Sainsbury Laboratory
Ioana Valea <ioana.valea@slcu.cam.ac.uk>
Feedback in discussion
The machine learning methods that might be used to address your problem are reliant on large training data sets to be effective, and it's not clear that this could be done within the timeframe.
On the most optimistic timescale, could we imagine project structure along the following lines?
Week 1 - scoping and planning Week 2-4 - build, program and test the image capture array and data annotation system Week 5 - collect and label/annotate data Week 6 - train machine learning system Week 7 - integrate and demonstrate results
Would 1 week of growing time be long enough to capture sufficient variety of images for your research question? This timescale would also require that your experts be available for data annotation / labelling to be done in parallel with the data collection.
Project proposal for Phenotyping
Internal project description Goal is to take pictures of growing strawberries at different developmental stages to observe potential phenotypical differences in Biosensor/Highlighter transformed plants. The phenotypic traits include: • Leaf enrichment of 2-month-old plants, i.e. transition from vegetative to reproductive stage o how and where the next leaves emerge • Runnering vs. flowering o Does exogenous GA affect the switch between them? o Does runnering exclude flowering? • Flower development o Is flowering induced by blue light periods in the morning hours?
• Fruit development
Materials: Plant material are 2-months old strawberries grown in a pot. The time-lapse pictures must be taken in light and dark conditions over a long period of time: o Runnering/Flowering 2-4 weeks of data collection o Flowering/Fruiting 5-6 weeks of data collection
Method: Arduino platform with control of light (array) LED lights in a growth chamber seem to work well. Lights and cameras (light and IR) must be under Arduino control. Humidity and room temperature can be regulated by the growth chamber. Cameras: three cameras per pot (light camera for side growth and from above, one IR camera from the side) At any one time at least two plants (WT vs. treatment/transformant) have to be able to be observed. If possible two pots per treatment, total at 4 pots at any one time is best. Each pot needs its own cameras and picture collection due to the analysis afterwards. Data analysis: Design a package for image-based tracking of strawberry development Jupyter/Python • Tracking of runner growth • Tracking of fruit development • Mature flower characterization Students should write code that could be eventually shared via the PlantCV/Fiji platform or use available one. The program must be able to identify and differentiate between runners, flowers, leaves and fruit. Also identifying colours and their intensity is important. Detecting size and changes in size is imperative. Being able to quantify changes in size and structure is important. E.g. different leaf shapes between treatments, different flower, fruit and runner sizes as well. Taking pictures with object identifiable by bare eye is also an important point.
Course of action
We will be firstly responsible for building a scaffold of how everything should be positioned in the chamber, students will be responsible for optimizing the positioning of cameras and sensors for optimal data collection. They will then continue with the Arduino programming and data analysis.
Project advertisement draft for students:
Are you interested in consulting for a team of international and energetic synthetic biologists at Sainsbury Laboratory to contribute developing a new phenotypic system for strawberries? We are currently designing an Arduino-based imaging set up to track developmental stages of woodland strawberry. We need you, a team of talented computer scientists knowledgeable in Python to develop the basic platform for the Arduino system and the subsequent image analysis preferably on PlantCV and/or Fiji.
