Excellence in Breeding: developing breeding programs in Africa and South Asia
Posted 29th July 2019 by Jane Williams
Kelly Robbins is Assistant Professor at Cornell University. He started his career in animal breeding and later went on to work as a quantitative geneticist at a large private seed company. We spoke to him ahead of his presentation at the 7th Plant Genomics and Gene Editing Congress: USA.
I came to Cornell University as director of a large project funded by the Bill and Melinda Gates Foundation called the Genomic Open-Source Breeding Informatics Initiative (GOBII for short).
With the cost of sequencing rapidly declining, it seems inevitable that routine of genomic information in breeding programs in Africa and South Asia will become more prevalent. However, there are certain capabilities, tools, and technologies that need to be in place to be able to exploit this type of information effectively and use it to drive higher rates of genetic gains.
I’m also involved in another project called Excellence in Breeding, also funded by the Bill and Melinda Gates Foundation. It has a similar theme to GOBII but on a broader scale – Excellence in Breeding is really an effort to work with breeding programs that are serving Africa and South Asia to help those programs improve their effectiveness in delivering varieties that farmers adopt.
The other area that I’ve been working on is in remote sensing and how it can be incorporated into a breeding program to make decisions that weren’t possible before. My focus is on being able to look at how plants/varieties develop throughout the growing season and how that impacts end-use phenotypes like yield and contributes to genotype by environmental interactions, with the goal being able to select on parameters of development curves.
What are the specific challenges and issues with the plant breeding programs in Africa and South Asia?
There’s a lot of the same challenges that you have in big breeding programs, just with added levels of difficulty. In particular, getting the information need to establish variety product profiles and define the right breeding objectives is very challenging. Running a variety testing trial anywhere in the world is a challenging and expensive task. Traditionally in plant breeding, you can take advantage of the ability to cheaply replicate varieties and design experiments with high power to select the best varieties. As a result, breeders often think of these trials as having a single purpose, to identify and advance the best variety. In other words, I’m going to plan this out, I’m going to figure out which variety is the best, and then I’m going to move on. That data has served its purpose and therefore, it’s no longer useful.
I come from an animal breeding background where you can’t do that. In animal breeding you can sometimes go back decades and pull out data to use in genetic evaluations. It’s a very different philosophy – using data as a tool for decision making and then discarding it rather than data as a resource that can be leveraged to make better decisions moving forward.
When you start talking about implementing some of these new technologies to improve genetic gains with genomic information, genomic selection, remote sensing, etc, you must start thinking about data in a different way.
It’s not just something that you collect, make a single decision about and then throw away. It becomes a resource that you build up over time. As you build it up, it becomes more and more powerful and you can do more with it. Industry have thought about it as an investment for some time and treat data in that regard, whereas public sector programs still don’t really think about it in that way.
One of the fundamental challenges is putting in place the capabilities and mindset of keeping this data. The concept of fair data is big now in science generally, but it also applies to breeding. The data needs to be findable, accessible, interpretable, and reusable because we need to use that to train the predictive models that we’ll need to move these programs forward.
There is a huge amount of effort and resources that go into putting into place these foundational capabilities to be able to store and keep and curate data and make it available and accessible to make the right decisions.
The other area that we work on in these programs, is to really think through how you would rethink or redesign your breeding programs to take full advantage of having access to this information, being able to use genomic information to predict future performance, the ability to use technologies like proximal sensing to get phenotypes on traits that you didn’t used to be able to collect, at least in the early stage trials.
You have to build to a routine implementation. A lot of times in the public sector, you do a little proof of concept. You get a paper. Everyone feels good about it, but no one really goes to that next step, which is usually fairly difficult and time-consuming – making it a routine part of your breeding programs. We’re really trying to break that cycle of doing some proof of concepts and never really getting into impactful routine implementation.
We work with programs to help the establish strategies for developing proof of concepts that can move towards large-scale implementation. Sometimes they just need some help with project management; other times, they really need resources to figure out how to develop training sets, how to run the predictions they need, custom software developed for them to be able to implement that within the field. We help across that spectrum with these programs.
Are there any examples of implementation where these programs have rethought the way that they’ve done things?
The one that I’ve been involved in most closely is the East African CIMMYT Maize Program. They have leadership and innovative breeders that understand what it takes to drive new methods through to implementation. They’ve thought a lot about how to do it and how to step towards optimal implementation. The first cycle we went through with them was a success in terms of the data and the results.
Now we’re tweaking the implementation and improving it, and looking ahead to the next steps. In a very cost-effective way, we’ve been able to show that we can predict maize varieties using customized training sets that appear to be as accurate as the first stage trials. We have five locations of data at the current cost of genotyping which they can do for less than $10 a sample. It’s a huge saving. You can implement this genomic prediction without increasing your breeding budget. There is a great potential there to improve the rates of genetic gain.
What’s one thing you are most excited to see in the future?
A real change of culture in these breeding programs. In industry, we talk frequently of continuous improvement – no matter how good your program is, there are always areas that can be improved. You are constantly looking for and using tools to identify ways to improve the way your breeding program operates.
If we can get these breeding programs equipped to constantly tweak and improve their breeding processes and to have a mindset continuous improvement, then I would be overjoyed.
Kelly Robbins, Assistant Professor, Cornell University, will be speaking on ‘Technology driven crop improvement for Africa and South Asia’ at the Plant Genomics & Gene Editing Congress: USA.
Leave a Reply