Klick Health

Generating drug brand names with neural networks

VP Strategy

Read More

Creating new drug brand names is hard. Apart from the general challenge of branding, there’s the added challenge of needing a name that physicians, patients and pharmacists won’t confuse with existing names—and that won’t imply claims not supported by a drug’s label. But we live in the age of artificial intelligence. Could it help?

It’s a long, laborious, and expensive process to get to an approvable drug name. Drug branding agencies typically generate up to 5,000 names to get to one that’s approvable. And the FDA rejects about 20% to 35% of those that are proposed. 

Inspired by the work of research scientist and neural network hobbyist Janelle Shane (who’s used neural networks to generate names for everything from guinea pigs to craft beers), I wanted to find out if AI could make this process any easier.  So I took a stab at training a neural network on drug brand names and letting it loose to create its own.

Have you taken your Lutana today?

Here are some of my favorites, in alphabetical order, from a list of 100 generated in less than five seconds:

How it works

The names were generated through the use of a recurrent neural network (more specifically: LSTM using Keras with Tensorflow). This type of neural network can learn sequences of characters, and then predict the probability of a character following other characters. For example, when it sees “queen,” “quart,” “question,” and other words that start with “q,” it learns that the letter “u” tends to follow the letter “q.” If you then provide it with the letter “q” and ask it to predict the next letter, it would give you “u” with high probability. By doing this repeatedly, it can generate words and sentences. For the list above, I provided it with a list of existing drug brand names, from which it learned generalities about their character sequences.

Some future ideas to explore

The work to generate this list is a proof of concept, with many limitations. But I hope you agree that it demonstrates the power of neural networks to automate many aspects of the drug commercialization process.

Here are two things I would consider to build upon this work in future:

So there you have it, the ability to generate hundreds or thousands of drug brand names instantaneously, and a path towards improving the results over time.

Your creative Technologists

Is your brand adapting to the constantly changing market place, and taking full advantage of the latest in emerging tech to remain competitive? Here at Klick Health we’re actively experimenting with and applying machine learning and artificial intelligence to healthcare. Want to learn more? We can’t wait to learn together.

More About the Author

Simon Smith

Simon is an original dot-com veteran who's passionate about creating scalable solutions to meaningful problems. A health and technology junky, he's known to read ISIs and program apps in his spare time, when he's not planning strategies to drive our clients' digital success.

More from this author

Go from news to action. Klick Wire

Weekly Digital Health Newsletter

Klick Health will NEVER spam you. Read our privacy policy

Thank you! You're now signed up to get the Klick Wire every week - news from the world of digital health marketing.

Sorry there seems to be a problem. Please try again later.