AgriPlex Genomics Interview with NRGene’s Nir Kfir, PhD.

AgriPlex Genomics Director of Global Business Development, Giulia Bonciani, recently interviewed Nir Kfir, Project Director at NRGene, to discuss the six-year working relationship between the two companies. Their discussion highlights the transition from microarrays to sequencing, NRGene’s SNPer solution, and what makes their partnership with AgriPlex successful. See their full interview below.

Giulia: Thank you so much for taking the time to speak with us at AgriPlex Genomics. We wanted to ask you some questions to help the

community learn more about how our companies work together.

Tell us about your background. How long have you been in the biotech industry? How many of these years have been at NRGene?

Nir: I have a PhD in Molecular Genetics from Tel Aviv

University and have been in the biotech industry since

2015, about 11 years. I’ve spent nearly 7 years at NRGene, making the transition from cancer

therapeutics to agricultural genomics. What I didn’t

expect was how much there still is to discover. Seven years in, I’m still learning, and honestly, that’s one of the things I love most about being here.

At AgriPlex we are proud to have been collaborating with NRGene for six

years. What prompted NRGene to start working with AgriPlex back then?

Back then, NRGene was best known for its core strengths in genome assemblies and genomics breeding platforms for large seed companies. Around the time I joined, we began expanding into genotyping - moving from “big reference genomes” to practical, scalable tools breeders could use routinely. We developed skim sequencing (now often called low-pass sequencing) and imputation-based analysis pipelines, and we started building custom genotyping panels across multiple crops to capture broad genetic variation at the right cost and throughput. To make those solutions real-world ready, we needed a genotyping partner with strongtargeted sequencing capabilities, consistent data quality, and a team that could

collaborate closely on development and optimization. We evaluated several technologies and providers, and targeted sequencing stood out

as the best fit for the performance and flexibility we were aiming for. When we were introduced to Dr. Beni Kaufman and learned about AgriPlex’s services and approach, it clicked immediately. The technology aligned with our roadmap, and the partnership felt collaborative from day one - which is why it has grown into a six-year relationship.

Can you tell us a bit about SNPer from NRGene? How is NRGene able to

take a high density SNP list from a microarray, trim it down to less than a

thousand SNPs and accurately impute back up from it?

SNPer is one of NRGene’s most widely used genotyping solutions. The idea is simple: instead of running a high-density microarray on every sample, we design a minimal marker set - typically 500 to 1,000 SNPs - and then accurately impute back up to a much higher density using a reference dataset. Depending on the customer’s needs, we can impute to anything from a few thousand SNPs up to hundreds of thousands or more, including public arrays or custom high-density sets designed around specific germplasm. The key is how we design that minimal panel. We start by analyzing existing genotyping data (either the customer’s legacy data or public datasets). If that isn’t available, we generate a reference by sequencing a representative germplasm diversity set at moderate coverage (typically 5x to 10x). With that high-density reference in hand, our Guided Locus Search (GLS) algorithm selects a compact set of “haplotype anchor” SNPs - markers chosen to capture the linkage disequilibrium structure, so each anchor efficiently represents the haplotype block around it. Operationally, imputation relies on two pieces of information: (1) minimal-panel genotypes for the breeding populations (the progeny samples), and (2) high-density genotypes for the parental lines as the reference. The parental lines high-density genotypes can come from whole-genome sequencing or existing array data, and once

they’re in our database, only new parents need high-density genotype profiling. That’s where the economics really change: most samples in a breeding program are progeny, so you get near high-density resolution at a fraction of the per-sample cost. Because we impute within the haplotype boundaries defined by the parents for each population, the accuracy is typically very high. In our internal benchmarking, we usually see imputation accuracy in the 92% to 98% range, depending on crop, population structure, and the reference data available.

How does AgriPlex fit into the picture when it comes to SNPer?

AgriPlex is the platform partner that brings SNPer panels to life in the lab.

NRGene designs the minimal imputation panel - typically 500 to 1,000 SNPs - and we select markers that are optimized for AgriPlex’s targeted sequencing (GBTS), meaning they amplify cleanly, are technically robust, and deliver consistent calls. Over the years, we also aligned our variant discovery and filtering approach to the characteristics of the GBTS workflow so the panel performs reliably across samples and runs. Once the panel is developed and validated, AgriPlex supports customers either by providing kits and genotyping/calling tools for teams that sequence in-house, or by running the genotyping as a service through their lab. They then deliver the minimal-

panel genotype data, and NRGene completes the picture by running SNPer imputation to generate the high-density genotypes used for downstream breeding decisions. So it’s a very complementary split: AgriPlex ensures the minimal panel is reliable and scalable in practice, and NRGene turns that low-cost data into high-density insight through imputation.

Many customers ask us at AgriPlex if we can work with “difficult” species

like wheat. Would you be willing to share about your experience working

with AgriPlex and tricky genomes?

One of the milestones that helped define NRGene early on was our contribution to assembling the wheat genome that was published in Science. That experience - and the work that followed in other highly complex crop genomes - built deep internal expertise in handling difficult genomic architecture and designing variant discovery workflows that hold up in the real world. When it comes to genotyping panels, that matters because “difficult” species usually fail in the details: inconsistent amplification, noisy calls, and markers that don’t transfer well across diverse germplasm. Our approach is to start from diversity - using legacy datasets when available, and when not, generating the right reference data - and then applying crop-specific filtering and selection so the final SNP set is both informative and technically robust.

AgriPlex has been a strong partner in making that practical. We’ve developed panels together for several challenging species, and hops is a great example. It took multiple rounds of optimization and validation to get the panel performing exactly as we wanted. What I appreciate about the collaboration is the problem-solving mindset - at one point, an AgriPlex team member even went out and bought hop leaves from a store so we could run extra validation without consuming the customer’s limited samples. That’s thekind of hands-on teamwork that makes “difficult genomes” doable.

https://www.science.org/doi/10.1126/science.aan0032

What would you say is the “ideal” SNP density for genomic selection and

why?

This is a tricky question, it depends on many factors. There isn’t a single “ideal” SNP density for genomic selection - it depends on the biology of the trait and the structure of the breeding program. The goal is to have enough markers to capture the genetic variation across the breeding germplasm, but not so many that you’re just adding noise relative to the amount of phenotype and training data you have. In practice, a range of about 5,000-25,000 SNPs is often a good working density for many programs, with the exact sweet spot shifting based on genome size, diversity, and how predictable the trait is. If you have a very large, high-quality training set with strong phenotypes, moving to higher densities can improve predictions, but it’s not automatically better in every case. That’s also why we like the SNPer approach: you can genotype cost-effectively with a minimal panel and then impute up to the density that fits your model - whether that’s 10K, 25K, or much higher - without paying high-density costs on every sample.

What are the main hurdles a lab faces before being able to confidently

implement imputation pipelines? How does NRGene help?

The main hurdle is designing the minimal imputation panel. Choosing the most

informative SNPs that can accurately predict the alleles of nearby SNPs - what we call haplotype anchors - is not trivial. Once you have good minimal-panel genotypes for the progeny and high-density genotypes for the parental lines, running the imputation itself is relatively straightforward. We do, however, include several controls so we can monitor and report imputation accuracy, which is critical for confidence. Another major hurdle is pedigree integrity. If a progeny sample is inconsistent with its parents’ genotypes and shows many “impossible alleles,” we can’t reliably impute it. The upside is that this process helps uncover mislabeled samples or pedigree errors. We often see about 3% to 5% of samples in a batch fall into this category. In some cases, we can identify a different parent within the batch that matches the progeny

genotypes and correct pedigree assignment of these samples. If we can’t resolve it, we flag the sample and don’t provide imputed high-density data.

For years the community has enjoyed speculating on when genotyping

technology will shift entirely from microarrays to sequencing. Where do

you see the future of genotyping technology?

I think this shift is already happening and it’s accelerating. We’re seeing well-known microarrays become unavailable as major providers discontinue legacy products, which pushes programs to look for sequencing-based alternatives.

At the same time, targeted sequencing approaches like amplicon-based genotyping and GBTS are becoming much more common, and hybrid-capture based genotyping is also expanding. Arrays are responding by getting cheaper - partly because array vendors can spread costs across very large multi-crop products, and because sequencing-based genotyping has created real price competition. In some sectors there’s also strong momentum to keep arrays, for example where workflows are deeply standardized and the ecosystem is built around them. But longer term, it’s hard to see arrays winning. Sequencing keeps improving on every front: instruments, chemistry, throughput, and library prep, and it gives you flexibility that fixed arrays can’t. So, I expect genotyping to continue moving toward sequencing as the default, with arrays gradually becoming more niche over time.

What’s your favorite thing about working with our team at AgriPlex?

There is a lot that I like and enjoy in our partnership with Agriplex, but I can say that my favorite thing about Agriplex is the Agriplex team. I get great service and support from the team. They are very responsive and feel they always go the extra mile to accommodate my requests and our customer’s; requests and needs. There are a lot of those…….

Any other thoughts you would like to share with our readers?

If I could leave readers with one practical takeaway, it would be this: if you’re

transitioning to targeted sequencing, start small. Very cost-effective panels of 50 to 150 SNPs can get the per-sample cost down to under $5 at scale, and they already deliver a lot of value. These small panels are ideal for routine QC of production materials, marker-assisted selection screening, and checking diversity or identifying off-types early. And once that foundation is in place, there’s a lot you can build on - at NRGene we also apply compact panels to additional downstream analyses that add real breeding value across both

breeding and production programs.