An Efficient Faiss-Based Search Method for Mass Spectral Library Searching

With billions of spectra acquired by MS/MS experiments, spectral library search tool, which is one of the most important tools for spectra identification, needs to be optimized to deal with big data. In this study, we propose an efficient spectral library search method based on faiss, which is a library for efficient similarity search proposed by Facebook-AI research team. Our faiss-based spectral library searching method contains two parts, spectral library's index construction and spectral library searching. In previous one, we embedded spectral library's spectra into vectors and then used faiss index add API to add these vectors to index file; In latter one, we created spectral library searching modular based on faiss index search API and then used this searching modular to search acquired spectra against spectral library's index file. The results show that our proposed method can search out more translation-supported identifications than SpectraST, which verify the effective identification performance of our method, but our method also searches out more unreliable identifications. And our method has a huge advantage in run-time, SpectraST took more than 4 days while our method only took 23 seconds on the test dataset.