Modeling seed viability over time to decide when to regenerate seed lots in long-term storage
Philip Dixon (Iowa State University )
Friday 21st March, 2014 15:00-16:00 Maths 204
Plant germplasm storage facilities (“seed banks”), such as Kew’s Millennium Seed Bank, provide long-term storage of seeds. By preserving different genetic varieties of crop plants, many of which are no longer grown commercially, the seed banks provide plant breeders access to a broader range of genetic material. By preserving different species of plants, the seed banks make it possible to reintroduce species that have be extirpated in the wild. Both functions require that seeds remain viable during storage. Current practice, both at Kew and at seed storage facilities in the United States, is to test the germinability of seeds on a regular basis. If germination falls below a critical percentage, a new collection of seeds needs to be obtained. The problem is how frequently should seed lots be tested? There are costs to both testing too frequently and testing too infrequently. We develop a method to identify a reasonable compromise.
Our method has three parts:
- model germination as a function of seed age in a way that combines data from many related seed lots,
- for each seed lot, estimate the distribution of age at which the seed lot is predicted to reach the critical germination percentage, and
- find the quantile of that distribution that balances the cost of premature testing and the larger cost of insufficient viability. The first part is a random coefficient regression model, the second is the Bayesian posterior distribution of time to reach the critical germination percentage, and the third is based on a smoothed receiver operating characteristic curve. These steps are illustrated using data from 2,833 accessions of maize.