Non-small cell lung cancer (NSCLC) encompasses distinct histopathological subtypes, namely adenocarcinoma (AC) and squamous cell lung carcinoma (SCC), which require precise differentiation for effective treatment strategies. In this study, we present a novel molecular diagnostic model that integrates tissue-specific expression profiles of microRNAs (miRNAs) obtained through next-generation sequencing (NGS) to discriminate between AC and SCC subtypes of NSCLC. This approach offers a more comprehensive and precise molecular characterization compared to conventional methods such as histopathology or immunohistochemistry. Firstly, we identified 31 miRNAs with significant differential expression between AC and SCC cases. Subsequently, we constructed a 17-miRNA signature through rigorous multistep analyses, including LASSO/elastic net regression. The signature includes both upregulated miRNAs (hsa-miR-326, hsa-miR-450a-5p, hsa-miR-1287-5p, hsa-miR-556-5p, hsa-miR-542-3p, hsa-miR-30b-5p, hsa-miR-4728-3p, hsa-miR-450a-1-3p, hsa-miR-375, hsa-miR-147b, hsa-miR-7705, and hsa-miR-653-3p) and downregulated miRNAs (hsa-miR-944, hsa-miR-205-5p, hsa-miR-205-3p, hsa-miR-149-5p, and hsa-miR-6510-3p). To assess the discriminative capability of the 17-miRNA signature, we performed receiver operating characteristic (ROC) curve analysis, which demonstrated an impressive area under the curve (AUC) value of 0.994. Our findings highlight the exceptional diagnostic performance of the miRNA signature as a stratifying biomarker for distinguishing between AC and SCC subtypes in lung cancer. The developed molecular diagnostic model holds promise for providing a more accurate and comprehensive molecular characterization of NSCLC, thereby guiding personalized treatment decisions and improving clinical management and prognosis for patients.

Non-small cell lung cancer is the predominant form of lung cancer and is associated with a poor prognosis. MiRNAs implicated in cancer initiation and progression can be easily detected in liquid biopsy samples and have the potential to serve as non-invasive biomarkers. In this study, we employed next-generation sequencing to globally profile miRNAs in serum samples from 71 early-stage NSCLC patients and 47 non-cancerous pulmonary condition patients. Preliminary analysis of differentially expressed miRNAs revealed 28 upregulated miRNAs in NSCLC compared to the control group. Functional enrichment analyses unveiled their involvement in NSCLC signaling pathways. Subsequently, we developed a gradient-boosting decision tree classifier based on 2588 miRNAs, which demonstrated high accuracy (0.837), sensitivity (0.806), and specificity (0.859) in effectively distinguishing NSCLC from non-cancerous individuals. Shapley Additive exPlanations analysis improved the model metrics by identifying the top 15 miRNAs with the strongest discriminatory value, yielding an AUC of 0.96 ± 0.04, accuracy of 0.896, sensitivity of 0.884, and specificity of 0.903. Our study establishes the potential utility of a non-invasive serum miRNA signature as a supportive tool for early detection of NSCLC while also shedding light on dysregulated miRNAs in NSCLC biology. For enhanced credibility and understanding, further validation in an independent cohort of patients is warranted.

Non-small-cell lung cancer (NSCLC) poses a challenge due to its heterogeneity, necessitating precise histopathological subtyping and prognostication for optimal treatment decision-making. Molecular markers emerge as a potential solution, overcoming the limitations of conventional methods and supporting the diagnostic–therapeutic interventions. In this study, we validated the expression of six genes (MIR205HG, KRT5, KRT6A, KRT6C, SERPINB5, and DSG3), previously identified within a 53-gene signature developed by our team, utilizing gene expression microarray technology. Real-time PCR on 140 thoroughly characterized early-stage NSCLC samples revealed substantial upregulation of all six genes in squamous cell carcinoma (SCC) compared to adenocarcinoma (ADC), regardless of clinical factors. The decision boundaries of the logistic regression model demonstrated effective separation of the relative expression levels between SCC and ADC for most genes, excluding KRT6C. Logistic regression and gradient boosting decision tree classifiers, incorporating all six validated genes, exhibited notable performance (AUC: 0.8930 and 0.8909, respectively) in distinguishing NSCLC subtypes. Nevertheless, our investigation revealed that the gene expression profiles failed to yield predictive value regarding the progression of early-stage NSCLC. Our molecular diagnostic models manifest the potential for an exhaustive molecular characterization of NSCLC, subsequently informing personalized treatment decisions and elevating the standards of clinical management and prognosis for patients.

Drug discovery remains a time-consuming and costly process, necessitating innovative computational approaches to accelerate early stage target identification and compound development. We introduce AgentMol, a modular multimodel AI system that integrates large language models, chemical language modeling, and deep learning–based affinity prediction to automate the discovery pipeline. AgentMol begins with disease-related queries processed through a Retrieval-Augmented Generation system using the Large Language Model to identify protein targets. Protein sequences are then used to condition a GPT-2–based chemical language model, which generates corresponding small-molecule candidates in SMILES format. Finally, a regression convolutional neural network (RCNN) predicts the drug-target interaction by estimating binding affinities (pKi). Models were trained and validated on 470,560 ligand–protein pairs from the BindingDB database. The chemical language model achieved high validity (1.00), uniqueness (0.96), and diversity (0.89), whereas the RCNN model demonstrated robust predictive performance with R² > 0.6 and Pearson’s R > 0.8. By leveraging LangGraph for orchestration, AgentMol delivers a scalable, interpretable pipeline, effectively enabling the end-to-end generation and evaluation of drug candidates conditioned on protein targets. This system represents a significant step toward practical AI-driven molecular discovery with accessible computational demands.