Team work - Acknowledgements

Data Scientists

  • Linley Jesson
  • Mark Wohlers
  • Harris Lin
  • Blake List
  • Daniel Bentall
  • Patrick Connelly
  • Peter Alspach
  • Nathan Tomer
  • … many others!

Breeders

  • John McCallum
  • Richard Volz
  • Ron Beatson
  • Kerry Templeton
  • Alastair Curie
  • Satish Kumar
  • Robert Shaffer
  • Liz Popowski
  • … many others!

I was going to talk about this…

… and its health benefits …

… and how we use genomic selection …

… to optimise properties …

… and how we use AI to improve our genomic selection results …

… but IP said NO!

So instead … I will talk about beer!!!

We aim to breed hop varieties with a unique aroma

Selection for hops is based on sensory comments of beer

Examples:

  • "This beer is delicious, it has a fruity aroma.",
  • "This beer tastes bad, too bitter. Lovely after taste though.",
  • "A very boring flavour palette. I would not recommend it."

Analyses sensory comments= subject specific lexicon + valence shifters:

Subject specific lexicon:

  • Positive sensory words: Bitter, cat pee, sweaty, unique, different
  • Negative sensory words: Sweet, ok, normal

Examples of importance of lexicon & valence shifters:

  • "I really enjoy the flavours and aromas of this beer. It has nice hoppy characteristics and a hint of cat pee." - Positive
  • "I appreciate the aroma, but the taste is too bitter for me." - Negative
  • "This beer has aromas of pineapple and citrus. Quite fruity on the nose. It tastes ok" - Neutral

Update the original lexicon with new words and their sentiments manually

In R: existing packages (SentimentAnalysis, SentimentR, lexRankr, lexicon, textstem) + own (adjusted) code:

hops_lexicon <- sentimentr::update_key(
  lexicon::hash_sentiment_jockers_rinker, 
  x = data.frame(words = c("interesting", "different", "unusual"),
                 polarity = c(0.80, 0.7, 0.60),
                 stringsAsFactors = FALSE)
  )
 
#hops_lexicon
hops_lexicon[which(hops_lexicon$x == "interesting")]
##              x    y
## 1: interesting 0.75

Hapi Talk

To improve our beer sensory specific lexicon we've build our own shiny app…

Current results

Outcomes

  • Wine and beer sensory lexicon IP
  • Faster data collection
  • Faster data analysis
  • More reliable selection processes (based on quantitative scores)

Thank you

And oh yes… if you choose so, hops can also be healthy.

www.plantandfood.co.nz