Scholarlink

Background <p>Dysregulated lipid metabolism is common in patients with gastrointestinal (GI) cancer. This study investigated the ability of plasma lipidome variation during systemic treatment to classify GI cancer patients according to tumor response.</p> Methods <p>Non-targeted plasma lipidomics with LC-qTOF mass spectrometry was conducted in patients with advanced GI adenocarcinomas before and after three months of systemic standard-of-care antitumor treatment. Bayesian ANOVA with repeated measures and an uninformative prior was used to screen for analytes whose before-and-after variation could be associated with tumor response. Suitable candidates with a Bayes Factor (BF) &gt; 7 for the interaction were used to train a KNN model to classify patients as responders (PR + CR) <i>vs</i>. non-responders (PD + SD) according to RECIST 1.1 criteria.</p> Results <p>Thirty patients were included (18 colorectal, 6 gastric, 6 biliopancreatic). The cohort included 10 responders (all partial responses, 33%) and 20 non-responders (14 stable disease, 6 progressive disease). Plasma lipidomics identified 262 analytes on 10 lipid species: phosphatidylcholines (PCs), non-polar and polar lyso-PCs, sphingomyelins, lysophosphatidylethanolamines, cholesterol esters, acylglycerides, fatty acids, hormones, and bile acids. An increase in abundance after treatment of five PCs (PC32:2, PC33:2, PC36:2, PC36:5, PC38:5) was associated with tumor response (BF &gt; 7 for interaction). After excluding collinear PCs (retaining PC32:2, PC33:2, PC36:2), a KNN model (optimized: <i>k</i> = 7, Manhattan distance, inverse weighting, LOOCV) achieved consistent performance over 20 runs (mean ± SD): AUC 0.86 ± 0.23; Precision 0.76 ± 0.17; Recall 0.81 ± 0.15; F1 score 0.80 ± 0.13; MCC 0.66 ± 0.23. None of the other nine lipid classes contributed more than one analyte with a BF &gt; 7 for the interaction with tumor response.</p> Conclusions <p>Tumor response is associated with variations in plasma lipidome in patients with advanced GI cancer. The plasma abundance of PCs increased after treatment in responder patients, suggesting that further investigation may be warranted on PCs as a potential biomarker.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Bayesian selection of lipidome dynamics features for a K-nearest neighbors (KNN) classifier model of tumor response in patients with advanced gastrointestinal adenocarcinoma undergoing systemic treatment

  • 期刊论文

  • 发表日期：

    2025年12月17日

查看全文
  • Vicente Valentí,
  • Javier Ramos,
  • Laura Fernández-Sénder,
  • Óscar Villuendas,
  • Begoña Rodríguez,
  • Eugenia Sopena,
  • Marta Peña,
  • Carlos Alonso-Villaverde

摘要

Background

Dysregulated lipid metabolism is common in patients with gastrointestinal (GI) cancer. This study investigated the ability of plasma lipidome variation during systemic treatment to classify GI cancer patients according to tumor response.

Methods

Non-targeted plasma lipidomics with LC-qTOF mass spectrometry was conducted in patients with advanced GI adenocarcinomas before and after three months of systemic standard-of-care antitumor treatment. Bayesian ANOVA with repeated measures and an uninformative prior was used to screen for analytes whose before-and-after variation could be associated with tumor response. Suitable candidates with a Bayes Factor (BF) > 7 for the interaction were used to train a KNN model to classify patients as responders (PR + CR) vs. non-responders (PD + SD) according to RECIST 1.1 criteria.

Results

Thirty patients were included (18 colorectal, 6 gastric, 6 biliopancreatic). The cohort included 10 responders (all partial responses, 33%) and 20 non-responders (14 stable disease, 6 progressive disease). Plasma lipidomics identified 262 analytes on 10 lipid species: phosphatidylcholines (PCs), non-polar and polar lyso-PCs, sphingomyelins, lysophosphatidylethanolamines, cholesterol esters, acylglycerides, fatty acids, hormones, and bile acids. An increase in abundance after treatment of five PCs (PC32:2, PC33:2, PC36:2, PC36:5, PC38:5) was associated with tumor response (BF > 7 for interaction). After excluding collinear PCs (retaining PC32:2, PC33:2, PC36:2), a KNN model (optimized: k = 7, Manhattan distance, inverse weighting, LOOCV) achieved consistent performance over 20 runs (mean ± SD): AUC 0.86 ± 0.23; Precision 0.76 ± 0.17; Recall 0.81 ± 0.15; F1 score 0.80 ± 0.13; MCC 0.66 ± 0.23. None of the other nine lipid classes contributed more than one analyte with a BF > 7 for the interaction with tumor response.

Conclusions

Tumor response is associated with variations in plasma lipidome in patients with advanced GI cancer. The plasma abundance of PCs increased after treatment in responder patients, suggesting that further investigation may be warranted on PCs as a potential biomarker.