The impact of leukocyte-associated RNA retention on platelet RNA-based cancer detection.

Platelet RNA profiling is a promising liquid biopsy approach, offering minimally invasive insights into systemic diseases, such as cancer. Yet, the effect of leukocyte-associated RNA retention on platelet transcriptomic signatures remains unclear. This study provides the first comprehensive evaluation of how leukocyte-associated RNA retention influences platelet RNA sequencing and machine learning (ML)-based cancer detection. Using the largest publicly available platelet transcriptomic dataset (2351 human RNA samples) and single-cell/single-platelet RNA sequencing from 29 donors, we quantified leukocyte-associated RNA retention in matched sample groups (low vs. high retention) via a novel 132-gene leukocyte marker panel. Diagnostic implications were assessed using 10 ML algorithms for binary classification (cancer vs. healthy), examining the contribution of leukocyte-derived transcripts. Interestingly, models trained on high-retention samples often showed slightly improved performance, suggesting complementary diagnostic value from leukocyte signals. Importantly, post hoc correction using remove unwanted variation (RUV) and linear modeling effectively mitigated retention effects. However, adding the leukocyte-associated RNA retention score to the full-cohort classifier changed cancer-detection AUC only minimally (0.9247 to 0.9249). Thus, whether this signal should be corrected, recalibrated, or reported should depend on the study question rather than assuming routine removal.