Optimization and testing of mass spectral library search algorithms for compound identification.

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Spectroscopy

Background:

• Accurate identification of unknown compounds from mass spectra is crucial in chemistry.
• Existing library search algorithms vary in their effectiveness for low-resolution mass spectral data.

Purpose of the Study:

• To optimize and compare five common algorithms for identifying unknown compounds using low-resolution mass spectra.
• To determine the most effective algorithm and optimal parameters for spectral library searching.

Main Methods:

• Five algorithms (dot-product, Euclidean distance, absolute value distance, probability-based matching, Hertz et al.) were tested.
• Algorithms were optimized by adjusting mass weighting and intensity scaling factors.
• Performance was evaluated based on the rank of the correct compound in the search results.

Main Results:

• The dot-product algorithm demonstrated the highest accuracy (75% for rank 1), outperforming other methods.
• Euclidean distance (72%) and absolute value distance (68%) also showed strong performance.
• Optimal parameters included square root intensity scaling and square or cube mass weighting.

Conclusions:

• The dot-product algorithm is the most effective for low-resolution mass spectral library searching.
• Optimization of parameters like mass weighting and intensity scaling is critical for improving identification accuracy.
• Minor improvements were observed by adding a term to weight common peaks in the dot-product algorithm.