Phenotype Classification using Proteome Data in a Data-Independent Acquisition Tensor Format.

Area of Science:

• Proteomics
• Bioinformatics
• Machine Learning

Background:

• Data-independent acquisition (DIA) mass spectrometry (MS) generates complex data.
• Existing DIA software often requires peptide precursor identification for analysis.
• Phenotype prediction from MS data is crucial for biological and clinical applications.

Purpose of the Study:

• To develop a novel approach for phenotype prediction using DIA-MS data.
• To introduce a new data format, DIA tensor (DIAT), for simplified DIA-MS data analysis.
• To apply deep learning models for phenotype classification using the DIAT format.

Main Methods:

• Conversion of DIA-MS data into the novel DIA tensor (DIAT) format.
• Utilizing DIAT files for direct input into deep neural networks.
• Application of deep learning models for phenotype prediction in hepatocellular carcinoma and thyroid nodule datasets.

Main Results:

• Achieved 96.8% accuracy in distinguishing malignant from benign hepatocellular carcinoma samples.
• Classified thyroid nodules with 91.7% accuracy in an independent test cohort.
• Outperformed existing deep learning models based on peptide and protein matrices.

Conclusions:

• The DIAT format facilitates direct deep learning application on DIA-MS data.
• This novel strategy enables accurate biological and clinical phenotype classification.
• Future work will focus on interpreting the deep learning models derived from DIAT analysis.