Volume 361, 24 April 2026, 121243Author links open overlay panel, , , , , Highlights•

The clinical application of herbal medicine is challenged by considerable heterogeneity in therapeutic outcomes among patients.

•

Our analysis suggests that CX3CL1 is the candidate biomarker for predicting the efficacy of Macaranga sinensis Müll.Arg.

•

Future prospective studies with larger samples and multi-omics integration are warranted to validate these biomarkers.

•

Our work highlights integrating machine learning with proteomic data to identify predictive biomarkers of herbal medicine on clinical efficacy.

AbstractEthnopharmacological relevance

Despite the fact that herbal medicine has been used for a long time, their clinical application is challenged by unclear active ingredients and poorly understood mechanisms of action, resulting in considerable heterogeneity in therapeutic outcomes among patients.

Aim of the study

To explore the use of OLink-based biomarkers to predict the efficacy of Macaranga sinensis Müll.Arg in individuals with long COVID-related fatigue.

Materials and methods

This nested case-control study recruited long COVID participants who had routinely taken Macaranga sinensis Müll.Arg for more than three months. Case (response) and control (non-response) group were defined based on the change in Brief Fatigue Inventory (BFI) score. Plasma samples were analyzed using OLink. Mann–Whitney U test, Lasso and Ridge regression model, Random Forest, and Support Vector Machine (SVM) were used for differential expression analysis, feature selection, and biomarker identification, respectively.

Results

A total of 55 long COVID fatigue patients was included in this study (27 in case group and 28 in control group). Differential expression analysis filtered in a total of 13 potential biomarkers (log2 fold change >0.5; p < 0.05). Feature selection further selected 11 biomarkers, including uPA, TRAIL, IL-10, IL-18R1, CX3CL1, EPHB4, COL1A1, Flt3L, EGFR, IL-1RT2, and IFN-γ (all p < 1e-6). Random Forest and SVM identified the key biomarker of CX3CL1 (F1-score of 0.72).

Conclusions

Our exploratory analysis suggests that CX3CL1 is the candidate biomarker for predicting the efficacy of Macaranga sinensis Müll.Arg, warranting further investigation in larger studies. Our work highlights the translational potential of integrating statistical modeling and machine learning approaches with proteomic data to identify predictive biomarkers for herbal medicine efficacy in clinical settings.

Graphical abstractDownload: Download high-res image (292KB)Download: Download full-size imageKeywords

Herbal medicine

Botanical drug

Proteomics

Biomarker

OLink

Efficacy

AbbreviationsTCM

Traditional Chinese medicine

FDA

Food and Drug Administration

PEA

Proximity Extension Assay

PCFS

Post-COVID-19 Functional Status

BFI

Fatigue Inventory Form

C19-YRS

COVID-19 Yorkshire Rehabilitation Scale

ISI

Insomnia Severity Index

HADS

Hospital Anxiety and Depression Scale

SGRQ

St. George's Respiratory Questionnaire

SF-12

Short Form 12 health survey

NPX

Normalized Protein eXpression

SVM

Support Vector Machine

Data availability

Data is in supplement.

© 2026 The Authors. Published by Elsevier B.V.