Enhancing Top-Down Proteomics Data Analysis by Combining Deconvolution Results through a Machine Learning Strategy.

Area of Science:

• Proteomics
• Computational Biology
• Biochemistry

Background:

• Top-down mass spectrometry (MS) is crucial for characterizing proteoforms, but data analysis is complex.
• Deconvoluting isotopic distributions is a critical, yet challenging, step in top-down MS data processing.
• Existing deconvolution algorithms yield varied accuracy, impacting proteoform identification.

Purpose of the Study:

• To develop a machine learning strategy for integrating and improving deconvolution results in top-down MS.
• To enhance the accuracy and confidence of proteoform identification through improved peak detection.

Main Methods:

• A machine learning strategy was designed to process and combine peak lists from multiple deconvolution algorithms (THRASH, TopFD, MS-Deconv, SNAP).
• Ensemble methods, including a random forest algorithm, were employed to generate consensus peak lists.
• Performance was evaluated using recall and precision values.

Main Results:

• The random forest machine learning algorithm achieved a recall of 0.60 and a precision of 0.78.
• This consensus approach significantly outperformed the best single deconvolution algorithm (recall 0.47, precision 0.58).
• The strategy improved the detection of true positive peaks and filtering of false positives.

Conclusions:

• The developed machine learning strategy effectively enhances proteoform identification accuracy and confidence in top-down proteomics.
• This method shows significant promise for high-throughput proteoform analysis.
• Integrating multiple deconvolution algorithms via machine learning offers a robust solution for complex MS data interpretation.