An unsupervised machine learning method for assessing quality of tandem mass spectra

Area of Science:

• Proteomics
• Computational Biology
• Mass Spectrometry

Background:

• Tandem mass spectrometry generates vast amounts of data, but many spectra are low quality, hindering peptide identification.
• Current quality assessment methods often rely on supervised learning, requiring unavailable training datasets for new data.
• Efficient quality assessment is crucial for reducing search time and false positives in proteomics.

Purpose of the Study:

• To develop an unsupervised machine learning method for tandem mass spectra quality assessment.
• To enable quality assessment without the need for pre-existing labeled training datasets.
• To improve the efficiency of peptide identification pipelines in proteomics.

Main Methods:

• An unsupervised machine learning approach was developed for quality assessment of tandem mass spectra.
• Conditional probabilities of spectra being high quality were estimated using individual feature assessments.
• A constraint optimization problem was formulated and solved with a convergent algorithm.

Main Results:

• The proposed method effectively assesses tandem mass spectra quality without requiring training data.
• Searching only high-quality spectra identified by the method saved 56% and 62% of database search time in experiments.
• This approach resulted in minimal loss of high-quality spectra.

Conclusions:

• The unsupervised method demonstrates strong performance in tandem mass spectra quality assessment.
• The developed approach for estimating conditional probabilities is effective and efficient.
• This method offers a practical solution for improving proteomics data analysis pipelines.