Can You Combine Short-Read and Long-Read Data in the Same Study?

One of the biggest reasons labs hesitate to adopt long-read sequencing has little to do with read length. It is the data they already have. Years of short-read resequencing, array genotypes, and GBS calls represent real investment, published results, and trained models. Starting over is not an option most programs can take.

They do not have to. Short-read and long-read data can be combined in the same study, and in many cases long reads make an existing dataset more valuable. The key is choosing the right integration strategy for your question and avoiding a handful of technical traps that produce misleading results.

Three Ways to Combine Short-Read and Long-Read Data

Strategy 1: Joint Genotyping on Shared Sites

The simplest approach is to call variants from both datasets and merge them at the sites both technologies can genotype reliably. This preserves continuity with earlier analyses and works well for SNP-based population structure, diversity summaries, and relatedness estimates.

The trade-off is that the merged dataset is limited to what the weaker dataset can see. Regions accessible only to long reads, and variant classes such as large structural variants, drop out of the shared set. Use this strategy when continuity matters more than new discovery.

Strategy 2: Long Reads as a Reference Panel for Legacy Data

Sequence a representative subset of the population with long reads, build a phased reference panel that includes structural variants, and impute into the existing short-read, array, or GBS data. The legacy samples gain variants they could never have detected directly, including common structural variants in strong linkage disequilibrium with the surrounding markers.

This is the most powerful way to upgrade historical data, because each long-read sample added to the panel can improve imputation for every legacy sample. Validate carefully: imputed structural variants should be checked against directly genotyped samples before they drive decisions.

Strategy 3: Long Reads for Discovery, Legacy Samples for Scale

Use long reads to find and characterize variants in a focused set of lines, then screen the broader collection with a targeted assay. A structural variant resolved to base-pair breakpoints with long reads can be converted into a simple PCR or KASP marker and tested across archived DNA, germplasm collections, or historical breeding lines. Long reads provide the precision, and the legacy collection provides the scale.

StrategyWhat you gainTrade-off
Joint genotyping on shared sites
Priority: continuity
One merged SNP set for population structure, diversity summaries, and relatedness, consistent with earlier analyses Limited to sites both technologies can call. Structural variants and regions only long reads can reach drop out.
Long-reads as a reference panel
Priority: upgrading legacy data
Structural variants and phased haplotypes imputed into existing short-read, array, and GBS samples Imputed structural variants must be checked against directly genotyped samples before they drive decisions.
Long-reads for discovery, legacy samples for scale
Priority: discovery at scale
Variants resolved to base-pair breakpoints, converted to PCR or KASP markers, and screened across archived collections The screen only tests variants already found in the discovery set.

The Technical Traps

Mismatched Reference Assemblies

Legacy data is often aligned to an older reference assembly. Converting coordinates between assembly versions, a process called liftover, loses variants in rearranged or newly assembled regions. Where raw reads are available, realign everything to the same assembly rather than lifting over variant calls.

Inconsistent Variant Representation

Different variant callers can describe the same insertion or deletion in different ways, especially in repetitive sequence. Normalize and left-align variants across all datasets before merging, or identical variants will appear as mismatches.

Batch Effects Between Technologies

Short-read and long-read data differ in error profiles and in which regions they can call. If technology tracks with a biological variable, for example if all older lines were sequenced with short reads and all newer lines with long reads, platform differences can masquerade as biological differences. Include technology as a covariate, and make sure it is not confounded with the comparison you care about.

Treating Missing as Absent

When short reads fail to detect a variant, the variant is not necessarily absent. It may sit in a region short reads cannot align to. Merging a long-read call set with a short-read call set and treating every short-read gap as a reference allele creates false differences between groups. Use callable-region masks for each technology.

Build a Bridge Before You Cross It

The most useful single step in any integration project is resequencing a subset of legacy samples with long reads. These bridge samples show exactly how the two technologies agree and disagree in your species: concordance at shared sites, which regions short reads missed, and how accurately structural variants impute into the legacy data. A few dozen well-chosen bridge samples turn assumptions into measurements.

Keep your raw data, too. FASTQ and aligned read files from earlier studies can be realigned to improved references and reprocessed with better tools. Variant call files alone lock a dataset into the decisions made when it was generated. Long-read low-pass (LRLP) sequencing keeps a bridge set affordable, at roughly half the cost of standard long-read sequencing, so it can be large enough to represent the haplotypes in your population.

Step 1

Start with your legacy collection

Short-read, array, or GBS data, plus the raw FASTQ or aligned reads wherever you kept them.

Step 2

Resequence a bridge subset with long reads

A few dozen well-chosen samples, sequenced with both technologies.

Step 3

Measure, don't assume

  • Concordance at shared sites
  • Regions short reads missed
  • How accurately structural variants impute

Your Existing Data Becomes More Valuable

Moving to long-read genotyping does not require abandoning the data that built your program. Integrated thoughtfully, legacy datasets become the scale layer of a study, and long reads become the resolution layer. As a long-read reference panel grows, it raises the value of every historical sample imputed against it.

That turns genotype data into a cumulative asset. Programs that plan the integration deliberately keep every dataset they have paid for working, while gaining access to the variants their earlier methods could not see.

The Bottom Line

  • Short-read and long-read data can be combined in the same study. The right strategy depends on whether your priority is continuity, discovery, or scale.

  • Joint genotyping on shared sites preserves continuity but limits the dataset to what short reads can see.

  • Using long-read low-pass (LRLP) data as a reference panel brings structural variants and phased haplotypes into legacy short-read, array, and GBS datasets through imputation.

  • Realign to a common reference, normalize variant representation, control for technology batch effects, and never treat a short-read gap as a confirmed absence.

  • Resequence a set of bridge samples with long reads to measure concordance in your species before relying on combined results.


Veil Genomics offers the first commercially available long-read low-pass sequencing service on PacBio HiFi. If you are planning a transition from short-read, array, or GBS data, we can help you choose bridge samples and an integration strategy.

See example per-sample pricing by genome size in our 2026 Pricing Guide. Talk to a scientist or request a quote.

Related Reading

Next
Next

Can You Use Low-Coverage Sequencing for Genomic Selection?