Modeling yeast dynamics during fermentation

New models for the dynamics of yeast in suspension during brewery fermentations are presented

Transcript

1. Modeling yeast dynamics during fermentation A p o w e r t o o l f o r t h e q u a l i t y t o o l b o x M B A A d i s t r i c t S E F a l l m e e t i n g O c t o b e r 2 0 , 2 0 1 8 A r t h u r R u d o l p h c r a f t b e e r s c i @ g m a i l . c o m @ c r a f t b e e r s c i
2. BIO • Quality manager at First Magnitude Brewery – Started in brewhouse & taproom in 2015 – Became quality manager in 2016 • PhD candidate in the Department of Biology, University of Florida – Researching flocculation and the evolution of flocculins genes • 2010 B.S. Integrative Biology from the University of Illinois • Over 10 years of research experience in genetics, genomics, and evolutionary biology
3. Outline • Methods for quantifying yeast • The benefits of monitoring your yeast • Current models of yeast population dynamics • New models of yeast population dynamics • What these new models can do for you!
4. Methods for quantifying yeast • Microscope and hemocytometer – Cheap, well studied, decent accuracy, fairly slow – Viability testing – www.hemocytometer.org • Automated cell counter – More expensive, better accuracy, faster – Many models allow viability testing • Turbidity measurement, absorbance @ 600nm – Very fast, very cheap if you’ve got a spec – No viability – Measurements do not correlate across beer types
5. The benefits of monitoring your yeast • Yeast population dynamics directly impact flavor – Pitch rate, oxygenation, fermentation temp, etc. matter for flavor profile and population dynamics – Population dynamics are an easy/cheap QC endpoint for monitoring flavor profile consistency • Gives a heads up on yeast quality for harvesting and re-pitching before your next brew day • Helps differentiate between hung fermentations from PYF or high mash temps • Monitor flocculence of the yeast across successive batches – Major consideration when deciding how many time to re-pitch yeast Your yeast have something to say to you
6. Why do we need a model of yeast population dynamics? 0 1 2 3 4 5 6 0 1 2 3 4 5 6 7 Cells in suspension Time (days)
7. Current models of yeast population dynamics • ASBC yeast-14 method (shown here) – Max population size – Max population growth – Time of max population • Paired logistic models (Lake et. al. 2009) – Many parameters • Model free spline fits – No predictive power
8. New models of yeast population dynamics Data collected by Mario Guadalupe Daqui, Presented at Brewing Summit 2018 ASBC yeast-14
9. What these new models can do for you! • Monitor for consistency – For small or large data sets • Optimize sampling times – Save time and money by minimizing sampling effort – Get the most out of your effort • Detect PYF positive malt
10. Why does parameter number matter? 0 1 2 3 4 5 6 0 1 2 3 4 5 6 7 Cells in suspension Time (days)
11. W h y d o e s p a r a m e t e r n u m b e r m a t t e r ? Step model 4 parameters
12. W h y d o e s p a r a m e t e r n u m b e r m a t t e r ? Gamma model 2 parameters
13. Yeast-14 population dynamics • The basic Normal model performs best • Yeast-14 method only lasts 78 hours, – less impact of tail
14. Monitoring for consistency: Conclusions • The Step model works best when there are many data points • The Gamma model works best when there are few data points • The Normal model fits data from the yeast-14 method best
15. Optimize sampling time • Improve model fits by sampling the peak – Add additional sampling during that time OR – Move sampling from tail to peak • This pattern is true regardless of relative sample number 0 1 2 3 4 5 6 0 1 2 3 4 5 6 7 Cells in suspension Time (days) MSE OF GAMMA MODEL K THETA ONE DAILY 0.0045 31 TWO DAILY 0.0258 1694 TWO DAY 2 0.0020 10 TWO DAY 3 0.023 11
16. Premature Yeast Flocculation • Picture/graph! • Caused by fungal infection of barley during growth or malting • Sticky yarn hypothesis • Anti fungal hypothesis • More common in wet years • Results in – under attenuated beer – Package over pressurization – Extra time & labor costs ASBC Buzz Vol 78, Num 6
17. Detecting PYF • Use Linear Discriminate Analysis (LDA) • Compare models fit to data with infected vs normal malt • Leave One Out Cross Validation • Plot Gamma model parameters Gam Gam+ Norm Tilt Step CV correct 90% 83% 90% 93% 80%
18. Detecting PYF • Use Linear Discriminate Analysis (LDA) • Compare models fit to data with infected vs normal malt • Leave One Out Cross Validation • Plot Gamma model parameters Gam Gam+ Norm Tilt Step CV correct 97% 97% 95% 98% 96%
19. Detecting PYF: Conclusions • The fermentation and data collection methods in yeast-14 work well for detecting PYF • All 5 models perform similarly well at identifying PYF with LDA – Given the similar performance, it’s best to use the simplest model, the Gamma • Even without performing a LDA, just examining the parameters can indicate PYF – If the k parameter is greater than 1.7 AND – If the theta parameter is less than 30 – You should be concerned about PYF
20. Cheers to: • Dr. Colette St. Mary • Dr. Andrew MacIntosh • Dr. Alex Speers • First Magnitude Brewery • University of Florida • MBAA
21. Models • Gamma • Gamma+ • Normal • Tilt • Step Eq. 1 () = −1− Eq. 2 () = + −1− Eq. 3 () = ∗ −1 2 ∗ −µ 2 Eq. 4 = ∗ + ∗ −1 2 ∗ −µ 2 Eq. 5 () = ∗ − 1 2 ∗ −µ /(1+()∗) 2 1+()∗ + ( − 1+ ∗() )
22. Citations • Lake, J. C., Speers, R. A., Gill, T. A., Reid, A. M. and Singer, D. S. (2009), Modelling of Yeast in Suspension During Malt Fermentation Assays. Journal of the Institute of Brewing, 115: 296-299. doi:10.1002/j.2050-0416.2009.tb00384.x