Application of Eight Machine Learning Algorithms in the Establishment of Infertility and Pregnancy Diagnostic Models: A Comprehensive Analysis of Amino Acid and Carnitine Metabolism.

Abstract

To explore the effects of altered amino acids (AAs) and the carnitine metabolism in non-pregnant women with infertility (NPWI), pregnant women without infertility (PWI) and infertility-treated pregnant women (ITPW) compared with non-pregnant women (NPW, control), and develop more efficient models for the diagnosis of infertility and pregnancy, 496 samples were evaluated for levels of 21 AAs and 55 carnitines using targeted high-performance liquid chromatography with tandem mass spectrometry (HPLC-MS/MS). Three methods were used to screen the biomarkers for modeling, with eight algorithms used to build and validate the model. The ROC, sensitivity, specificity, and accuracy of the infertility diagnosis training model were higher than 0.956, 82.89, 66.64, and 82.57%, respectively, whereas those of the validated model were higher than 0.896, 77.67, 69.72, and 83.38%, respectively. The ROC, sensitivity, specificity, and accuracy of the pregnancy diagnosis training model were >0.994, 96.23, 97.79, and 97.69%, respectively, whereas those of the validated model were >0.572, 96.39, 93.03, and 94.71%, respectively. Our findings indicate that pregnancy may alter the AA and carnitine metabolism in women with infertility to match the internal environment of PWI. The developed model demonstrated good performance and high sensitivity for facilitating infertility and pregnancy diagnosis.