Herbal Medicine in the Molecular Era: Integrating Traditional Chinese Herb Signatures with Chemotherapy Efficacy and Tumor Microenvironment Using Machine Learning Models

Traditional Chinese Medicine (TCM) offers a rich repertoire of bioactive compounds whose molecular targets can now be mapped systematically against highthroughput cancer data. Here, we combine networkbased herbgene associations with machine and deep learning to both predict therapeutic response and elucidate tumor microenvironment (TME) modulation in metastatic melanoma. First, we derived regulatory signatures for three herbsWuHuaGuo (Fig), JiuCengTa (Basil), and Codonopsis pilosula (DangShen)and mapped their putative gene targets onto TCGA melanoma transcriptomes. KaplanMeier and Cox analyses confirmed that high herb regulatory scores correlate with improved overall survival and enriched immune infiltration. Next, we trained four classifiers (logistic regression, random forest, knearest neighbors, and neural network) on GDSC cellline expression and logIC data to predict chemosensitivity using each herbs signature. Nonlinear models (RF, NNET) achieved robust discrimination (AUC>0.8) for DNA damagetargeting (PARP) and epigenetic (HDAC) inhibitors, highlighting the power of ensemble and deeplearning approaches to capture complex, nonlinear patterns in herbregulated pathways. To validate our in silico predictions, we treated B16F10 melanoma cells with Fig and Basil extracts, observing dosedependent growth inhibition (distinct IC values) and, in coculture assays, significant upregulation of Tcell activation markers CD44 and CD69. Together, these results showcase a unified computationalexperimental framework that leverages TCM herb signatures and modern machine learning to forecast drug response and enhance antitumor immunity, offering a blueprint for discovering novel adjuvant therapies in melanoma.