The Use of Machine Learning Algorithms and the Mass Spectrometry Lipidomic Profile of Serum for the Evaluation of Tacrolimus Exposure and Toxicity in Kidney Transplant Recipients

Identification of urinary metabolites correlated with tacrolimus levels through high-precision liquid chromatography-mass spectrometry and machine learning algorithms in kidney transplant patients

Background and aim

Tacrolimus, a widely used immunosuppressive drug in kidney transplant recipients, exhibits a narrow therapeutic window necessitating careful monitoring of its concentration to balance efficacy and minimize dose-related toxic effects. Although essential, this approach is not optimal, and tacrolinemia, even in the therapeutic interval, might be associated with toxicity and rejection within range. This study aimed to identify specific urinary metabolites associated with tacrolimus levels in kidney transplant patients using a combination of serum high-precision liquid chromatography-mass spectrometry (HPLC-MS) and machine learning algorithms.

Methods

A cohort of 42 kidney transplant patients, comprising 19 individuals with high tacrolimus levels (>8 ng/mL) and 23 individuals with low tacrolimus levels (<5 ng/mL), were included in the analysis. Urinary samples were subjected to HPLC-MS analysis, enabling comprehensive metabolite profiling across the study cohort. Additionally, tacrolimus concentrations were quantified using established clinical assays.

Results

Through an extensive analysis of the HPLC-MS data, a panel of five metabolites were identified that exhibited a significant correlation with tacrolimus levels (Valeryl carnitine, Glycyl-tyrosine, Adrenosterone, LPC 18:3 and 6-methylprednisolone). Machine learning algorithms were then employed to develop a predictive model utilizing the identified metabolites as features. The logistic regression model achieved an area under the curve of 0.810, indicating good discriminatory power and classification accuracy of 0.690.

Conclusions

This study demonstrates the potential of integrating HPLC-MS metabolomics with machine learning algorithms to identify urinary metabolites associated with tacrolimus levels. The identified metabolites are promising biomarkers for monitoring tacrolimus therapy, aiding in dose optimization and personalized treatment approaches.

LipidSIM: Inferring mechanistic lipid biosynthesis perturbations from lipidomics with a flexible, low-parameter, Markov modeling framework

Lipid metabolism is a complex and dynamic system involving numerous enzymes at the junction of multiple metabolic pathways. Disruption of these pathways leads to systematic dyslipidemia, a hallmark of many pathological developments, such as nonalcoholic steatohepatitis and diabetes. Recent advances in computational tools can provide insights into the dysregulation of lipid biosynthesis, but limitations remain due to the complexity of lipidomic data, limited knowledge of interactions among involved enzymes, and technical challenges in standardizing across different lipid types. Here, we present a low-parameter, biologically interpretable framework named Lipid Synthesis Investigative Markov model (LipidSIM), which models and predicts the source of perturbations in lipid biosynthesis from lipidomic data. LipidSIM achieves this by accounting for the interdependency between the lipid species via the lipid biosynthesis network and generates testable hypotheses regarding changes in lipid biosynthetic reactions. This feature allows the integration of lipidomics with other omics types, such as transcriptomics, to elucidate the direct driving mechanisms of altered lipidomes due to treatments or disease progression. To demonstrate the value of LipidSIM, we first applied it to hepatic lipidomics following Keap1 knockdown and found that changes in mRNA expression of the lipid pathways were consistent with the LipidSIM-predicted fluxes. Second, we used it to study lipidomic changes following intraperitoneal injection of CCl4 to induce fast NAFLD/NASH development and the progression of fibrosis and hepatic cancer. Finally, to show the power of LipidSIM for classifying samples with dyslipidemia, we used a Dgat2-knockdown study dataset. Thus, we show that as it demands no a priori knowledge of enzyme kinetics, LipidSIM is a valuable and intuitive framework for extracting biological insights from complex lipidomic data.

Artificial Intelligence Applications in Clinical Chemistry

Artificial intelligence (AI) applications are an area of active investigation in clinical chemistry. Numerous publications have demonstrated the promise of AI across all phases of testing including preanalytic, analytic, and postanalytic phases; this includes novel methods for detecting common specimen collection errors, predicting laboratory results and diagnoses, and enhancing autoverification workflows. Although AI applications pose several ethical and operational challenges, these technologies are expected to transform the practice of the clinical chemistry laboratory in the near future.