The following supporting information can be downloaded at https://www.mdpi.com/article/10.3390/metabo14010002/s1, Figure S1. Averaged Fourier transform–infrared (FT–IR) spectrum of baseline fecal samples from healthy controls and patients with ulcerative colitis. Table S1. Institutional and collection data information of samples from the development and validation models. A, The Catholic University of Korea St. Vincent’s Hospital; B, Chung-Ang University Hospital; C, Seoul National University Hospital; D, SNU Boramae Medical Center; E, Ewha Womans University Mokdong Hospital; F, Chosun University Hospital; G, Daejeon St. Mary’s Hospital; H, Severance Hospital; I, Keimyung University Dongsan Medical Center; J, Korea University Anam Hospital; K, Inje University Seoul Paik Hospital; L, Inje University Haeundae Paik Hospital; M, Chonnam National University Hospital; N, Kangbuk Samsung Hospital; O, Inha University Hospital; P, Kyungpook National University Hospital; Q, KyungHee University Medical Center. Table S2. Assignment of FT–IR spectra of fecal samples from patients with ulcerative colitis and healthy controls. ^(a) Detailed information on each reference used for spectral assignment is listed in the reference section below. Table S3. Parameters of OPLS–DA models applying various normalization and scaling methods for discriminating baseline fecal samples in three comparison cases. CV-ANOVA, cross validated-analysis of variance; HC, healthy controls; OPLS-DA, orthogonal partial least squares-discriminant analysis; UC, ulcerative colitis; UV, unit variance; RM, patients in remission; NRM, patients not in remission; W8, baseline fecal samples from patients after 8 weeks of adalimumab treatment; W56, baseline fecal samples from patients after 56 weeks of adalimumab treatment. Bold characters represent the best parameters of the selected model. Table S4. Comparison of absorbance intensity of major assigned spectral peaks from fecal samples in three types of discrimination models. HC, healthy controls; NRM, patient not in remission; RM, patient in remission; UC, ulcerative colitis; VIP, variable importance of projection; W8, baseline fecal samples from patients after 8 weeks of adalimumab treatment; W56, baseline fecal samples from patients after 56 weeks of adalimumab treatment. The Direction of comparison is UC/HC and RM/NRM with a fold change threshold of 1.0. Significant differences between the two groups are indicated with asterisks (*) (p t-test. Bold characters represent selected wavenumbers satisfying both VIP values over 1.0 and p-values (Student’s t-test) below 0.05. Table S5. Comparison of 10-fold cross-validation performance, by various machine learning algorithms using baseline fecal samples, of the remission prediction model at 8 and 56 weeks of adalimumab treatment in patients with UC. DT, decision tree; KNN, K-nearest neighbors; LR, logistic regression; OPLS–DA, orthogonal partial least squares–discriminant analysis; RF, random forest; ROC–AUC, receiver operating characteristic–area under the curve; SVM, support vector machine; UC, ulcerative colitis. LR, KNN, DT, RF, and SVM were performed by SciKit-Learn software (version 0.24.0) and parameters were selected by the function of “GridSearchCV” in SciKit-Learn software (version 0.24.0). OPLS–DA was performed using SIMCA software (version 15.0.2), and the parameters were selected using the “autofit” function in SIMCA. The 95% confidence intervals are presented within parentheses. Table S6. Comparison of the predictive performances of various ML algorithms (with internal and external validation) of the remission prediction model at 8 and 56 weeks of adalimumab treatment in patients with ulcerative colitis. The 95% confidence intervals are indicated within parentheses, with the parameter values in the development model. DT, decision tree; KNN, K-nearest neighbors; LR, logistic regression; rbf, radial basis function; RF, random forest; ROC–AUC, receiver operating characteristic-area under the curve; SVM, support vector machine; UC, ulcerative colitis. LR, KNN, DT, RF, and SVM were performed by SciKit-Learn software (version 0.24.0) and parameters were selected by the function of “GridSearchCV” in SciKit-Learn software (version 0.24.0).