Time-series Variant Calling — Longitudinal Somatic Mutation Detection
Time-series Variant Calling · Also known as: longitudinal variant calling, temporal somatic mutation detection, serial variant calling, time-course variant detection
Time-series variant calling is a bioinformatics pipeline that identifies and tracks genomic variants — typically somatic mutations — across multiple sequencing samples collected from the same subject at different time points. It is most widely applied in cancer genomics to reconstruct tumour evolution, monitor minimal residual disease, and detect the emergence of therapy-resistant clones. By jointly modelling variant allele frequencies across the temporal dimension, the method distinguishes true somatic changes from sequencing noise and estimates clonal dynamics over time.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use time-series variant calling when you have sequencing data from the same individual at two or more distinct time points and your question concerns how the variant landscape changes over time — most commonly in cancer (tumour evolution, therapy resistance, minimal residual disease monitoring) but also in infectious disease (viral or bacterial evolution within a host) and evolutionary biology experiments. The approach is essential when single-timepoint data cannot distinguish a new mutation from one present but undetected earlier. Do NOT use it as a replacement for standard somatic variant calling when only a single time point is available — the longitudinal joint model requires at least two time points to function. Avoid it when samples from different time points differ substantially in sequencing depth or platform, as this introduces systematic biases that are difficult to correct; in such cases, run independent per-timepoint analysis and interpret conservatively.
Strengths & limitations
- Detects low-frequency variants that fall below single-timepoint calling thresholds by borrowing statistical power across the temporal series.
- Distinguishes founder (clonal trunk) mutations from subclonal or newly acquired mutations, enabling tumour phylogeny reconstruction.
- Directly quantifies clonal dynamics — expansion, contraction, or extinction of cell populations — linking genomic change to clinical events.
- Sensitive for monitoring minimal residual disease or early relapse using liquid biopsy serial data.
- Jointly models tumour purity and copy-number changes at each time point, reducing false positives caused by ploidy shifts.
- Requires at least two matched, high-quality sequencing samples from the same subject; sample attrition or quality failure at any time point degrades the analysis.
- Computationally intensive: joint models over many time points and thousands of variants are memory- and CPU-demanding.
- Tumour heterogeneity and spatial sampling bias (biopsy only captures one region) can mislead clonal inference even with perfect temporal coverage.
- Liquid biopsy time-series at very low tumour fractions (<0.1%) requires ultra-deep sequencing (>10,000×) and error-correction protocols (e.g., UMIs) to avoid noise dominating the signal.
- Clinical annotation of detected variants depends on curated databases that may lag behind current literature.
Frequently asked
How many time points do I need for a meaningful analysis?
Two time points allow detection of newly acquired or lost variants, but phylogenetic reconstruction and clonal dynamics modelling are severely underpowered. Three to five time points are generally considered the practical minimum for reliable clonal inference. More time points improve resolution of the evolutionary trajectory, but the marginal gain diminishes beyond about eight to ten densely sampled points.
Can I use time-series variant calling with single-cell sequencing data?
Yes, but the workflow differs substantially. Bulk longitudinal calling aggregates VAFs across thousands of cells, while single-cell approaches resolve per-cell genotypes directly. Single-cell data at two or more time points enable trajectory analysis with tools like Cassiopeia or CARLIN, which assign mutations to individual lineages. The statistical models differ; bulk joint-VAF models should not be applied directly to sparse single-cell matrices.
Which tools are commonly used for the joint modelling step?
PyClone-VI is widely used for clonal clustering of VAFs across time points. Mutect2 (GATK) supports multi-sample joint calling in tumour-only or matched-normal modes. ClonEvol and Timescape handle phylogenetic visualisation of clonal dynamics. For liquid biopsy at very low tumour fractions, ichorCNA (copy number) combined with ddPCR or ctDNA-specific callers (e.g., ctSNV) are common. Tool choice depends heavily on sequencing depth and available matched normal data.
What sequencing depth is required?
Standard WGS at 30–50× is sufficient for detecting clones present at roughly 10% or more of cells. Targeted panel sequencing at 500–1000× extends detection to 1–2% VAF. Liquid biopsy monitoring of ctDNA at <1% tumour fraction requires 5,000–30,000× depth combined with unique molecular identifiers (UMIs) for error suppression. The required depth should be estimated based on the expected minimum clone size before sequencing is initiated.
How do I handle copy-number changes that shift VAF across time points?
Apparent VAF changes can reflect copy-number evolution rather than changes in the fraction of cells carrying a mutation. All time-series analyses should include copy-number profiling (e.g., FACETS, PURPLE, or CNVKit) and convert observed VAFs to cancer cell fractions (CCF) by correcting for local copy number and tumour purity at each time point. Failing this correction is the most common source of spurious clonal dynamics.
Sources
How to cite this page
ScholarGate. (2026, June 3). Time-series Variant Calling. ScholarGate. https://scholargate.app/en/bioinformatics/time-series-variant-calling
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- RNA-seq Differential ExpressionBioinformatics↔ compare