Validation and Community Sharing of Ocean Spectral Libraries Generated by Machine Learning for Data Independent...

Area of Science:

• Marine biology
• Proteomics
• Bioinformatics

Background:

• Ocean metaproteomics reveals microbial community structure and function.
• Marine samples present challenges due to high biological diversity and large dynamic range of peptides.
• Data dependent acquisition (DDA) methods struggle with the complexity of ocean metaproteomic samples.

Purpose of the Study:

• To evaluate the capabilities of data independent acquisition (DIA) for ocean metaproteomic analysis.
• To assess the impact of spectral library size and composition on peptide identification.
• To compare DIA performance against DDA methods in complex marine environments.

Main Methods:

• Characterization of DIA mode for ocean metaproteomics.
• Construction of spectral libraries using machine learning (ML) to filter false positives.
• Comparison of DIA with 1D and 2D DDA, with and without gas phase fractionation.
• Evaluation of spectral libraries of varying sizes and origins, including the Ocean Protein Portal (OPP) and metagenomic open reading frames (ORFs).

Main Results:

• DIA outperformed DDA methods in ocean metaproteomic analysis.
• Larger spectral libraries consistently improved peptide identification, irrespective of sample geographic origin.
• The OPP spectral library demonstrated superior performance compared to smaller, campaign-specific libraries.
• Extremely large libraries from all ORFs in a metagenome were unworkable, hindering false discovery rate control.

Conclusions:

• DIA is a powerful approach for ocean metaproteomics.
• Optimized spectral libraries are crucial for accurate and efficient peptide identification.
• Previously generated spectral libraries can serve as valuable community resources, reducing redundant efforts in future proteomic studies.