HCsig

Copy number signatures in cervical samples enable early detection of high-grade serous ovarian carcinoma

Introduction

Ovarian cancer remains a leading cause of mortality, with nearly two-thirds of patients succumbing to the disease due to late-stage diagnosis and a lack of effective screening methods. Early detection is crucial, as women diagnosed with localized disease have a survival rate exceeding 90%. Most high-grade serous tubo-ovarian carcinomas (HGSC) originate from epithelial cells in the fallopian tubes, with precursor lesions such as serous tubal intraepithelial carcinomas (STICs) and earlier dysplastic cells harbouring TP53 mutations. These findings suggest that early tumor-driving mutations may be detectable in asymptomatic women. Using shallow whole-genome sequencing (sWGS), we investigated CNAs in cervical samples from HGSC patients and controls. We applied HGSC-derived copy number feature components and developed novel signatures (HCsig) using the machine learning-based unsupervised clustering SRIQ method. A centroid-based prediction model based on HCsig achieved promising sensitivity and specificity in an external validation cohort. These findings highlight the potential of sWGS on standard cervical samples for non-invasive early detection of HGSC.

Bioinformatic workflow

BINP52_CNA_Framework, a pipeline to generate copy number profiles and detect copy number signatures from shallow whole genome sequencing (sWGS) samples.

The version of tools and packages to be used will be specified in each step (see Chapter 3). The scripts within the pipeline are based on Snakemake (6.15.1), Python (v3.11.6), R (v4.3.2), and Java (JDK 11).

• (1) Preprocessing. This step includes quality assessment and quality trimming on the raw reads. (Fastp will be used for QC and trimming, together with fastqc and multiQC to generate the QC reports.)

• (2) Alignment. The human reference genome will be indexed. The reads will be mapped to the reference genome. (BWA will be used for both indexing and alignment.)

• (3) Clean-up. After alignment, the SAM files will be sorted and the PCR duplicates will be marked and removed. Also, the .sorted.deduplicated.sam will be converted to BAM files. GATK software (v4.1) was applied for base quality score recalibration (BQSR). The BAM files will be indexed for later analysis. (Picard will be used for sorting SAM, marking duplicates, removing duplicates and converting SAM to BAM. samtools will be used for generating the clean_up stats and for indexing the BAM files.)

• (4) Relative copy number profile. The BAM files will be analyzed through fixed-size binning, filtering, correction, and normalization to generate the read counts per bin. This data will then be used for the segmentation of bins and for generating the relative copy number profile. (QDNAseq will be used for this step.)

• (5) Ploidy and cellularity solutions. The output file from QDNAseq contains a relative copy number, and we need to estimate ploidy and cellularity in our samples to generate our final absolute copy number profile for comparison. (Rascal will be used for this step to find the solutions that best fit our study samples.)

• (6) Absolute copy number profile. We will further use other information (such as TP53 allele frequency) to infer the tumour fraction to select the best ploidy and cellularity solution. We apply this best solution to our relative copy number profile and generate the final absolute copy number profile for each sample. (Rascal will be used for this step.)

• (7) Comparison with the recent HGSC signatures (n=7). The functions should be loaded from the github repository: https://bitbucket.org/britroc/cnsignatures.git .

• (8) Comparison with the Pan-Cancer signatures (n=17). The package CINSignatureQuantification will be used to generate the samply-by-component matrix for the Pan-Cancer chromosomal instability signatures.

• (9) Comparison with the panConusig signatures (n=25). Tools including Battenberg (alleleCounter, impute2 and beagle5 were included in this package), ASCAT.sc and panConusig will be used in this step.

• (10) Generate and validate HCsig signatures (n=5) and the HGSC prediction model. Tools including SRIQ clustering, and CentriodClassification will be used in this step.

Analysis steps predicting HCsig:

(1) Preprocessing, (2) Alignment, (3) Clean-up, (4) Relative copy number profile, (5) Ploidy and cellularity solutions, (6) Absolute copy number profile,

snakemake --use-conda  --configfile config/config.yaml --cores 30  --snakefile workflow/Snakemake_HCsig_Pipeline.smk

(7) Generate sample-by-component matrix using the GitHub package CNsignatures

snakemake --use-conda  --configfile config/config.yaml --cores 30  --snakefile workflow/Snakemake_QDNASeq_RASCAL_CN_matrix.smk

(10) Predict HCsig signatures (n=5) using CentriodClassification.

java -jar -Xmx10g CentriodClassification.jar /path/to/input_folder/ sample-by-component_matrix_filename (without extension) /path/to/output_folder/