Background <p>This study aims to identify branched-chain amino acid (BCAA) plasma metabolites and gene signatures that enhance prognostic assessments in non-small cell lung cancer (NSCLC) patients receiving immunotherapy.</p> Methods <p>Plasma metabolites were measured using untargeted UPLC-MS/MS (<i>n</i> = 94 and 40), with lymphocyte subset tests on 72 patients. BCAA-related subtypes were identified in NSCLC datasets (<i>n</i> = 274, 176, and 196). A prognostic risk model was developed and validated in NSCLC (<i>n</i> = 16, 27, 24, and 339), melanoma (<i>n</i> = 25), and pan-cancer ICIs cohorts (<i>n</i> = 330 and 81). Immune cell infiltration and prognostic signatures were validated using mIF (<i>n</i> = 21 in CHCAMS), scRNA-seq (<i>n</i> = 8 and 21), and spatial transcriptomics (<i>n</i> = 2 and 6). Cell and animal experiments involving <i>HMGCS1</i> were conducted in a lung cancer model. Additionally, based on our previous findings that B cells with higher malignancy exhibited enhanced cholesterol homeostasis pathways in diffuse large B-cell lymphoma (DLBCL), we further analyzed the prognostic value of HMGCS1 using our spatial transcriptomics (<i>n</i> = 10) and immunohistochemistry (IHC, <i>n</i> = 39) in DLBCL.</p> Results <p>Our plasma metabolite analysis showed higher L-leucine levels were associated with better prognosis and had higher T cell counts and CD4⁺ T cell counts (<i>P</i> &lt; 0.05). In GEO datasets, four NSCLC subtypes were identified, showing distinct prognostic outcomes and tumor microenvironment. Five BCAA-related genes (<i>ACAT2, ALDH2, HMGCS1, MLYCD,</i> and <i>PPM1 K</i>) formed a prognostic risk model for NSCLC, validated through Kaplan–Meier and ROC curve analyses in ICI cohorts (<i>P</i> &lt; 0.05). <i>HMGCS1</i> was an independent prognostic value in ICI cohorts and was negatively correlated with CD8⁺ T cell infiltration, while positively correlating with tumor severity, cholesterol homeostasis, and BCAA degradation across multiple platforms, including GEO datasets, our mIF cohort, public scRNA-seq, and spatial transcriptomics (<i>P</i> &lt; 0.05). And our cell and animal function experiments found <i>HMGCS1</i> overexpression promotes metabolic pathways and accelerates tumor growth, whereas <i>HMGCS1</i> knockdown suppresses tumor progression in a mouse model treated with PD-1 monoclonal antibody (<i>P</i> &lt; 0.05). In DLBCL, high <i>HMGCS1</i> expression was associated with shorter overall survival, enriched in B cells and relapsed patients, correlated with cholesterol homeostasis and amino acid degradation pathways, and its prognostic value was further validated at the protein level by our IHC cohort (<i>P</i> &lt; 0.05).</p> Conclusions <p>This study identifies a BCAA-related plasma metabolites and gene signature as effective prognostic markers for NSCLC patients receiving immunotherapy, with <i>HMGCS1</i> as a key prognostic factor influencing tumor progression and immune response.</p>