ATAClone: Cancer Clone Identification and Copy Number Estimation from Single-cell ATAC-seq

The paper introduces ATAClone, a novel computational tool that accurately identifies cancer clones and estimates absolute copy number variations from single-cell ATAC-seq data by leveraging stably accessible genomic regions and automatic resolution optimization, thereby improving the disentanglement of genetic and non-genetic factors in tumor heterogeneity.

The Gist

Imagine you are a detective trying to solve a mystery inside a chaotic city (a tumor). This city is made up of millions of citizens (cells). Some citizens are the "good guys" (healthy cells), but many are "rebels" (cancer cells).

The problem is that these rebels aren't all the same. They belong to different gangs, or clones. Each gang has a unique history: some have stolen extra bags of gold (extra DNA), while others have lost their wallets (missing DNA). These differences in their "wealth" (DNA copy number) are what make them distinct gangs.

However, there's a catch. When scientists try to look at these cells using a special camera (single-cell ATAC-seq), the picture is blurry. The camera sees that some cells are "louder" or "brighter" than others. Usually, scientists think this means the cells are doing different things (like singing different songs). But often, the cells are just singing the same song; they just have different amounts of amplifiers (different amounts of DNA).

If you don't account for this, you might think two gangs are totally different when they are actually the same gang, just with different bank accounts. Or, you might miss the fact that one gang is actually two different gangs that just happen to look similar right now.

Enter: ATAClone (The Smart Detective)

The authors of this paper, Lachlan Cain and Anna Trigos, built a new tool called ATAClone. Think of it as a super-smart, automated detective that can sort through the chaos of the tumor city and figure out exactly which gangs exist and how much "wealth" (DNA) each one has.

Here is how ATAClone works, using simple analogies:

1. The "Stable Neighborhood" Filter

Imagine trying to count how many people live in a neighborhood, but the houses keep changing their paint colors every day. It's hard to tell if a house is new or just repainted.

• The Old Way: Scientists looked at every open door in the city. But in cancer, doors open and close randomly based on what the cell is doing, which creates noise.
• ATAClone's Way: It only looks at the "Stable Neighborhoods"—the doors that are always open, no matter what the cell is doing. By ignoring the doors that change, ATAClone can clearly see the size of the house (the DNA amount) without being distracted by the noise.

2. The "Automatic Sorting Hat"

Usually, to find the gangs, a scientist has to guess how many gangs there are and tell the computer how to group them. It's like a teacher trying to sort students into groups without knowing how many classes there are, and they have to guess the rules.

• ATAClone's Way: It runs a simulation (a practice run) first. It asks, "If there were no gangs, how would the computer sort these cells?" Then, it compares the real data to this "no-gang" simulation. It automatically figures out the perfect number of groups needed to spot the real differences, without the human having to guess. It's like a sorting hat that knows exactly when a student belongs to a new house.

3. Counting the "Total Gold" (Absolute Copy Number)

Most tools can tell you that Gang A has more gold than Gang B, but they can't tell you if Gang A has 3 bags or 6 bags. They only know the ratio.

• ATAClone's Way: It looks at the total weight of the DNA in the cell. If Gang A has twice as much total DNA as a healthy cell, and their gold distribution looks like it's been doubled, ATAClone can say, "Aha! This gang is polyploid (they have double the normal set of chromosomes)." It can count the actual number of bags, not just compare them to each other.

Why is this a Big Deal?

1. It's Automatic: You don't need to be a math wizard to use it. You just feed it the data, and it handles the messy parts (filtering bad cells, finding the right groups, and counting the DNA).
2. It's Accurate: The authors tested it against other tools (like a tool called RIDDLER) and against real "gold standard" data from bulk sequencing. ATAClone was much better at finding the true gangs and counting their DNA correctly.
3. It Reveals Evolution: By knowing exactly which gangs exist and how much DNA they have, scientists can reconstruct the "family tree" of the tumor. They can see how the cancer evolved, which gangs are becoming dominant, and why some might be resistant to drugs (often because they have extra DNA).

The Bottom Line

ATAClone is a new, automated tool that helps scientists stop getting confused by the "volume" of cancer cells and start understanding their true "identity." It separates the signal (the real genetic differences) from the noise (random biological changes), allowing researchers to map out the evolutionary history of a tumor with much greater clarity. It turns a blurry, confusing mess of data into a clear map of the criminal gangs inside a cancer.

Technical Summary

1. Problem Statement

Single-cell analyses of cancer often rely on unsupervised clustering to identify distinct cell populations. However, in cancer genomics, these clusters are frequently driven by DNA Copy Number Variants (CNVs) rather than true biological differences in gene expression or epigenetic regulation.

• Confounding Factors: Ignoring clonal CNVs can lead to misinterpretation of differential expression results and tumour heterogeneity studies.
• Limitations of Existing Tools:
  • Most tools require manual user input for quality control (QC), normalization, and clustering resolution.
  • They typically estimate only relative copy numbers, failing to detect ploidy differences (e.g., whole-genome doubling) or infer absolute copy numbers.
  • Existing methods often struggle to distinguish between technical noise and genuine clonal signals in scATAC-seq data.

2. Methodology: The ATAClone Workflow

ATAClone is an automated R package designed to infer clonal structure and estimate absolute copy numbers from single-cell ATAC-seq (scATAC-seq) data (both standalone and multiome). The workflow consists of four distinct stages:

A. Feature Creation

• Stably-Accessible Regions: Instead of sample-specific peak calling, ATAClone uses a pre-computed list of 76,951 stably-accessible regions (consensus peaks across 16 organ systems). This minimizes non-genetic signals (differential accessibility) to isolate copy number signals.
• Binning: Fragments overlapping these regions are aggregated into genomic bins (default target width: 10 Mb) to create a bin-by-cell barcode count matrix. This improves the signal-to-noise ratio.

B. Quality Control (QC)

ATAClone implements automated filtering based on several novel metrics:

• Empty Droplets & Debris: Filters based on total fragments in stably-accessible regions and "excess" zero bin counts (indicating random chromosome loss).
• Transposition Efficiency: Uses a Poisson regression approach to measure the fraction of fragments in accessible regions, identifying cells with low transposition efficiency.
• Barcode Sequence Bias: Identifies a systematic bias in 10X Genomics multiome assays where specific cell barcode sequences yield lower ATAC-seq coverage. ATAClone calculates a "barcode odds" metric to filter these low-coverage barcodes.

C. Clone Identification (Automated Clustering)

• Normalization: Applies a fractional power transformation to bin counts to correct for heteroscedasticity (Gamma-Poisson distribution), removing library size scaling.
• Dimensionality Reduction: Performs PCA, discarding components correlated with technical covariates (Total DNA, Transposition Efficiency).
• Graph-Based Clustering: Constructs a KNN graph and applies the Leiden algorithm.
• Automated Resolution Selection: A key innovation is the use of Monte Carlo simulations. ATAClone generates a "null" dataset (simulated without biological variation) to determine the optimal clustering resolution parameter. It selects the highest resolution where the Type I error rate (false positive clustering) remains below a threshold (e.g., 0.05).

D. Absolute Copy Number Estimation

• Reference Selection: Uses internal non-tumour cells (identified by flat relative copy number profiles) as a reference.
• Ploidy-Aware Fitting: Unlike tools that only provide relative ratios, ATAClone infers absolute copy numbers. It utilizes:
  • Total DNA content differences between cells.
  • Clone-to-clone copy number differences.
  • A weighted least-squares approach to fit integer copy number values, even in the presence of polyploidy (e.g., distinguishing between a diploid gain and a tetraploid baseline).

3. Key Contributions

1. Fully Automated Pipeline: Eliminates the need for manual QC thresholds and clustering resolution tuning, reducing user bias and increasing reproducibility.
2. Absolute Copy Number & Ploidy Inference: Capable of detecting whole-genome doubling and distinguishing clones with different ploidy levels, a feature missing in most existing scATAC-seq tools.
3. Stably-Accessible Regions: Introduces a robust feature selection strategy that decouples copy number estimation from cell-type-specific chromatin accessibility.
4. Novel QC Metrics: Identifies and corrects for a previously unreported systematic bias in 10X multiome assays related to cell barcode sequences.
5. Simulation-Based Clustering: Provides a statistically rigorous method for automatically selecting clustering resolution to control Type I errors.

4. Results & Validation

The authors validated ATAClone using multiple datasets:

• Reproducibility (Kidney Cancer Replicates): Applied to four kidney cancer replicates with different nuclear isolation protocols. ATAClone consistently identified the same major clones and technical artifacts across protocols, demonstrating robustness to sample preparation differences.
• Sensitivity (scmixology2 Lung Cancer Mixture): Tested on a mixture of 5 lung cancer cell lines. ATAClone achieved high homogeneity (0.97), correctly grouping cells by cell line. It successfully identified sub-clones within cell lines driven by specific CNVs (e.g., whole chromosome deletions/amplifications).
• Specificity (Non-Cancer Control): Applied to normal PBMCs. While it identified 4 clusters, these were driven by biological cell types rather than CNVs, and the estimated copy number differences were sub-integer and non-contiguous, confirming low false-positive rates for clonal CNV calling.
• Comparison with RIDDLER: In a metastatic prostate cancer sample, ATAClone identified fewer, more biologically coherent clusters separated by large chromosomal events compared to RIDDLER, which produced sparse, fragmented CNV calls.
• Accuracy vs. Bulk WGS: In a cohort of 22 metastatic prostate cancer samples with matched bulk Whole Genome Sequencing (WGS), ATAClone's copy number estimates showed a high Pearson correlation (mean 0.868) with bulk-derived estimates (PURPLE pipeline), significantly outperforming RIDDLER (mean 0.665).
• Ploidy Detection: Successfully identified samples with mixed ploidy (e.g., clones with different total DNA content but similar relative profiles), inferring whole-genome doubling events.

5. Significance

• Disentangling Genetics from Epigenetics: ATAClone allows researchers to separate the effects of genetic CNVs from non-genetic regulatory changes, leading to more accurate interpretations of tumour heterogeneity and evolution.
• Evolutionary Insights: By detecting ploidy changes and absolute copy numbers, the tool provides deeper insights into the evolutionary history of tumours, including the timing of whole-genome doubling events.
• Clinical Relevance: Improved detection of clonal diversity and ploidy is crucial for understanding drug resistance mechanisms and tailoring cancer therapies.
• General Utility: The automated approaches for normalization, bias correction, and clustering resolution selection introduced in ATAClone offer general advancements applicable to broader single-cell sequencing analyses.

Availability: The tool is implemented as an R package available at https://github.com/TrigosTeam/ATAClone.