GenomeOne Brings Genomic Prediction Into Breeding Decisions
Reading time: 10 min
Sow 100 seeds from a cross and a program might advance a handful of them. The rest are judged on disease resistance, appearance, taste and whatever else the eye can reach, then marked as failures. Some of them carried the trait the program spent the next several seasons hunting for.
Genomic data changes which plants survive that decision, and it changes it early, before the traits have shown themselves in the field.
GenomeOne is a new Phenome Networks product for the genomic side of a breeding program, and it works against the phenotypic records the program has already been keeping for years.
In this article
A line that looks wrong can carry the trait a program needs, and reading genetics early significantly increases how many lines stay relevant.
The output is a prediction that supports a decision, grounded in the relationship between genetics and traits you measured yourself.
Phenotypic data never leaves PhenomeOne. GenomeOne reads it in place and matches genetic samples to it automatically.
Three starting points, from a small program using an outside supplier to a breeder making advancement calls, and what changes in each.
Genomics delivered as an ongoing subscription rather than a funded project, which is what puts it within reach of smaller programs.
Every company’s data stays its own and trains only its own model, logically separated from every other customer, with a dedicated private cloud available as an upgrade.
Table of Contents
The Lines You Throw Away Too Early
Why a Round Tomato Takes So Many Seasons
What the Prediction Actually Hands You
Two Systems That Share One Set of Records
Where Genotype Data Sits and Waits
Three Ways Programs Arrive Here
The Lines You Throw Away Too Early
GenomeOne Brings Genomic Prediction Into Breeding Decisions
Selection in an early generation is mostly a visual exercise. Someone walks the plot and reads fruit shape, color, vigour, disease symptoms, taste where it can be assessed, and marks what continues. It has worked for a century and it still works.
It also discards information that was never visible in the first place.
A plant or a fruit can look entirely unsuitable for continued breeding and still carry, somewhere in its sequence, a gene or a genomic region tied to a trait the program badly wants. Disease resistance is the obvious case. Cull that line on appearance and the resistance goes back into the ground with it.
Keep it instead, and it becomes a parent. Cross it with a line that already shows the external characteristics you want and you are working toward offspring that combine both, rather than choosing between them. That is a different kind of decision from the one appearance alone allows, and it can be made at an earlier stage of the program.
The effect on the shape of a program is not subtle. Out of those 100 seeds, the traditional method advances only the small number that pass on looks. Reading the genetics identifies potential in lines that would have been discarded earlier, which significantly increases the number of lines still relevant to the work. You end up with more to choose from, not less.
The real argument for genotyping early is not speed for its own sake. It is that decisions can rest on genetic information from the first stages, instead of waiting several seasons for traits to express themselves in the field.
Why a Round Tomato Takes So Many Seasons
Say the goal is a round-shaped tomato. In each generation the team grows a large number of lines, examines the phenotypic results, and selects the most suitable ones to carry into the following season. Then it does it again. Reaching the intended variety often takes many generations and many growing seasons.
Most of what gets grown along the way turns out to be irrelevant to the result.
That is not a small accounting problem. Every one of those lines consumed time, growing area, and the working hours of the people collecting phenotypic data in the field, plus the money underneath all of it. The cost is real whether or not the line ever contributed anything.
Bringing the genomic dimension into the same process gives you information about the plants’ DNA before all the traits have actually been expressed. Identifying markers or DNA regions associated with the trait you want, running genomic prediction, and classifying lines into populations all point at the same practical outcome, which is choosing in a more targeted way which lines are worth advancing. Fewer wasted cycles, and the desired result arrives sooner.
What the Prediction Actually Hands You
Ask what comes out at the end of a genomic prediction run and the honest answer is not a report, and not a score anyone should act on blindly. What you receive is a prediction that lets you decide how to continue the breeding process, built from genomic and phenotypic information together.
It rests on the relationship between the plant’s genetic information and the traits that were genuinely measured. Not measured in general, measured by you, in your trials. Where a marker or a genomic region associates with a trait the program cares about, that association becomes usable on plants nobody has grown out yet. This is roughly the general logic behind genomic selection as the literature describes it.
The range of traits is wide, and which ones are workable depends on the trait itself and on how much genomic and phenotypic information exists for it. Programs work with taste, color, disease resistance, sugar content, resistance to weather conditions, water consumption, and the ability to increase the amount of produce obtained from the same growing area. Anything else that matters to a particular breeding program belongs on that list too.
Accumulation matters more than the choice of algorithm. The more genomic and phenotypic information a program collects, the larger the body of data the prediction rests on, which is also why refreshing training data across successive cycles is treated as routine rather than as a one time exercise.
Two Systems That Share One Set of Records
Two Systems That Share One Set of Records
PhenomeOne manages the phenotypic information a breeding program collects. Traits of different plants and lines, color, shape, disease resistance, field trial results, images, data gathered across seasons, historical information, and the relationships between generations and breeding lines.
Breeding companies accumulate enormous quantities of this over the years. At some of them it runs to decades, covering thousands of different crops, lines and trials. The point of holding it in one system is that it stays accessible, rather than living in separate files and spreadsheets that only one person fully understands.
GenomeOne focuses on the genomic information. Upload it, process it, run analyses on it, and combine it with the phenotypic information already sitting in PhenomeOne. Putting the two together is what makes genomic prediction possible and what supports the decision at the end of it.
The connection between the two deserves precision, because it is easy to describe wrongly. Information does not pass from PhenomeOne to GenomeOne in the sense of moving a database from one system into another. GenomeOne makes use of the phenotypic information where it already is, alongside the genomic information, and it links genetic samples automatically to the phenotypic records that match them. Connecting the phenotypic data is a button, not a project.
So nobody exports phenotypic files, re-uploads them somewhere else, and hand matches genetic information against phenotypic information. If you have ever renamed spreadsheet columns to line up a sample sheet with a plot list, you already know exactly what that removes from the week.
Results travel back the other way as well. Analyses can be displayed inside PhenomeOne, which matters because the breeder who has to make the decision works there and not inside an analysis tool. Different genetic studies belonging to the same genetic map and the same crop can be linked and shown on a single screen, so the comparison happens in one place. Instead of knowing only how a plant looked or behaved in the field, you can look at the genetics behind it and use both when deciding which lines to advance and which crosses to make, exactly as the breeding workflow in PhenomeOne already handles the phenotypic half.
See what GenomeOne would do with your data
Walk through genomic prediction on your own crops and records with our team, and see where it changes a decision.
Where Genotype Data Sits and Waits
Getting genomic data is no longer the hard part of genomics. Using it is.
Raw genomic information arrives, and then it has to be processed. Analyses have to be run on it with external tools. It has to be combined with the phenotypic information, which means exporting that information first and then matching it to the genetic samples by hand. Every one of those steps needs professional knowledge and tools such as Python or R, plus somebody who is genuinely fluent in both.
That is a real constraint on a team already stretched across several crops. It is also the same failure mode that quietly loses phenotypic records between disconnected systems, and it is why the raw file so often sits untouched while the season closes around it.
Three Ways Programs Arrive Here
Programs reach genomic prediction from three fairly distinct starting points, and what changes is different in each.
A small company using an outside genomics supplier
Many smaller companies hand the processing and the analyses to a third party, because they do not have the time, the knowledge or the tooling to do it in-house. The work comes back as a one-off project, usually an expensive one, and the moment a follow up question appears it is another engagement. Inside GenomeOne the same company uploads its raw genomic information, processes it, chooses filters and parameters, connects the phenotypic information from PhenomeOne at the click of a button, picks the analysis it wants and runs it without writing code. The matching between phenotypic information and genetic samples happens automatically. Genomics becomes something the company does continuously rather than something it commissions.
A mid-size or large company with a bioinformatics team
Here the capability exists in-house, and the cost is the specialists’ time. Data processing, filtering, writing code, combining genomic and phenotypic information, running analyses, producing results, all of it demanding real command of Python, R and bioinformatics. GenomeOne moves that work into the interface. The team uploads the genomic information, processes it, sets parameters, draws on the phenotypic information already in PhenomeOne and runs the analysis, without the code. Results can surface in PhenomeOne so the breeder sees them in their own working environment. The specialists get their time back for the questions that actually need them.
A breeder making the advancement calls
The breeder advances through generations and seasons, examining lines on phenotypic results and deciding what carries forward. The process is long, a large share of the lines prove irrelevant later, and time, growing area, personnel and resources went into all of them regardless. With genomic information alongside the phenotypic, the breeder can run genomic prediction and use markers, marker-trait associations and population classification to read the potential of different lines at an earlier stage. Decisions that once required a data analyst or a bioinformatician become decisions the breeder can make.
The Vocabulary That Travels With This Work
These terms are standard in the industry, and they turn up in every conversation about genomic work. Worth being exact about them, because two of them get used interchangeably when they should not be.
Which Crops This Fits
The practical test is simple. If genotyping is performed for the crop, the approach applies to it.
That covers vegetables including tomato and pepper, field crops including wheat and rice, and a further group taking in flowers, berries and other vegetatively propagated crops. The industry can now genotype a very wide range of species, considerably wider than most breeders assume when they first ask the question.
There is an honest limit to state alongside that. Some crops still have no genomic component in their breeding process at all, not because the technology cannot reach them but because it has not become part of how those programs work. Worth checking against your own crop rather than assuming either way.
What It Costs and What It Saves
Genomics has been sold as a project for a long time. A company that wants this kind of work engages an external genomics company, which puts a dedicated team on a one-off piece of work and delivers a result at the end. Projects of that shape can run into the hundreds of thousands of dollars. One example given was roughly 250 thousand dollars across about three years.
That model carries an obvious consequence. When genomics is a capital project, only organizations that can fund capital projects do genomics.
GenomeOne offers the same processes on an ongoing subscription basis, inside a system the company is already using day to day, priced annually according to the number of users. It is significantly cheaper than the existing alternatives and the barrier to entry is significantly lower. On the information available, it is currently the only offering that delivers processes of this kind under a subscription model rather than as a commissioned project.
The point of pricing it that way is access. Smaller companies that previously struggled to compete with large organizations holding substantial budgets and dedicated teams can work with tools that used to sit well out of reach.
The savings show up in more than one column. Bioinformatics teams get their working hours back from data processing, code writing and the business of combining information held in different systems. Time to results shortens. So does the number of breeding cycles, and with it the growing area, the personnel needed to collect phenotypic data in the field, and the financial resources underneath all of them. A program running genomics and marker data this way adds the capability without first hiring for it.
