CANNABIS ANALYTICS Toward a Chemotype-Based Labeling System
SAVINO SGUERA Formulation Chemist, 2010-2011 B.S., Biomedical Engineering, 2008 Founder & CSO, 2013-Present Laboratory Director, 2011-2013 Laboratory Director, 2014-2016
INTRODUCTION Cannabis-related data has grown exponentially in the last decade Mostly chemical profile (cannabinoids, terpenes, contaminants) Includes patient/user observations, genetic data, cultivation practices and observations Number of potentially-active analytes in chemical profile exceeds other products Challenge for the industry, opportunity for knowledge Good data gathering practices and bioinformatic techniques give insight to underlying mechanisms This leads to new treatments, products, techniques
TERPENES AND TERPENOIDS How are genetic variants organized according to terpene profile? How does chemical profile translate to medical efficacy? How do genetic and environmental factors translate to a specific chemotype?
DATA ACQUISITION Based on data gathered in Nevada and Massachusetts Terpenes: GC/MS, headspace, full evaporation technique (FET) a-pinene, camphene, myrcene, b-pinene, carene, a-terpinene, limonene, cymene, eucalyptol, g-terpinene, d-terpinene, linalool, isopulegol, caryophyllene, humulene, nerolidol, guaiol, caryophyllene oxide, bisabolol, ocimene Sampling Preparation Label Analysis Chromatography
PREVIOUS STUDIES Hazekamp (2015 & 2016) No significant clustering around strain names Indica/Sativa could be predicted by select terpenes
DiscOmic System Performed PCA and other unsupervised learning algorithms to see how the terpenes related to each other Reduced data to visually and numerically represent a complex chemical profile Goal: quantitative but simple system for doctors, patients, and recreational users alike Can be applied to many similar complex phytochemical mixtures (wine, hops, coffee, kratom, etc) Sampling Preparation Label Analysis Chromatography
PREVIOUS STUDIES Fischedick 2017 Strains can be subdivided into overlapping groups based on dominant terpenes
RECOMMENDATION CHALLENGES No objective way to recommend or prescribe Currently done by budtender or herbal specialist Relies on olfactory interpretation of chemical profile Strain names are not standardized, and do not guarantee a particular chemical profile Need: summarize chemical profile in easy-to-interpret way
DiscOmic System Split into 4 Principal Components: - Find Eigenfunction of Data Covariance Matrix - Calculate first 4 components - Normalize around 0 Generate Label: - Color by sign of PC value (+/-) - Radius is absolute PC value
DiscOmic System Features: - Quantitative interpretation of terpene profile - Visual comparison between strains - Avoids strain name - 80% data variance retained Difficulties: - Values are abstract and non-intuitive
SELF ORGANIZING MAPS Powerful tool to visualize how chemical profiles are organized Samples become grouped by similarity and viewed as a 2D grid
SELF ORGANIZING MAPS CBDA Plane THCA Plane 1.5 0.0 29 23 14.3 2 0.0 0.0 23 1.0 0.0 0.0 5 Low Concentration High Concentration 14 20
SELF ORGANIZING MAPS b-pinene a-pinene 6.8 1.7 2.84 0.8 2.4 0.94 1.18 0.5 1.33 0.58 0.89 0.5 0.4 0.7 2.9 1.1 1.2 0.56 0.96 2.1 0.7 0.4 0.27 5.6 1.3 0.64 Nerolidol 1.6 0.41 0.8 0.92 Humulene b-caryophyllene 1.8 0.42 0.6 1.5 0.37 1.61 0.4 1.7 2.7
SELF ORGANIZING MAPS 0.9 7.3 d-limonene 4.0 1.0 0.59 2.0 a-terpinene 0.000 2.09 1.25 0.7 0.6 0.001 0.69 2.8 0.40 0.51 0.95 d-terpinene 0.001 0.0 l-linalool 1.25 g-terpinene 0.0 0.005 0.007 0.011 0.0 0.033 0.000 0.049 0.6 0.029 0.9 0.233 0.001 0.0 0.042 4.6 0.175 0.13 0.00 0.03 0.02 0.11 0.00 0.33 0.004 d3-carene 0.004
SELF ORGANIZING MAPS 7.2 3.9 0.23 0.40 4.0 0.13 3.3 1.6 0.46 0.22 0.09 1.9 1.1 0.27 1.10 Ocimene b-myrcene 5 4 Ocimene 0.6 3 2 1 0 0 5 Myrcene 10 15 20
OPPOSING SUBGROUPS Combine correlated terpenes into subgroups - 1: caryophyllene, humulene, nerolidol - 2: a-pinene, b-pinene - 3: limonene, linalool - 4: a-terpinene, d-terpinene, g-terpinene, d3-carene - 5: b-myrcene - 6: ocimene
AXES 1 0.8 0.7 0.8 Group 4 Group 2 0.6 0.6 0.4 0.5 0.4 0.3 0.2 0.2 0.1 0 0 0.1 0.2 0.3 0.4 Group 1 0.5 0.6 0.7 0.8 0 0 0.2 0.4 0.6 Group 3 Above: Normalized Scores for Combined Terpene Subgroups 0.8 1
AXES 1 0.8 0.7 0.8 0.6 0.5 0.6 0.4 0.4 0.3 0.2 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0 0 0.2 0.4 0.6 - Assign Sample to 1 of 3 Clusters for each axis - Sample can be Group 1 dominant, Group 2 dominant, or split - Sample can be Group 3 dominant, Group 4 dominant, or split 0.8 1
AXES 1 0.8 0.7 0.8 0.6 0.5 0.6 0.4 0.4 0.3 0.2 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0 0 0.2 0.4 0.6 0.8 Caryophyllene, Humulene, Nerolidol Limonene, Linalool a,b-pinenes a,d,g-terpinenes, Carene 1
MYRCENE / OCIMENE Ocimene Myrcene 5 Ocimene 4 3 2 1 0 0 5 Myrcene 10 15 20
Caryophyllene, Humulene, Nerolidol Dominant Group a,d,g Terpinene, Carene Dominant Group High Myrcene, Low Ocimene a,b Pinene Dominant Group Limonene, Linalool Dominant Group High Ocimene, Low Myrcene
Thickness of radii correspond to combined subgroup concentration Ratio of split colors correspond to ratio of each group w.r.t. total
Gorilla Glue Juicy Fruit SFV OG Platinum Blue Dream Sour Sage Platinum Blue Dream
Gorilla Glue Juicy Fruit SFV OG Platinum Blue Dream Sour Sage Platinum Blue Dream
Similar Chemical Profiles Gorilla Glue Juicy Fruit SFV OG Platinum Blue Dream Sour Sage Platinum Blue Dream Similar Chemical Profiles
Caryophyllene, Humulene, Nerolidol, Limonene, Linalool, Ocimene Gorilla Glue Juicy Fruit SFV OG Platinum Blue Dream Pinenes, Terpinenes, Carene, Ocimene Sour Sage Platinum Blue Dream
Gorilla Glue SFV OG Myrcene, Terpinenes, Carene Sour Sage Almost equal Groups 1 and 2 Juicy Fruit Platinum Blue Dream Platinum Blue Dream
LIMITS OF THE MODEL Requires diverse dataset for universal use Different models for flower, extract, etc. Sample prep and data acquisition methods must match to combine data
DATA AGGREGATION CHALLENGES LC/UV Benefits Difficulties LC/MS -integrates with -improved id cannabinoids and quant -inexpensive -resolve early monoterpenes -id/quant at low conc. -id at low concentrations GC/MS, liquid -faster separation - NIST Lib. ID -inlet maintenance -analyte breakdown GC/MS, headspace -clean loading -low interference -sesquiterpene quant. -more time/cost
DATA AGGREGATION CHALLENGES Method standardization = more data = higher value data In-depth experiments to compare results of different methods Find way to aggregate data from different methods/ different analytes
TERPENES AND TERPENOIDS How are genetic variants organized according to terpene profile? How does chemical profile translate to medical efficacy? How do genetic and environmental factors translate to a specific chemotype?
CURRENT PROJECTS Using Artificial Neural Networks to discover connections between chemical profile and: Medical efficacy via patient-reported survey Heart Rate Variability using Lief Wearable Device
FUTURE Use observed data to design full clinical trial Correlate chemical profile and observed effects with: Metabolite levels in fluids/tissue ECG, EEG, MEG, HRV, GSR Psychological evaluations Impairment Patient genetics
REFERENCES Fischedick Justin T.. Cannabis and Cannabinoid Research. March 2017, 2(1): 34-47. https://doi.org/10.1089/can.2016.0040 Hazekamp Arno, Tejkalová Katerina, and Papadimitriou Stelios. Cannabis and Cannabinoid Research. September 2016, 1(1): 202-215. https://doi.org/10.1089/can.2016.0017 All numerical calculations performed using Octave Labels generated using code from D3: Data Driven Documents SOM Source Code: Melikerion by FinnDiane Data courtesy of MCR Labs and Confident Cannabis