Genomic signatures in cell-free DNA and their role in early detection of cancer

Cancer can be treated with curative intent when diagnosed in its early stages. Attention has focussed increasingly on the use of circulating tumour DNA (ctDNA). Work by our group and others has pushed the boundaries of information that can be obtained by analysing cell-free DNA.

Detection accuracy depends on the burden of disease, with 1 tumour genome present per 5 ml of plasma for patients with early stage disease. To mitigate this, recent efforts attempt to improve detection by collecting information across multiple genomic loci.

Research groups including our own have developed methods that leverage low-depth whole genome sequencing (WGS) and use patterns such as copy number changes, differences in genomic coverage, and features of DNA fragmentation (“fragmentomics”) to develop machine learning (ML) classification models.

In this primer project, we aim to improve performance of classification models based on WGS data by adding information from mutation signature patterns: groups of mutations associated with cancer aetiology, risk factors and exposures.

We propose to investigate mutation signatures in ctDNA using a combination of available data novel data we will generate.

We will seek to identify patterns related to mutation signatures in low-depth (0.4x) data we have previously generated across hundreds of samples; public datasets at ~1x coverage; and additional WGS data we will generate at deeper coverage (~15x) through this award. We will develop algorithms to reduce background noise and extract mutational signature information.

This will be used to develop ML classifiers to detect cancer DNA.

Classifiers will be tested in cross-validation and on held-back data, and will be combined with other metrics we previously developed to quantify copy-number changes and fragmentation patterns in cell-free DNA.

Based on learnings, we will generate a further preliminary validation dataset from approx. 100 samples, focussing on plasma samples from patients with early-stage cancer.

If successful, these results can be used in other Early Detection programmes and may offer opportunities for expansion or validation as a future project proposal.

In addition to improvements to ctDNA detection, this proposed project aims to support training and development of bioinformatic skills.

The named researcher has a background in physics, and the primer award proposal intends to offer her an opportunity to improve her skills in machine learning and to transition into computational cancer genomics.

This application therefore addresses key gaps in methods and skills that are essential for effective development and future research in early detection of cancer.