Colorectal cancer (CRC) remains a major cause of cancer mortality, creating an ongoing need for robust prognostic biomarkers that are grounded in biology. This study aimed to discover and validate a compact, mechanistically coherent tissue-based gene signature for CRC prognosis by integrating bulk transcriptomics, single-cell RNA sequencing, and machine learning approaches.
The investigators used paired tumor and adjacent normal tissues from 15 CRC patients together with public transcriptomic datasets (TCGA and GSE231559) to screen candidate genes. Two machine learning feature-selection algorithms—LASSO and SVM-RFE—were applied to derive a minimal predictive panel. Experimental validation included quantitative PCR (qPCR) in the paired-sample cohort. Single-cell RNA-seq analysis and immune deconvolution methods were used to explore cellular localization and tumor microenvironment correlations. The authors also constructed a clinically interpretable nomogram and assessed its calibration.
Using the combined analytic pipeline, three genes—ABCE1, ATAD5, and FUT4—were selected as a concise signature. These genes emerged from integrative analysis of patient samples and public datasets and were prioritized by both LASSO and SVM-RFE, indicating consistent selection across machine learning approaches.
The three genes were reported as consistently upregulated in CRC tumor tissues compared with adjacent normal tissues. Upregulation was confirmed by qPCR in the study’s paired-sample cohort. The consistent overexpression across datasets and experimental validation supported the robustness of the signature as a tumor-associated molecular classifier.
The signature genes were linked to complementary oncogenic processes:
ABCE1: associated with protein synthesis and immune modulation pathways, suggesting roles in translational control and interactions with immune components.
ATAD5: implicated in the DNA replication stress response, consistent with a role in genomic stability and replication-associated repair mechanisms.
FUT4: connected to cell adhesion and immune evasion, pointing to potential involvement in cell–cell interactions and modulation of anti-tumor immunity.
Together, these functional annotations indicate that the signature spans metabolic/translation processes, genomic instability responses, and immune-related mechanisms relevant to CRC biology.
Single-cell RNA sequencing localized expression of the signature genes to specific cellular compartments within the tumor ecosystem; the abstract reports that single-cell analysis clarified the cellular distribution of the three genes. Immune deconvolution of bulk transcriptomes revealed that the signature correlated with a macrophage-dominated tumor microenvironment, implying an association between signature expression and innate immune infiltration.
The three-gene signature demonstrated high diagnostic performance, with an area under the receiver operating characteristic curve (AUC) reported as 0.85. Prognostically, the signature stratified patients into distinct risk groups with divergent survival outcomes, indicating its potential utility for risk classification in CRC patients.
A clinically interpretable nomogram incorporating the three-gene signature was developed to facilitate individualized risk prediction. The nomogram was reported to have excellent calibration, suggesting agreement between predicted and observed outcomes in the evaluated datasets. The authors propose that this compact, biologically informed panel could serve as a tissue-based molecular classifier for CRC and may point to actionable axes—metabolism/translation, genomic instability, and immune suppression—for therapeutic development.
The abstract highlights discovery and validation across paired clinical samples and public cohorts, machine learning selection, single-cell localization, and immune correlation. Specific details such as cohort sizes beyond the 15 paired samples, independent external validation cohorts, follow-up duration for survival analyses, and statistical model coefficients were not reported in the abstract and therefore are not described here. The authors conclude that the three-gene signature—ABCE1, ATAD5, FUT4—is a promising prognostic marker for CRC with mechanistic links to core tumor processes and potential translational relevance.