<![CDATA[Background. Chemotherapy response is highly variable, leading to ineffective treatment and toxicity. Reliable, non-invasive biomarkers to predict response a priori are urgently needed. Circulating microRNAs (miRNAs) are stable liquid biopsy candidates, but previous studies often lack robust validation. Purpose. This study aimed to identify and validate a novel, non-invasive circulating miRNA signature to accurately predict chemotherapy response using a large-scale bioinformatic approach. Method. A comprehensive in silico study was conducted. We aggregated and harmonized 948 patient samples from five public datasets (GEO, TCGA). A machine learning pipeline (LASSO + Random Forest) was applied to a Training Set (n=664) to discover a predictive signature. The signature was then validated in an Internal Testing Set (n=284) and a separate External Validation Cohort (n=120). Results. We aggregated and harmonized 948 patient samples from five public datasets (GEO, TCGA). A machine learning pipeline (LASSO + Random Forest) was applied to a Training Set (n=664) to discover a predictive signature. The signature was then validated in an Internal Testing Set (n=284) and a separate External Validation Cohort (n=120). We identified and validated a 7-miRNA circulating signature (c-miRSig). The model demonstrated high accuracy in both the internal (AUC 0.89) and external (AUC 0.86) validation sets. Conclusion. The signature was also a powerful prognostic tool, significantly stratifying patients for progression-free survival (p < 0.001). Functional analysis linked the signature to key chemoresistance pathways (PI3K-Akt, ABC transporters). The c-miRSig is a robust, non-invasive biomarker with dual predictive and prognostic power. This computationally validated signature provides a strong foundation for a clinically viable test to personalize chemotherapy, sparing non-responders from toxic, ineffective treatment.]]>
