancer types (Clinical Proteomics Tumor Atlas Consortium).

Main Results:

• Lupine outperforms existing state-of-the-art imputation methods.
• Lupine successfully identifies differentially abundant proteins and Gene Ontology terms.
• The method learns a meaningful representation of proteins and patient samples.

Conclusions:

• Lupine offers a significant advancement in handling missing data in proteomics.
• The approach of joint learning from multiple datasets is effective for improving imputation accuracy.
• Lupine is an open-source Python package, promoting accessibility and further research.