1. Slow is Smooth & Smooth is
Fast!
Understanding the Kinetics &
Thermodynamics of Cannabis Extraction
Dr. Markus Roggen
President & CSO
2. Introduction
DELIC Labs is a research venture that seeks to add fundamental scientific
insight to the field of cannabis and mushroom production.
We seek to support the cannabis and mushroom industries by
establishing a centralized hub in Vancouver, BC, for collaborative
research focused on:
• Process Design
• Process Optimization
• Process Analytics
• Formulation Research
3. Collaborative Research
DELIC Labs collaborates with academic, industry and private groups
around the globe. Some highlights of those collaborations are:
• University of British Columbia, Vancouver
• Loyalist College, Belleville
• Via Innovations by Dr. Monica Vialpando
• Veridient Science by Dr. Linda Klumpers
Fundamental Collaboration
4. Research Topics
• Chemometrics and data analytics for process control and optimization
• Kinetic studies to understand mechanisms
• In-process analytics for process control
• Computational studies to understand mechanisms
• Process development, like crystallization
Fundamental Cannabis and Mushroom Chemistry
5. Abstract
• CO2 Extraction Optimization and its Limitations
• Optimization beyond Single Case
• Flow Rate and Yield
• Gaussian Processes and Bayesian Optimization
• Machine Learning and Artificial Intelligence
8. CO2 Extraction Optimization and its Limitations
• Linear optimization, even multifactorial, is cost prohibitive
• Looking at one parameter at time is to slow
• Optimizing more than 3 factors takes to long
Temperature (˚C)
Pressure (psi)
Yield THC (g)
34
60 1100
1900
47
1500
0
9. CO2 Extraction Optimization and its Limitations
• Just looking at pressure and temperature is not enough
• We look at >20 controls, every minute, all the time
10. CO2 Extraction Optimization and its Limitations
• Optimization is conventionally done empirically
• explore how altering conditions changes the outcome
• This is resource intensive (time, materials, money)
• Discounts or does not use all the data generated
• Results are often biased by the optimization and not entirely generalizable
to different inputs or extractors
• Garbage in, garbage out
• Not for material
• But for data
11. Optimization beyond Single Case
• Producers: XXX
• Instrument Types: 6
• Individual Runs: XXXX
• Datapoints: ~100,000
12. Flow Rate and Yield
• Counterintuitive observation from dataset:
• Slower flow rate decreases CO2 needed
• Slower flow rate increases extract purity
13. 0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Gaussian Processes and Bayesian
Optimization
Linear Models
• Classical dimensionality reduction techniques and linear regression
models probe the relationship between various input variables and yield
• Challenges in proper cross validation
• Test errors were noisy
• Linear techniques do not provide good accuracy
• Do not provide a built-in uncertainty estimate
14. Gaussian Processes and Bayesian
Optimization
Gaussian Processes (GPs)
• Fit an entire family of curves to the observations
• Infinite number of functions fit our finite set of observations
• GPs assign a probability to each of these functions
• Mean of probability distribution represents most likely prediction
• Use the variance of the distribution as an uncertainty estimate.
15. Gaussian Processes and Bayesian
Optimization
Bayesian Optimization (BO)
• GP model provides mean and uncertainty
• BO identifies highest uncertainty of GP
• BO choose the next best point to sample based on both its uncertainty and
the function value at that point
18. Gaussian Processes and Bayesian
Optimization
• Bayesian optimizer outputs a set of experimental conditions to further to
refine model
19. Gaussian Processes and Bayesian
Optimization
• Bayesian optimizer outputs a set of experimental conditions to further to
refine model
20. Machine Learning and Artificial Intelligence
• Working model for subset of data
• Can see effect of input material on optimal conditions
• Probe effects of more factors
• Expand to whole dataset, fill gaps in dataset
• Partner with more producers to get everyone better results
• Biggest problem is bad data
21. Expertise
CSO: Dr. Markus Roggen
Dr. Roggen has been actively involved in the cannabis industry for over 5 years in executive
positions overseeing production, R&D and process optimization for multiple producers. Dr.
Roggen is also a trusted advisor and mentor for multiple startups, startup accelerators and
organizations.
DELIC Labs Team
Our team covers a wide range of expertise,
including analytical chemistry, process
chemistry, engineering physics, data science
and statistics.
Scientific Advisor: Prof. Glenn Sammis
Prof. Sammis is an Associate Professor in the Chemistry Department at the University of British
Columbia. He has built an internationally recognized research group working on the
development of novel synthetic methods for the preparation of natural products and
pharmaceuticals.