Copy Number Variation Inference

The foundational approach that smooths gene expression signals across contiguous genomic regions to reduce the noise inherent in single-cell data.

• Mechanism: A sliding window of 50–100 genes averages relative expression values, transforming sparse transcriptomic signals into a continuous profile.
• Key Assumption: Chromosomal amplifications or deletions affect multiple adjacent genes, producing a sustained deviation from the diploid baseline.
• Example: inferCNV uses a 101-gene window with dynamic thresholding to call CNV states in tumor cells versus normal reference cells.
• Limitation: Window size selection creates a trade-off between spatial resolution and noise suppression; small windows retain false positives, while large windows blur breakpoint boundaries.

A probabilistic framework that models chromosomal segments as discrete copy number states with defined transition probabilities between adjacent genomic positions.

• Mechanism: An HMM treats the true copy number as a hidden state sequence and the observed expression deviations as emissions, applying the Viterbi algorithm to decode the most likely state path.
• Advantage: Explicitly models the biological expectation that CNV breakpoints are rare, penalizing implausible state transitions and reducing false-positive calls.
• Example: HoneyBADGER uses a hierarchical Bayesian HMM to integrate expression and allelic imbalance signals for joint CNV and loss-of-heterozygosity detection.
• Key Parameter: The transition probability matrix controls the expected CNV segment length; low transition probabilities favor fewer, longer segments.

A strategy that contrasts tumor or perturbed cells against a matched normal reference to isolate somatic copy number alterations from germline variation and technical artifacts.

• Mechanism: Gene expression in each cell is subtracted from the average expression of a panel of normal diploid cells, centering the baseline at zero and revealing deviations.
• Reference Selection: Normal cells can be intrinsic (e.g., immune infiltrate within the same sample) or extrinsic (e.g., a separate matched normal tissue dataset).
• Example: CopyKAT uses a Gaussian mixture model to initially separate aneuploid tumor cells from diploid stromal cells, then uses the diploid cluster as an internal reference for CNV inference.
• Critical Consideration: Contamination of the reference set with aneuploid cells produces a shifted baseline that masks true CNV signals.

A probabilistic approach that computes the posterior probability of each copy number state given the observed expression data, incorporating prior biological knowledge about CNV frequency and length distributions.

• Mechanism: Bayes' theorem combines a likelihood model (how expression varies with copy number) with a prior distribution (expected CNV patterns) to produce a posterior probability for each genomic segment.
• Advantage: Provides uncertainty quantification for each CNV call, enabling downstream filtering based on confidence thresholds rather than binary accept/reject decisions.
• Example: CONICSmat applies a Bayesian mixture model across chromosome-wide expression matrices, estimating the posterior probability that a given cell harbors a CNV at a specific locus.
• Prior Specification: Informative priors can encode known cytoband-level CNV hotspots from TCGA or COSMIC databases to improve sensitivity in commonly altered regions.

An enhancement that combines total expression shifts with allele-specific expression ratios to distinguish copy-neutral loss of heterozygosity from true amplifications and deletions.

• Mechanism: Single nucleotide polymorphism-aware analyses track the ratio of reference to alternate allele reads; a deviation from the expected heterozygous 0.5 ratio indicates allelic imbalance.
• Complementary Signal: Total expression increases can result from either amplification or transcriptional upregulation, but allelic imbalance provides orthogonal evidence of genomic alteration.
• Example: Numbat jointly models haplotype-aware expression and allele counts using a phylogenetic tree structure to reconstruct clonal CNV evolution in single cells.
• Requirement: Requires sufficient SNP coverage in transcriptomic reads, limiting applicability in low-coverage or 3'-biased single-cell RNA-seq protocols.

A modern approach that employs autoencoder architectures to learn a latent representation of expression data from which technical noise is removed before CNV inference.

• Mechanism: Variational autoencoders or denoising autoencoders compress high-dimensional gene expression into a low-dimensional bottleneck and reconstruct a denoised output, preserving biological variation while suppressing dropout and amplification artifacts.
• Advantage: Learns noise structure directly from the data without requiring explicit parametric assumptions about error distributions.
• Example: DeepCNV trains a convolutional autoencoder on single-cell expression matrices, then applies circular binary segmentation to the denoised output for breakpoint detection.
• Integration: Often combined with downstream HMM or segmentation algorithms rather than replacing them entirely, serving as a preprocessing enhancement layer.

A critical preprocessing step that adjusts raw gene expression counts to account for differences in sequencing depth and capture efficiency between cells. Without proper normalization, systematic technical artifacts can mimic or mask genuine copy number signals. Methods like library-size normalization or SCTransform regularized negative binomial regression stabilize variance across cells, ensuring that the averaged gene expression used in CNV inference reflects true chromosomal dosage rather than technical noise.

A feature selection method that identifies genes with the highest cell-to-cell expression variance, often excluding those driven by technical noise. In CNV inference, this step is crucial: using only the most informative genes sharpens the chromosomal signal by reducing background noise. Genes with low variance contribute little to detecting amplifications or deletions, so filtering to the top 1,000–5,000 highly variable genes improves the signal-to-noise ratio of the moving average across genomic coordinates.

The computational alignment of multiple single-cell datasets from different conditions, technologies, or donors into a shared latent space. When inferring CNVs across patient cohorts, batch effects can confound chromosomal signals. Methods like Harmony or scVI correct for these technical artifacts while preserving genuine biological variation. Effective integration ensures that inferred copy number alterations are driven by true genomic aberrations rather than inter-sample technical discrepancies.

The computational reconstruction of transcription factor–target gene interactions from single-cell expression data. CNVs often encompass transcription factors and their downstream targets, causing coordinated expression changes across entire regulons. By integrating CNV inference with regulon analysis, researchers can distinguish whether expression changes are driven by cis-acting dosage effects (direct CNV of the gene) or trans-acting regulatory perturbations (CNV of an upstream regulator), providing mechanistic insight into oncogenic programs.

The process of assigning known biological identities to single-cell clusters by comparing their gene expression signatures to reference databases or curated marker gene panels. CNV inference is highly sensitive to cell type composition: different lineages exhibit distinct baseline expression profiles that can be mistaken for copy number alterations. Accurate annotation allows analysts to perform within-lineage CNV comparisons, ensuring that inferred amplifications in tumor cells are measured against appropriate normal epithelial controls rather than stromal or immune cells.

A computational method that predicts the future transcriptional state of individual cells by modeling the ratio of unspliced precursor mRNA to mature spliced mRNA. When combined with CNV inference, RNA velocity can reveal whether copy number alterations are associated with specific differentiation trajectories or transitional cell states. This coupling helps identify the point in a developmental or oncogenic process where genomic instability arises, linking chromosomal aberrations to dynamic transcriptional programs.