Additional file 1 of Picky with peakpicking: assessing chromatographic peak quality with simple metrics in metabolomics

Additional file 1: Table S1 Model parameters and significance according to both logistic and random forest models: Model outputs from the full (all parameters) model trained on the combined Falkor and MESOSCOPE datasets. The parameter name is reported alongside its more detailed name, with the logistic model estimate, standard error, test statistic, and p-value for each predictive term. We also included the estimate obtained from the elastic net model (α = 0.5) and the accuracy and Gini index decreases per term. NA values are reported for several terms that were removed from the non-regularized regression because they were too highly correlated with other parameters. Table S2 Confusion matrices for logistic, regularized, and random forest regressions: Confusion matrices reported in long format for the different models reported in this manuscript. Models were all both internally tested using the same dataset for testing and training as well as cross-validated against the other fully labeled dataset. For the two-parameter model, we also report the values obtained when training on the two datasets combined (train = Both). For each model and test-train set we report the absolute number of false positives (manually labeled as "Bad" but model labeled as "Good"), false negatives (manually labeled as "Good" but model labeled as "Bad"), true positives (both manual and model labelled as "Good"), and true negatives (both manual and model labeled as "Bad") as well as the calculated false discovery rate (FDR, calculated as FP/(FP + TP)) and the fraction of good features found (GFF, calculated as TP/(TP + FN)). Unless otherwise specified, all values reported here used a likelihood threshold of 0.5 for categorization. Table S3 Internal standards used for normalization: Isotope-labeled internal standards used for the best-matched internal standard normalization procedure as denoted in Boysen et al. (2018).