Systems and methods for measuring similarity between mass spectra and peptides
Saeed,
Fahad; Tariq,
Muhammad Usman; ,
US Patent 11,251,031
(2022).
Abstract
Systems and methods for measuring cross-modal similarity between mass spectra and peptides are provided. A deep learning network can be used and, by training on a variety of labeled spectra, the network can embed both spectra and peptides onto a Euclidean subspace where the similarity is measured by the L2 distance between different points. The network can be trained on a novel loss function, which can calculate the gradients from sextuplets of data points.