Intrahepatic cholangiocarcinoma (iCCA) is a hepatobiliary malignancy which accounts for approximately 5–10 % of primary liver cancers and has a high mortality rate. The diagnosis of iCCA remains significant challenges owing to the lack of specific and sensitive diagnostic tests available. Hence, improved methods are needed to detect iCCA with high accuracy. In this study, we evaluated the efficacy of serum amino acid profiling combined with machine learning modeling for the diagnosis of iCCA. A comprehensive analysis of 28 circulating amino acids was conducted in a total of 140 blood samples from patients with iCCA and normal individuals. We screened out 6 differentially expressed amino acids with the criteria of |Log2(Fold Change, FC)| > 0.585, P-value < 0.05, variable importance in projection (VIP) > 1.0 and area under the curve (AUC) > 0.8, in which amino acids L-Asparagine and Kynurenine showed an increasing tendency as the disease progressed. Five frequently used machine learning algorithms (Logistic Regression, Random Forest, Supporting Vector Machine, Neural Network and Naïve Bayes) for diagnosis of iCCA based on the 6 circulating amino acids were established and validated with high sensitivity and good overall accuracy. The resulting models were further improved by introducing a clinical indicator, gamma-glutamyl transferase (GGT). This study introduces a new approach for identifying potential serum biomarkers for the diagnosis of iCCA with high accuracy.
Read full abstract