Summary

Hi-C sequencing captures the three-dimensional architecture of the genome by quantifying interaction frequencies between all pairs of genomic loci. The resolution of a Hi-C experiment — and therefore the biological features it can reliably detect — is determined primarily by sequencing depth. A/B compartments are detectable at 50–100 million read pairs, TAD boundaries require 100–200 million, and chromatin loops at kilobase resolution need 500 million or more. This guide maps sequencing depth to the features each depth tier can resolve, helping researchers plan a Hi-C project that matches their biological question without paying for unnecessary coverage. For detailed project guidance, explore the Hi-C service page for sample requirements and submission guidelines.

Figure 1. Hi-C contact maps at three resolution tiers — compartments (50–100 kb bins), TADs (25–50 kb bins), and chromatin loops (5–10 kb bins) — and the sequencing depth each tier requires in a mammalian genome.

Key Takeaways

• A/B compartments are detectable at 50–100 million read pairs; TAD boundaries at 100–200 million; chromatin loops at 500+ million
• Resolution in Hi-C is defined by bin size, and achievable bin size depends on sequencing depth
• Biological replicates (2–3 per condition) are more important than ultra-deep sequencing of a single sample for differential analysis
• Low-input and single-cell Hi-C have fundamentally different resolution limits than standard Hi-C
• Matching read depth to the feature of interest saves 50–80% of sequencing cost compared to a default "deep sequencing" approach

What Resolution Means for Your Question

Hi-C resolution is defined by the bin size used to build the contact matrix. A 100 kb bin means each cell in the matrix represents the summed interactions between two 100 kb genomic intervals. Smaller bins give finer spatial resolution but contain fewer reads per bin, making the matrix sparser and statistical detection of features less reliable.

The relationship between sequencing depth and achievable resolution follows a practical rule: more reads allow smaller bins while maintaining sufficient per-bin coverage for feature detection. For a mammalian genome (~3 Gb), the bin size that can be reliably analyzed depends on the total number of mapped read pairs.

The following table maps sequencing depth to the features resolvable at each tier:

Fewer reads are needed for smaller genomes. A Drosophila Hi-C experiment at 100 million read pairs achieves roughly 5–10 kb resolution, comparable to 500 million reads in human. The genome size difference is the main scaling factor.

How Many Reads for Compartments

A/B compartments are the largest and most robust feature in Hi-C data. They represent the segregation of active (A compartment, gene-rich, transcriptionally active) and inactive (B compartment, gene-poor, repressed) chromatin at megabase scale. Compartments are detected by principal component analysis of the contact matrix at 100–200 kb resolution.

50–100 million read pairs is sufficient for compartment detection in a mammalian genome. At this depth, the first principal component (PC1) clearly separates A and B compartments, and the compartment profile correlates well with replication timing, histone marks, and gene expression data from matched samples.

Compartment detection is robust enough that even datasets with 25–50 million read pairs can reveal the major compartment switch on the X chromosome or between differentiation states. The trade-off is that finer subcompartments (A1, A2, B1–B3) require higher depth or dedicated algorithms.

If compartments are the primary question — for example, comparing the overall chromatin state landscape between two cell types — depth above 100 million read pairs produces diminishing returns. The compartment profile plateaus, and additional reads contribute more to TAD and loop detection than to compartment resolution.

Read Depth That Resolves TAD Boundaries

Topologically associating domains (TADs) are sub-megabase regions where sequences interact more frequently with each other than with sequences outside the domain. TAD boundaries are enriched for CTCF binding sites and housekeeping genes, and they are largely conserved across cell types.

100–200 million read pairs is the recommended range for TAD detection in mammalian Hi-C experiments. At 25–50 kb bin size, TAD boundaries identified at this depth match well between replicates and across published datasets. The insulation score — a common TAD boundary metric — stabilizes at approximately 150 million read pairs.

Figure 2. TAD boundary detection sensitivity at different sequencing depths — the insulation score profile at 50 M, 100 M, 200 M, and 500 M read pairs showing that weak boundaries become detectable only at higher depth.

Below 100 million reads, TAD boundary detection becomes less reliable. Weak boundaries are missed, and the apparent TAD size distribution shifts toward larger domains as smaller TAD structures become indistinguishable from noise. This effect is particularly problematic for detecting cell-type-specific TAD changes, where the biological signal of interest may be a boundary weakening or shifting rather than a complete loss.

For studies comparing TAD boundaries between conditions — such as upon CTCF depletion, cohesin knockout, or during differentiation — 150–200 million read pairs per sample provides enough sensitivity to detect boundary shifts. Higher depth is warranted only when sub-TAD structures or intra-TAD contact changes are the targets.

What Loop Detection Actually Requires

Chromatin loops are point-to-point interactions between specific loci, typically mediated by CTCF and cohesin through loop extrusion. Identifying loops requires calling significant interaction peaks against a background model, which demands higher read depth than either compartment or TAD detection.

500 million to 1 billion read pairs is the typical range for comprehensive loop detection at 5–10 kb resolution. The landmark Rao et al. 2014 study achieved ~10,000 loops in GM12878 cells at 4.9 billion contacts and 1 kb resolution, but most loop-focused studies operate at lower depth with adequate results.

The number of loops detected scales with sequencing depth but plateaus beyond approximately 1 billion read pairs for most human cell types. At 500 million reads, typical detection ranges from 5,000 to 10,000 loops. At 200 million, loop detection drops to 1,000–3,000 loops, and only the strongest interactions are recovered.

For projects where the goal is to identify enhancer-promoter loops linked to differentially expressed genes, 200–500 million read pairs is a practical target that balances detection sensitivity with sequencing cost. The specific loops of interest — those anchored at genes that change expression — tend to be among the stronger interactions and require less depth than comprehensive whole-genome loop cataloging. For loop-focused projects, integrated calling and analysis pipelines are available through the Epigenomic Data Analysis service.

Replicates That Match Your Design

Biological replicates are essential in Hi-C experiments because the contact matrix captures both biological variation and technical noise, and the two cannot be distinguished from a single sample.

Minimum recommendations for replicates:

Combining data from replicates by summing contact matrices is a common practice that effectively increases sequencing depth. Two replicates at 100 million reads each, when combined, produce a 200 million read dataset for feature detection while also providing replicate-level statistics for differential analysis.

For differential interaction analysis — identifying contacts that change significantly between conditions — biological replicates are non-negotiable. Without replication, there is no basis to distinguish biological differences from technical variation. Most differential Hi-C methods (diffHiC, multiHiCcompare, HiCcompare) require at least two replicates per condition.

Controls and Technical Checks Worth Building In

Hi-C experiments have well-characterized failure modes that can waste sequencing budget if not caught early. Including the right controls and QC checkpoints prevents investing deep sequencing in failed libraries.

Key technical controls:

• Digestion efficiency check. Run the purified crosslinked DNA on a gel before and after restriction digestion. Complete digestion produces a smear shifted toward smaller fragments. Partial digestion generates a visible high-molecular-weight band — a warning that the library will have a low valid pair rate.
• Ligation control. The biotinylated ligation junction is the signature of a valid Hi-C contact. A qPCR assay targeting ligation junctions can estimate the library quality before sequencing. Libraries with fewer than 10–20% valid pairs after sequencing typically trace back to poor ligation efficiency.
• Sequencing QC metrics. After the first sequencing run (10–20 million reads), check the cis/trans ratio, the fraction of reads within 20 kb versus 200 kb, and the long-range contact enrichment. A cis/trans ratio below 1.0 for intra-chromosomal contacts suggests excessive noise from random ligation.
• ENCODE compatibility check. Compare insulation scores, compartment strength, and replicate reproducibility against published ENCODE Hi-C standards for the same cell type if available. The ENCODE Hi-C pipeline provides reference quality metrics.

These checkpoints are best inserted between library preparation and full-scale sequencing. A 10-million-read QC sequencing run costs much less than discovering at 500 million reads that the library has a 2% valid pair rate. CD Genomics incorporates these QC checkpoints in its standard Hi-C service workflow and provides pre-sequencing library QC reports.

From Resolution Plan to Project Plan

A Hi-C project plan should start from the primary biological question and work backward to the sequencing depth and replicate structure.

The decision sequence is:

1. Identify the target feature. Compartments, TADs, or loops? This sets the depth floor.
2. Determine the comparison type. Single-condition characterization, pairwise comparison, or multi-condition study? This determines the replicate structure.
3. Check sample constraints. Low cell numbers or precious samples may require low-input Hi-C, which has inherently lower resolution than standard Hi-C from high-quality inputs.
4. Budget for QC. Reserve 3–5% of the sequencing budget for pre-library QC (digestion check, ligation qPCR) and a shallow QC sequencing run.
5. Plan the bioinformatics pipeline. Tools for each feature type differ. Compartment analysis uses PCA (cooltools, Juicer). TAD calling uses insulation scores or directionality index (HiCExplorer, TopDom). Loop calling uses peak callers (Mustache, Peakachu, SIP).

The most common Hi-C planning mistake is sequencing too deeply for a compartment-level question, or not deeply enough for loop detection. A clear resolution plan saves 50–80% of sequencing cost compared to a default "deep sequencing" approach for projects with well-defined feature targets.

Figure 3. Hi-C project planning decision flowchart — from biological question to feature target, sequencing depth, replicate structure, and QC checkpoints.

Frequently Asked Questions

1) Can I detect TADs with 50 million read pairs per sample?

Partial TAD detection is possible at 50 million reads, but only strong, well-defined TAD boundaries are reliably recovered. At this depth, approximately 30–50% of TAD boundaries identified at 150 million reads will be detectable. For reliable genome-wide TAD analysis, 100–200 million read pairs per sample is recommended. If 50 million reads is the budget limit, consider focusing on compartment-level analysis or combining replicates computationally.

2) How many reads do I need for single-cell Hi-C?

Single-cell Hi-C typically generates 50,000–500,000 contacts per cell, orders of magnitude fewer than bulk Hi-C per cell. Resolution in single-cell Hi-C is described at the population level: pooling 50–100 single cells produces a pseudo-bulk contact matrix suitable for compartment detection. TAD-like domains can be called in individual high-coverage cells, but loop detection at single-cell resolution remains a research challenge rather than a standard deliverable.

3) What is the difference between Hi-C and Micro-C resolution planning?

Micro-C uses micrococcal nuclease digestion instead of restriction enzymes, producing shorter fragments and better short-range contact coverage. This makes Micro-C more sensitive for detecting loops at sub-kb resolution but does not change the depth requirements proportionally: Micro-C still needs 500+ million read pairs for comprehensive loop detection. The main difference is that Micro-C captures finer-scale contacts that Hi-C may miss due to restriction fragment size limitations, not that it requires less sequencing.

4) Should I sequence deeper or add more replicates?

If the goal is to detect features in a single condition, deeper sequencing of one or two samples is more effective. If the goal is to find differences between conditions, adding biological replicates provides more power than sequencing the same samples deeper. As a rule of thumb: for detection, go deeper; for comparison, add replicates. Three replicates at 200 million reads each is more informative for differential analysis than one sample at 600 million reads.

5) What is a cis/trans ratio and what should it be?

The cis/trans ratio is the fraction of intra-chromosomal read pairs (cis) versus inter-chromosomal read pairs (trans). In mammalian Hi-C, cis reads should substantially outnumber trans reads. A cis/trans ratio of 2.0–5.0 is typical for high-quality libraries. Ratios below 1.0 indicate excessive random ligation or poor crosslinking efficiency, and libraries with low cis/trans ratios produce noisy contact maps with reduced feature detection sensitivity.

Related CD Genomics Services

• Hi-C Service — Genome-wide 3D chromatin conformation capture
• Capture Hi-C — Targeted high-resolution contact mapping
• HiChIP Sequencing — Protein-anchored chromatin interaction profiling
• Epigenomic Data Analysis — Integrated 3D genome and epigenome bioinformatics

References

1. Lieberman-Aiden E, van Berkum NL, Williams L, et al. Comprehensive mapping of long-range interactions reveals folding principles of the human genome. Science. 2009;326(5950):289-293. doi:10.1126/science.1181369
2. Rao SSP, Huntley MH, Durand NC, et al. A 3D map of the human genome at kilobase resolution reveals principles of chromatin looping. Cell. 2014;159(7):1665-1680. doi:10.1016/j.cell.2014.11.021
3. Bonev B, Cavalli G. Organization and function of the 3D genome. Nature Reviews Genetics. 2016;17(11):661-678. doi:10.1038/nrg.2016.112
4. CD Genomics. Hi-C / Genome-Wide 3C Sequencing Service. Accessed June 2026.

Services mentioned in this article are provided for research use only and are not intended for clinical diagnosis, treatment, or personal health assessment.