What Impact Could Generative AI Have on Drug Discovery? Part I

Published in
6 min readOct 24


Image courtesy of Aqemia

By Florian Denis, with assistance from Marc Rougier, Louisa Mesnard and Anya Brochier

The sonic boom caused by the eruption of generative AI (GenAI) into our vocabulary has asked us to imagine the impact of this new technology on the world of healthcare and specifically on drug discovery. While most current projects are in the early stages of development, the combination of GenAI and drug discovery could not only lead to the creation of new treatments, but also new advances that nature itself is not yet able to offer.

Part I: The beginnings of this intertwined approach

AI has existed, often used as a “buzzword” across sectors, for the last decade, leading to significant advancements in technology and operational efficiencies. As the impact of AI begins to be felt across sectors, the intersection of AI with Healthcare is already here. Even so, the most advanced AI-driven medicine companies only have a portfolio of drug candidates in Phase 2 clinical trials. What does this mean? The next few years will be key in unlocking the power of this combination to create potent new cures for deadly diseases.

A BCG report from March 2022 states that , “Biotech companies using an AI-first approach [had] more than 150 small-molecule drugs in discovery and more than 15 already in clinical trials”. With this in mind, it’s not outrageous to imagine that the first drugs with active ingredients co-constructed with algorithms could be on the market in the next few years, even if the excitement surrounding the first molecules created by artificial intelligence has been tempered by recent news from from Exscientia and BenevolentAI.

For researchers, AI represents a cutting-edge assistant capable of analyzing a multitude of data in record time, better than any human brain. Today, drug development still takes on average 10–15 years of R&D with a cost of production (around $2.5 billion) that has increased dramatically over the last decade despite improvements in technology: this is the concept of Eroom’s law (Moore’s law spelled backwards). With AI, it could be possible to go much faster, cutting the cost and time of development and drug GTM by 30%, while reducing the risk of failure (i.e. the attrition rate) — currently at approximately 90% (depending on the therapeutic area).

Gen AI vs “Traditional” AI

While the recent introduction of GenAI into our vernacular could appear to render traditional AI obsolete, when looking at the trajectory of the intersection of AI and healthcare, it’s important to differentiate.

Traditional AI: When referencing traditional AI, we think of Deep Learning (DL), a subset of Machine Learning (itself a subset of AI) that focuses on artificial neural networks (a method that teaches computers to process data like the human brain) and their ability to learn and make decisions. It uses complex algorithms to learn how to recognize specific patterns based on a set of training data. These patterns are then used to classify unseen data, make predictions, create data clusters, or remove unnecessary information or noise.

To visualize how Deep Learning (DL) works, imagine teaching a child to recognize a task by showing an example multiple time. This is how scientists were able to teach machines to do single tasks that normally need human intelligence, such as understanding language, recognizing images, or making decisions. However, it was not possible with these techniques to teach machines to mimic one of human’s most unique characteristics: creativity.

Generative AI: Generative AI (a subset of DL) represents the next step in the evolution of artificial intelligence and is like a magical paintbrush for computers. From DL, scientists recently developed a brand-new category of generative models able to produce their own structured data in the form of texts, images or other media.

Just like traditional AI, GenAI models can learn patterns from (admittedly, much larger) training data sets, but they are also able to infinitely produce new patterns of data. Just like an artist who comes up with new ideas, GenAI comes up with its own unique creations, inspired by things it has seen before.

According to Maximilien Levesque, CEO of Aqemia, “ The concept of Generative AI has been envisioned for a while, but its realization has been propelled by modern hardware, which can process vast amounts of data, such as text, images or audio, enabling the generation of probable answers to problems posed”. Indeed, this amount of data implies thaumaturgical computing power, and there are even whispers that generative AI should enable Moore’s Law (a law establishing a correlation between the evolution of computer computing power and the complexity of computer hardware) to remain in force to ensure the evolution of microprocessor computing power.

To put it simply, traditional AI is great at spotting patterns, whereas generative AI shines in inventing patterns. Traditional AI can scrutinize data and convey its observations, whereas generative AI can take that very data and craft something entirely fresh from it.

From genetic code to the binary language of computing

From a philosophical perspective, while there are many mysteries shared between life and technology, there is a surprising relationship between the genetic code and the realm of computing.

Despite their seemingly disparate natures, the four constituents of the genetic code (A, C, G, and T) find an echo in the binary language of computing, hinging on only two digits (0 and 1).

This seemingly incongruent correlation shows that the four letters of the genetic code can cleverly map to binary digits, suggesting a profound connection between the intricacies of life and the foundational logic of computing.

The AI optimization equation

The intersection between AI and healthcare has accelerated, with recent developments in the proteomic field with protein-folding AI technologies seen by Deepmind’s (acquired by Google/Alphabet in 2014) AlphaFold program.

Before diving in further into the impact of AI on protein folding, here’s a quick download on how proteins are formed and what roles they play in the body.

Key facts to know about proteins:

  • Proteins are the functional building blocks of all life form that can play many important roles in the body: digestive, muscular, immune, metabolism or even transporter functions (see our handy chart above)
  • Proteins are made up of a linear chain of amino acids that wraps around itself to form a unique 3-D structure called “folding”
  • Folding is critical to the function of the protein, as it determines its ability to bind to other molecules and catalyze biochemical reactions
  • When we understand protein folding, we can understand how proteins interact, their role in biological processes, and how these processes can be manipulated to develop incredible new drugs and therapies

How does this relate to Deepmind?

AlphaFold, developed by Deepmind, is an incredibly sophisticated AI computing system based on deep learning (DL) that can dramatically reduce the time it would take to predict the 3-D structure and the folding of a protein. Previously, it would take months or years to isolate a protein’s structure with traditional technologies. With AI, our knowledge of protein structures increased from several hundred thousand structures to more than 200 million in the span of a couple of years.

Deepmind’s protein folding technologies are just one breakthrough in the intersection of AI and healthcare. At Elaia, we are actively engaging with and investing at this crossroads that’s offering so many avenues for innovation. Let’s take an example for our portfolio: Aqemia.

Aqemia: a drug discovery wunderkind

We’ve backed Aqemia, a next-gen pharmatech company generating one of the world’s fastest-growing drug discovery pipelines. Their mission is to quickly design innovative drug candidates for dozens of critical diseases. Why did we invest? Aqemia relies on their unique quantum and statistical mechanics algorithms to design novel drug candidates. The disruptive speed and accuracy of their technology platform allows them to treat and scale drug discovery projects as technology projects.

Stay tuned for Part II of our investigation into the intersection of GenAI and drug discovery, where we will look at how GenAI will accelerate healthcare innovations.