Method for controlling mass spectrometer using analysis data determined by two variables

Abstract

Provided is a method for determining a first N×K matrix S and second K×M matrix P, using factor number K, so that their product SP approximates to an N×M data matrix X obtained by analyzing a sample containing an unknown number of components. Multiple candidates of regularization parameter λr and one sparsity-inducing regularization function R(S,P) are prepared. For each regularization-parameter candidate λr, a candidate Sr of matrix S and candidate Pr of matrix P which minimize a loss function L(S,P)=D(X|SP)+λrR(S,P) are determined, where D(X|SP) is a distance function expressing the degree of difference between X and SP. For each combination of matrix element Xin X and corresponding matrix element (SrPr)in SrPr, a transformed value y=F(X(SrPr)) is determined using function Fwhich performs variable transform from probability distribution Pcorresponding to D(X|(SP)) into common probability distribution Pcommon, and goodness of fit between yand Pcommon is calculated. Among the candidates of λr, a candidate is found which yields the highest value of the goodness of fit, or one which yields the goodness of fit higher than a predetermined threshold and also has the largest value of λr. The matrix candidates Sr and Pr determined for the found λr are selected as the first and second matrices S and P.