Classifying Cannabis Strains: Room For Improvement

When consumers go into a cannabis dispensary, they typically hear about sativas, indices, and hybrids. However, as the cannabis industry has developed, these three classifications are no longer accurate descriptions of cannabis strains. To provide more reliable descriptions, cannabis entrepreneurs have to start using data-driven methods and deep learning to classify cannabis correctly and properly inform consumers.

Cannabis pioneers and entrepreneurs have used the terms Sativa, Indica, and hybrid to describe a strain’s effect for decades. But these terms only describe the width and height of a plant’s leaves and sometimes its chemotype. Not every Sativa is guaranteed to deliver more energy, just like every indica is not guaranteed to increase relaxation.

Ethan Russo, a neurologist who’s respected worldwide for his research in cannabis psychopharmacology, has weighed in heavily on this issue. He believes that classifying cannabis strains as sativas, indices, or hybrids for commercial use is “nonsense.” In an interview with Leafly, he said the clinical effects of cannabis have nothing to do with how a particular cannabis cultivar looks.

Jeffrey Raber, a chemist who founded the first independent testing lab to analyze cannabis terpenes for commercial use, echoed Russo’s sentiments. He told Leafly that there is no “factual or scientific basis” to make these broad recommendations. In fact, Raber said cannabis entrepreneurs should learn which standardized cannabis compositions lead to specific effects, while also taking into account the cannabis’ form factor, dosage and the person using it.

Collecting and interpreting data is the best way to learn the effects a cannabis strain will have. A data-driven approach forces cannabis entrepreneurs to base their classifications on hard facts and figures, not nomenclature.

By using data collection and analysis, it’s possible to prove that cannabis is not as simple as the industry once thought, but since cannabis entrepreneurs and users alike are fond of the Sativa, Indica and hybrid classification, moving away from this model might take a bit of time and effort, especially on the entrepreneur’s part.

Over the past five years, we used deep learning to scan the entire internet to create a master dataset that encompasses different cannabis strains. Our initial analysis included around 3,000 different strains, and we created a feature set based on the strains we studied.

A “feature set,” in this case, is the list of attributes that are informative for the classification or prediction of specific outcomes. An easy example to understand of a feature set is in the context of predicting the sale price of a house. In this case, a useful feature set would include essential attributes like the home’s number of bedrooms, number of bathrooms, square footage and previous sale price. All of these attributes are crucial facts that dictate how much money a seller can get. However, information such as the middle initial in the third homeowner’s name isn’t vital or insightful material that a seller would need to include in their feature set.

When applying this analogy to cannabis, we developed our feature set by looking at the cannabinoids and terpenes in each strain and the ratio in which they were found. We also looked at the opinion the public applied to each particular strain and paid specific attention to the words people used to describe every cannabis product.

Once we created the feature sets, we told the algorithm to find the best classifications for each strain. The algorithm generated six distinct groups, as opposed to the three traditional groupings of Sativa, Indica, and hybrid. Those six classes are what we call “activity groups,” the categories of strains and other cannabis products that are more likely to provide a specific effect because of their unique chemical composition.

The practice of using traditional cannabis terminology like Indica and Sativa to determine effects is an outdated model with no scientific backing. To educate consumers and the industry at large, cannabis entrepreneurs need to use data analysis and deep learning to classify cannabis strains and figure out which ones are most likely to have specific effects. Better classifications will lead to more consistent experiences and an industry that’s known for producing reliable products.

Nicco Reggente is a co-founder of Strain Genie, a Los Angeles-based startup that offers a cannabis DNA test to match patients and recreational users with the right cannabis products, strains, terpenes, and dosage recommendations. He received his Ph.D. in cognitive neuroscience from UCLA, where he focused on using machine learning and neuroimaging to predict the efficacy of treatment regimens.