What is a podcast?
For those of you who are newer to the medium, a podcast is like a pre-recorded radio show. In the same way that you turn on a talk radio show, you have to turn on a podcast. The major difference is that while our cars are equipped to find radio frequencies, they are not built to accommodate direct access to podcasts. On your smartphone or computer with internet access (since the files tend to be on the larger side), you can discover podcast shows of any kind, in any field, on any topic.
Listed above are some of the most used podcast hosts. iTunes and the iTunes Podcast app are preinstalled on your iPhone and are the simplest tools to use. You simply search for “WSU Wheat Beat Podcast” in the search bar, hit “subscribe” and the download arrow, and listen whenever it’s convenient for you.
If you use an Android or use another type of smartphone, you will need to find a different podcasting app because those devices don’t come with a preinstalled app like Apple. If you don’t know which podcast app you’d like, simply hit the “Android” link above and it will show you to several Android podcast apps for you to choose from.
After you download an episode, you can listen without using data any time of day. Our goal is to post a new podcast every other Monday. Your podcast app should automatically load our new episodes and download them for you (on WiFi), hands-free if you choose that in the app settings.
If you have further questions about what a podcast is, which app is best for you or need more assistance with getting started with podcasts, don’t hesitate to contact us.
Drew Lyon: Hello. Welcome to the WSU Wheat Beat podcast. I’m your host, Drew Lyon, and I want to thank you for joining me as we explore the world of small grains production and research at Washington State University. In each episode, I speak with researchers from WSU and the USDA-ARS to provide you with insights into the latest research on wheat and barley production. If you enjoy the WSU Wheat Beat podcast do us a favor and subscribe on iTunes or your favorite podcasting app and leave us a review while you’re there so others can find the show too.
[ Music ]
Drew Lyon: My guest today is Karansher Sandhu. Karan is a Ph.D. candidate in the Department of Crop and Soil Sciences. Karan is originally from India. He received his BS degree from Punjab Agricultural University. Karan joined WSU in fall of 2017, under the supervision of Dr. Arron Carter, who’s leading the WSU Winter Wheat Breeding Program. Karan’s current research focus is on breeding for grain protein content stability using genomic selection and high throughput phenotyping tools. He’s also focusing on identifying wheat lines having quicker canopy closure and cold tolerance to compete with weeds in the Pacific Northwest. Furthermore, he is applying machine and deep learning models for predicting grain yield and use quality traits and various other agronomic traits in the wheat breeding program. Hello, Karan.
Karan Sandhu: Hi, Drew, thanks for inviting me.
Drew Lyon: Nice to have you on. We’ve had you on before talking about an aspect of your work, but we’re going to dive into more the non-breeding aspects of it today.
Karan Sandhu: So, yes, I came here last year. So that time we just talk about protein content stability by using traditional breeding approaches. So that was one of my research objectives, yeah.
Drew Lyon: Okay, so why don’t you tell us a little bit about what machine and deep learning are and how you are using them in the wheat breeding program.
Karan Sandhu: So, yeah, machine learning and deep learning are not exactly separate from each other. So they are a branch of artificial intelligence. So machine learning uses various model. We strive to infer or learn the trend present in the data set and then try to make decisions of the prediction depending upon those. While the deep learning system is more complex, and it is also a branch of machine learning. So deep learning works like the way the human brain works. So it uses different sets of neuron, the way neuron interact with each other and give some kind of stimulus. So that’s how deep learning model works. So it uses different neuron. And the interaction of neurons decide the models. So, for example, you can take the example of our iPhone. So Siri works on the basis of deep learning model because we have a lot of data sets. So whenever we have a lot of data, we can use deep learning models in order to get more prediction or better decision-making power. In our Winter Wheat Breeding Program, we are using both machine and deep learning model for predicting traits like grain yield, grain protein content, several agronomic, end-use quality traits. So exactly we are implementing this thing for most of the traits which we are working on.
Drew Lyon: Okay, so you’ll have to help me through this. Do you have an example of something you’ve done with deep learning? And how exactly it takes a data set?
Karan Sandhu: Okay, so for deep learning, we use our like five-year data set from 2015 to 2020 for learning the information in the data. And then we made prediction for the grain yield. And that publication recently got accepted in the “Frontiers in Plant Sciences.” And we got an accuracy of 60 percent for predicting the grain yield. That means we are moving toward better prediction accuracy by using deep learning models in our program.
Drew Lyon: Okay, so what are some traits or aspect for which you’re using machine learning in your breeding program?
Karan Sandhu: So exactly we are using that thing for all the traits we are working on, which varies from grain yield, grain protein content end-use quality and disease resistance. So that means we are implementing this thing on whatever we can do. Because in our breeding program we have a lot of data set from 2000 to 2020. That means as the data set is increasing, our models are performing more better. So we can implement this one.
Drew Lyon: Okay, so anything you have, what a minimum of five years worth of data. How much data do you have to have before you feel like you can use machine learning?
Karan Sandhu: So for machine learning, if you have more than 500 plants, that means you are good to go. But if you want to proceed with the deep learning model, so I will suggest try to have minimum 20,000, sorry, 2,000 lines. That means, if you have more data, go for the deep learning model. On the other hand, if you are implementing deep learning model on a smaller data set, you will be getting over-fitting in the model. That means definitely you are getting good accuracy, but that’s not true.
Drew Lyon: Okay, so you might be making some false predictions, I guess.
Karan Sandhu: Yes, yeah.
Drew Lyon: Okay, what are the potential difficulties faced while you’re building a model for this deep learning? And as breeders who aren’t mostly trained in these tools, how do you go about using it?
Karan Sandhu: Yeah, that’s a really good question. So exactly, when I started working with Dr. Arron Carter, so I was not having any experience on machine learning and the deep learning. So the best thing we work this is– we tried to have some collaboration with the computer scientists and statistical genomics at the WSU. So those collaboration help us initiating several things, even different classes which are taught at WSU, like data science machine learning and the neural network. So I took all those classes just to build my machine learning, data science and deep learning skills. So once I got all those skills, so we tried to make prediction and let our collaborators know that we are getting these kind of results just to verify those things from computers scientists that we are on a good track. So that’s how we tried to build those skills. And I completely agree that plant breeders or the geneticists are not trained in this way. But I feel like this is a need of ours. So as we are moving ahead, we should learn these things in our plant breeding.
Drew Lyon: Okay, are you familiar with other breeding programs around the country or the world that might be trying this as well?
Karan Sandhu: Yeah, so exactly like you can take an example of Kansas State University or you can look at Florida. So most of the universities are applying this machine learning and the deep learning model. Even the companies like Bayer, Syngenta, they have adopted this thing like ten years back. But in our public sector, we are slowly and slowly making transition, yeah, but we are definitely on it.
Drew Lyon: Okay, so how do you think it will help predict traits for multi-environment situations like we have here in eastern Washington, where you have everything from the high rainfall area here in the Palouse to the very dry conditions out in the Horse Heaven Hills?
Karan Sandhu: Yeah, so exactly, yeah. We were also thinking of implementing models in certain areas, but progress is made, but that’s not that far. We are still working on it. The main thing is like, as we have different weather parameters or the different climatic zone in the eastern Washington, in order to make such kind of prediction, we need to include weather data set. So if we are having more weather information, that means we can encode for genotype by environmental interaction. So in case of plant breeding, genotype by environmental interaction plays a very critical role. For example, if we have planted a line Pullman, that will perform differently compared to a line which is performed in the Lind condition. Or you can take example of Ritzville. So in order to make those prediction accurate weather data set, enough previous year data set should be available to make such kind of prediction. And the other thing is like we need to make some better changes in the model which can account for G by E interactions. So definitely we are working on that, but we haven’t got that much accuracy yet.
Drew Lyon: Okay, so the goal would be maybe not to need as many sites eventually. Where you could just predict by growing it in one or two locations. You can predict how it might do in other environments?
Karan Sandhu: Yes. Yeah, definitely, yeah. So that’s what we are working on. Like this plant at two locations and then try to predict most of the other.
Drew Lyon: Okay, so I think you mentioned accuracy earlier of about 60 percent for one of the traits. What, in general, is your accuracy on predictions?
Karan Sandhu: So accuracy usually depends upon the trait which we are working on. For example, grain yield is a very complex trait because it is controlled by different genes and environmental interactions. So we are getting accuracy in the range of 40 to 60 percent. On the other hand, like traits like protein content or some kind of disease resistance, as those traits are controlled by some less number of genes, and we are getting an accuracy up to 70 percent. That means it depends, completely depends upon a trait and the environment on which you are working on. But definitely, we can improve the accuracy by including the genotype by environmental interactions.
Drew Lyon: Okay, but yield is what many growers are interested in, yet it’s one of the more difficult ones to figure out, huh?
Karan Sandhu: Yeah, it is really different, difficult to figure it out, yeah. But with the help of machine and deep learning model, we are thinking that we can improve it more.
Drew Lyon: Okay, a lot of breeding and scientists are using these high throughput phenotyping methods, where you just move these little plants through a system and take measurements. Can you use that kind of data? Or is all this data from field that you’re using?
Karan Sandhu: Yeah, definitely, we are using that kind of data set. So we are using drone technology for missing different reflections from the plant. So those reflection give us information about various physiological processes like chlorophyll status of a plant or any kind of water stress which the plant is facing. So we have included that data set in our machine learning and deep learning model to make prediction. And you won’t believe it that with the use of those data set, we improved our prediction accuracy by 10 to 17 percent for various trait. That mean if you are, if we are having more data sets, definitely we are getting more prediction accuracy. And high throughput data set is a key to increase our prediction accuracies.
Drew Lyon: Okay, so the bottom line is the more data, the better, huh?
Karan Sandhu: Yeah, definitely, yeah. [ Drew laughs ]
Drew Lyon: And if you listen or read in the popular press out there, there’s a lot of concerns about artificial intelligence, and yet it can be a very useful tool when used appropriately.
Karan Sandhu: Yes, yeah, definitely. But if we are getting accuracy, better accuracy and we’re not sure that whether it’s over-fit or some kind of false positives, so we need to think those things exactly that, are we going in the right direction? Or are we just give getting some false predictions?
Drew Lyon: Okay, I know one of the things that everybody’s doing these days is collecting a lot more data and then trying to figure out what to do with it. It sounds like you, artificial intelligence, and deep learning is a tool to help you use all those, all that data to actually come up with useful tools.
Karan Sandhu: Yeah, definitely, yeah. You can take the example of Tesla. Like they have used like previous years’ data set, and now the car is working very efficiently. Similarly, we are trying to go in that direction.
Drew Lyon: Okay, very interesting. It’s kind of outside the realm of agriculture as I’ve known it for many years. But the CS applying this new technology’s pretty exciting. We’ll be watching to see what comes out of this through your program. How much longer do you have to go yet?
Karan Sandhu: So I’m left with one and a half years. So I will be done by next summer, summer 2022. And hopefully, I will be done with good results, yeah.
Drew Lyon: Excellent. We’ll look forward to seeing that.
Karan Sandhu: Thank you. Thank you so much. Thank you for inviting me.
[ Music ]
Drew Lyon: Thanks for joining us and listening to the WSU Wheat Beat podcast. If you like what you hear don’t forget to subscribe and leave a review on iTunes or your favorite podcasting app. If you have questions or topics, you’d like to hear on future episodes please email me at drew.lyon — that’s email@example.com –(firstname.lastname@example.org). You can find us online at smallgrains.wsu.edu and on Facebook and Twitter @WSUSmallGrains. The WSU Wheat Beat podcast is a production of CAHNRS Communications and the College of Agricultural, Human and Natural Resource Sciences at Washington State University. I’m Drew Lyon, we’ll see you next time.