Sensitive and specific peak detection for SELDI-TOF mass spectrometry using a wavelet/neural-network based approach

Area of Science:

• Biomarker Discovery
• Mass Spectrometry
• Computational Biology

Background:

• Surface-Enhanced Laser Desorption/Ionization-Time of Flight (SELDI) mass spectrometry is widely used in clinical research for biomarker studies.
• Existing preprocessing algorithms for SELDI data often struggle with reproducibility, leading to miscalled peaks and systematic errors.
• Despite advancements, visual inspection reveals persistent issues in SELDI spectra preprocessing, impacting the reliability of biomarker identification.

Purpose of the Study:

• To address the ongoing reproducibility challenges in SELDI mass spectrometry data preprocessing.
• To develop and validate an improved algorithm for accurate peak detection in SELDI spectra.
• To enhance the reliability of biomarker discovery from SELDI data, particularly in clinical applications.

Main Methods:

• A novel wavelet/neural network-based algorithm was developed to automate peak calling in SELDI spectra.
• The algorithm incorporates wavelet denoising for optimal signal smoothing and a neural network trained on expert-identified peaks.
• The method was validated using data from a study on cervical mucus for early cervical cancer detection in HPV-infected women.

Main Results:

• The new algorithm achieved approximately 95% accuracy in calling peaks in test spectra.
• It effectively mimics the peak identification process of expert human users, improving upon existing automated methods.
• The application to cervical cancer biomarker data demonstrated the method's potential to overcome SELDI reproducibility issues.

Conclusions:

• The developed wavelet/neural network algorithm offers a significant improvement in SELDI data preprocessing accuracy and reproducibility.
• This advancement holds promise for more reliable biomarker discovery in clinical studies, including early cancer detection.
• The method provides a robust solution to persistent challenges in analyzing SELDI mass spectrometry data.