Abstract 7824: Plasma-only classification of CHIP, low-VAF germline, and somatic variants enables accurate tumor-fraction estimation without matched normal samples.

Abstract Background: Clonal hematopoiesis (CHIP) and germline variants commonly appear in cell-free DNA (cfDNA) and confound tumor genotyping. We analyzed historical paired plasma and Peripheral Blood Mononuclear Cell (PBMC)/buffy coat samples to (i) quantify CHIP prevalence and variant allele frequency (VAF) distributions, (ii) characterize low-VAF germline signals attributable to copy number variation (CNV) or alignment artifacts, (iii) identify tumor-derived somatic variants by integrating fragmentomics, CNV context, and longitudinal VAF dynamics, and (iv) benchmark plasma-only tumor-fraction (TF) estimation. Data were generated with the PredicineCARE and PredicineATLAS assays. Methods: We retrospectively profiled ∼1,000 plasma samples spanning prostate, breast, colorectal, lung, pancreatic, and other solid tumors with matched PBMC/buffy coat specimens. UMI-aware pipelines called single nucleotide variants/insertions/deletions/CNVs. Variants were annotated for CHIP drivers (e.g., DNMT3A, TET2, ASXL1, PPM1D, TP53, SF3B1/SRSF2/U2AF1, JAK2), population AF, hotspots, and an in-house knowledge base. Low-VAF germline events were adjudicated using a genome-wide germline SNP skeleton to explain deviations from the ∼50% heterozygous expectation under local CNV. We trained and locked a plasma-only classifier (CHIP/germline/somatic) and estimated mutation-derived TF from tumor-assigned variants, benchmarking both against matched-normal labels. Longitudinal analyses (baseline + follow-up, months apart) assessed VAF dynamics and further improved the tumor fraction detection sensitivity. Results: CHIP was prevalent and dominated by DNMT3A/TET2/ASXL1, with a minority of high-VAF clones; VAF distributions were summarized with and without TF normalization. Low-VAF germline signals were frequent but often CNV-driven, resolved by the SNP-skeleton model. On single-timepoint plasma, 95% of CHIP variants with VAF 2% were correctly classified. Incorporating longitudinal VAF dynamics further differentiated low-VAF CHIP when TF changed 2-fold between draws, while tumor-derived variants tracked response/progression. The plasma-only TF showed excellent concordance with matched-normal TF (concordance correlation coefficient 0.95). Conclusions: CHIP is common and often low-VAF; a non-trivial fraction of apparent low-VAF germline findings reflect copy-number-induced VAF shifts. A plasma-only strategy that integrates CNV-aware germline modeling, CHIP gene context, fragmentomics, and longitudinal VAF dynamics closely reproduces matched-normal truth, enabling accurate somatic calling and tumor-fraction estimation without matched PBMC/buffy coat. Citation Format: Tiantian Zheng, Yong Huang, Chao Dai, Junmei Wang, Xiaohong Wang, Pan Du.. Plasma-only classification of CHIP, low-VAF germline, and somatic variants enables accurate tumor-fraction estimation without matched normal samples abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 7824.