Bayesian Modeling in Biotech: Using PyMC to Analyze Agricultural Data (Indigo Ag)

Manu Martinet, Bill Engels and Thomas Wiecki

## Timestamps
00:00 Thomas Wiecki does PyMC introduction
02:49 Thomas introduces self
03:33 Manu Martinet introduces self
04:25 Bill Engels introduces self
05:10 Panel discussion begins
06:51 Testing crop yields on fields
08:16 How do you sell the product to farmers?
10:55 Data modeling and challenges
13:04 Goal of the project: Estimate the spatial pattern and remove it to get the treatment effect
15:20 Gaussian processes and how they are used
18:04 Spatial Gaussian Processes
19:09 Spatial effects
22:13 Examples fields to show the spatial components
24:28 Question: How does modeling the spatial component with a Guassian process compare with other simpler methods?
25:47 Question: With the Gaussian Process(GP) can you estimate the spatial scale?
28:06 Question: How does the Gaussian Process deal with latent variables?
30:08 Advantages of the a Bayesian framework
35:00 Collaboration between Indigo and PyMC Labs review
42:43 Question: What were the biggest challenges in the study?
45:29 Question: Is there any example online for PyMC based Hierarchical Gaussian Processes(GP) regression?
46:37 Question: How did the decomposition work out between signal, spatial and noise and how do you balance the confidence between what is signal and what is noise?
47:35 Question: How to effectively use Bayesian methods to substantiate product claims to regulatory bodies?
48:07 Thank you!
#BayesianModeling #BiotechDataAnalysis #PyMCAnalysis #AgriculturalData #StatisticalInference #MCMC #UncertaintyQuantification #ComputationalBiology #CropAnalysis #StatisticalModeling #DecisionMaking #ProbabilisticProgramming #DataScience #BayesianStatistics #ParameterEstimation #PriorDistribution #PosteriorDistribution #SensitivityAnalysis #BiotechResearch #AgTech

## Event Description
In this panel discussion we will discuss why Bayesian modeling is such a powerful tool for solving problems in biotechnology. As experiments are often complex it is important to build custom and causal models that accurately represent the structure of the experiment in the statistical model. As important decisions are made based on limited data, quantifying uncertainty at every level becomes important.

## About the speakers
Thomas Wiecki
Dr. Thomas Wiecki is an author of PyMC, the leading platform for statistical data science. To help businesses solve some of their trickiest data science problems, he assembled a world class team of Bayesian modelers founded PyMC Labs -- the Bayesian consultancy. He did his PhD at Brown University studying cognitive neuroscience.

Bill Engels
Bill Engels is a Data Scientist at PyMC Labs with experience solving problems using Bayesian methods in several different industries. He contributed and helps maintain the Guassian process module in PyMC. He has a Masters in Statistics from Portland State.

Louis-Emmanuel (Manu) Martinet
Louis-Emmanuel is a Lead Data Scientist in the Microbial Products department at Indigo Ag. He is responsible for designing analysis pipelines to understand the performance of candidate microbes in field trials across geographies and environmental conditions (e.g., drought, soil quality, …). He joined his current company after going through the Insight Data Science Fellowship program. Prior to that, he spent several years as a Post-doctoral fellow at Boston University, Harvard, and the Massachusetts General Hospital where he analyzed human electrophysiological brain recordings in order to study brain networks during seizures. He received a Ph.D. in Computer Science in 2010 from the Sorbonne University in Paris, France.

## Connecting with PyMC

#pymc #bayesian #python

[SjorsNKaon]

in the section about Spatial GP, how does the GP "know" that only the spatial part is being modeled instead of the combined effect of spatial+treatment on the yield?

[davidzapata]

please, subtitles in spanish!!!! from latam