Machine learning links PDF to cold colorectal cancer

A peer-reviewed experimental study combining colorectal cancer bioinformatics with machine learning has identified peptide deformylase, or PDF, as a candidate protein biomarker associated with immune-cold colorectal cancer. Using TCGA-COAD data, the researchers applied support vector machines, random forests and XGBoost to distinguish molecular patterns associated with immune-hot and immune-cold tumors, then assessed the candidate marker in the independent CPTAC-2 proteomics cohort. The study reports that high PDF expression was associated with poorer prognosis, low immune-cell infiltration and elevated oxidative phosphorylation, or OXPHOS. The authors also proposed a PCBP1–PDF–OXPHOS regulatory axis through which PDF may connect tumor metabolism with an immune-suppressive environment.

The investigation addressed the pronounced immune heterogeneity of colorectal cancer by examining molecular differences between immune-hot and immune-cold tumor subtypes. Differential-expression analyses were combined with multiple machine-learning frameworks, including support vector machines, random forests and XGBoost, to stratify colorectal tumors. CPTAC-2 supplied an independent proteomics cohort for evaluating the protein-level association. According to the study, PDF emerged as a candidate indicator of the cold phenotype, with higher expression linked to unfavorable clinical outcomes, diminished immune infiltration and an OXPHOS-high metabolic profile. The cross-cohort analysis therefore linked a computational subtyping result with a protein-level marker in a separate dataset, although the work did not establish PDF as a clinical test or demonstrate that it predicts response to immunotherapy.

Beyond the biomarker association, the researchers proposed a mechanism that could explain how PDF-high tumors may develop a metabolically active yet immune-excluded state. In the PCBP1–PDF–OXPHOS axis, PCBP1-associated regulation of PDF is described as supporting tumor bioenergetics through OXPHOS and limiting the accumulation of N-formyl peptides. The study further associates N-formyl peptides with recruitment of CD8-positive T cells and macrophages. Their reduction was interpreted as a possible contributor to suppression of immune-cell recruitment, providing a functional explanation for the observed links among PDF expression, oxidative phosphorylation and low immune infiltration. This mechanism remains a research finding rather than a demonstrated treatment strategy; the reported study does not show that targeting PDF or the proposed axis changes immunotherapy outcomes in patients.

The results provide a testable model for further research into metabolic and immune features of colorectal cancer. The machine-learning component was used to identify molecular distinctions between immune-hot and immune-cold subtypes, while the proteomics analysis supported PDF as a candidate protein-level marker of the cold phenotype. The observed relationships among PDF, prognosis, OXPHOS and immune infiltration are associations derived from the analyzed cohorts, and the proposed regulatory axis should not be read as proof that PDF causes immune resistance in people. The evidence also does not establish that PDF inhibition can reprogram cold tumors, that high PDF expression identifies patients with a different response to existing therapies or that the marker is ready for diagnostic use. The study’s supported conclusion is narrower: elevated PDF expression marks a colorectal cancer state associated with reduced immune infiltration, elevated oxidative phosphorylation and poorer prognosis, while the PCBP1–PDF–OXPHOS axis offers a mechanistic hypothesis for further biological and clinical testing.

Subject of Research: Machine-learning identification of peptide deformylase as a candidate biomarker for immune-cold colorectal cancer

Article Title: Identification PDF as a biomarker of “cold” CRC using integrated bioinformatics and machine learning

Article References: Identification PDF as a biomarker of “cold” CRC using integrated bioinformatics and machine learning. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: colorectal cancer, peptide deformylase, immune-cold tumors, machine learning, proteomics, oxidative phosphorylation, immune infiltration, TCGA-COAD, CPTAC-2, tumor metabolism, Identification, biomarker