Deciphering cancer origin from DNA evolution

The National Institutes of Health National Cancer Institute awarded The Trustees of Columbia University in the City of New York $428,975 on April 17, 2026, for Project Grant R01CA310281 under the Cancer Detection and Diagnosis Research program (CFDA 93.394).

Columbia will develop DECODE (Deciphering Cancer Origin from DNA Evolution), a novel computational approach for reconstructing tumor evolution from DNA sequencing data to predict patient outcomes and facilitate risk stratification and treatment selection. The method integrates the neutral tail in the site frequency spectrum—comprising recent mutations that signal a tumor's expansion mode—and applies mathematical corrections for sample-specific coverage and mutation-filtering effects to reduce overfitting, enhance detection of overlapping clusters, and optimize performance with current DNA-sequencing technologies. Columbia will apply DECODE to analyze tumor heterogeneity and infer tumor evolution history across primary cancers from the International Cancer Genome Consortium and The Cancer Genome Atlas, characterizing the site frequency spectrum neutral tail for each sample, estimating subclone numbers and frequencies, and inferring tumor evolution timelines using a mathematical framework to estimate tumor age, subclonal arrival times, and growth rates.

Performance occurs at Columbia University in New York, New York. The project period runs through March 31, 2031. The award is classified as a Project Grant, a research assistance type supporting investigator-initiated scientific inquiry consistent with NIH peer review and merit-based competition.

Abstract

DNA sequencing holds significant promise for precisely profiling individual tumors, which in turn can guide personalized cancer therapies and improve long-term patient outcomes. A key prognostic factor across tissue types is intra-tumor heterogeneity (ITH), resulting from somatic single-nucleotide variants and copy-number aberrations, that is highly predictive of patient survival and risks of metastasis and relapse. However, accurately estimating ITH and tumor growth from DNA data remains challenging, thereby impeding its broader adoption for clinical decision making. This proposed project will focus on developing decode (deciphering cancer origin from DNA evolution), a novel approach for reconstructing tumor evolution that can predict patient outcomes and facilitate risk stratification and treatment selection in clinical practice (aim 1). Decode will integrate the neutral tail in the site frequency spectrum (SFS), comprised of recent mutations that carry signals indicative of a tumor's expansion mode. Decode's enhanced accuracy will be based on mathematical corrections for sample-specific coverage and effects of mutation filtering. This will reduce the risk of overfitting, enhance the detection of overlapping clusters, and render decode ideally suited for data generated by current DNA-SEQ technologies. In aim 2, decode will be used to analyze the tumor heterogeneity and infer tumor evolution history in primary cancers from the international cancer genome consortium (ICGC) and the cancer genome atlas (TCGA). For each sample, the SFS neutral tail will be characterized and the number of subclones and subclonal frequencies will be estimated. The tumor evolution timelines from these samples will then be inferred, using a novel mathematical framework to estimate the age of each tumor's most recent common ancestor, subclonal arrival times and growth rates from decode's parameters. Finally, the ITH estimates and tumor and subclonal growth rates will be examined for predictive power of patient outcome in specific cancer types, toward finding potential biomarkers. Relapsed and metastatic tumors from the personalized oncogenomics (POG) cohort will also be studied (aim 3). For these samples, a multi-variate decode algorithm will be developed to consider mutations with distinct multiplicities occurring on different copy number backgrounds. The multi-variate decode will be able to analyze advanced cancers, which often exhibit variable copy numbers due to chromosomal instability. The comparison of ITH estimates and clonal evolutions inferred from decode between samples in POG and ICGC/TCGA will enable a deeper understanding into differences in selection forces driving relapses and metastases compared to DE novo cancers. In summary, decode is a novel approach toward reliable tumor evolution reconstruction, enabled by advances in mathematical modeling and parameter inference. Its accuracy, notably for samples sequenced with currently available technologies, will enable deeper understanding of tumor adaptation, clinical patient stratification and outcome prediction, toward the goal of personalized therapy treatment.