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

@theosanderson In mass spectrometry they generally use weighted or unweighted cosine similarity for this, see https://t.co/7MkY3tGNWa. https://t.co/qOwhxrCHkr. ChatGPT suggests inverse variance weights for the cosine similarity... https://t.co/OSG4guqhkc

@Zhixu_Ni Yes, I am certain this is the case. The disparity appears to come from failure to distinguish between pure dot product or the weighted- or 'optimised'- dot product, which is scaled to favour larger m/z and lower intensity fragments.(https://t.co/