PhosSight: A Unified Deep Learning Framework Boosting and Accelerating Phosphoproteome Identification to Enable...

Area of Science:

• Biochemistry
• Computational Biology
• Genomics

Background:

• Protein phosphorylation is crucial for cellular signaling, with mass spectrometry (MS) as the primary analysis tool.
• Current MS methods like Data-Dependent Acquisition (DDA) and Data-Independent Acquisition (DIA) have limitations including undersampling, missing values, and computational inefficiencies.
• Existing phosphoproteomics workflows struggle with data completeness and search speed.

Purpose of the Study:

• To introduce PhosSight, a unified deep learning framework to improve identification depth and accelerate phosphoproteomics data analysis.
• To address acquisition biases in DDA and DIA methods for more comprehensive phosphoproteome profiling.
• To enhance biological discovery in precision oncology through improved phosphoproteomics.

Main Methods:

• Developed PhosSight, a deep learning framework with PhosDetect, a model predicting peptide detectability using phosphorylation-specific features.
• Applied PhosSight to DDA by refining site localization and rescoring using predicted retention time, fragment intensity, and detectability.
• Utilized PhosSight in DIA for detectability-guided library pruning to accelerate searches and reduce noise.
• Benchmarked PhosSight on synthetic and real-world datasets, including a Uterine Corpus Endometrial Carcinoma (UCEC) cohort.

Main Results:

• PhosSight significantly augmented identification depth and accelerated search efficiency in both DDA and DIA modes.
• The framework successfully recovered marginal, low-abundance spectra in DDA and reduced non-detectable noise in DIA without compromising sensitivity.
• Application to a UCEC cohort improved data completeness and expanded the quantifiable phosphoproteome.
• Enabled the discovery of novel prognosis-associated kinase targets, such as MARK2.

Conclusions:

• PhosSight offers a unified deep learning solution to overcome limitations in current phosphoproteomics acquisition and analysis.
• The framework enhances data completeness and analytical efficiency, leading to expanded phosphoproteome coverage.
• PhosSight facilitates biological discovery in precision oncology by identifying novel kinase targets relevant to cancer prognosis.