Artificial intelligence (AI) has been steadily making inroads into every phase of drug discovery over the last decade. While AI may not have solved every problem in drug discovery, it's undoubtedly facilitating and transforming drug discovery to a new height. 

A notable advancement was the incredible success of AI systems for protein structure prediction. DeepMind's AlphaFold program was the first to crack one of the grand challenges in biology, called the "protein folding" problem. The ability to correctly predict the three-dimensional structures of proteins is ludicrously difficult and pertinent for understanding biological processes and elucidating how drug molecules bind to their target protein. The other significant breakthrough has been the success of AI-driven generative design platforms in delivering clinical candidates at accelerated timelines. Exscientia, a UK-based AI-driven PharmTech company, was the first to announce the discovery of an AI-designed molecule to enter clinical trials. Currently there are more than 15 AI-designed clinical stage assets, thus validating the transformative potential of AI-driven generative platforms in the making drug discovery faster, cheaper, and better. 

As 2022 draws to a close, it's no doubt the future of AI-driven drug discovery is rosy and will continue to progress as we enter 2023. Accordingly, the top five developments to watch out for in the next twelve months, in my opinion, are

Next-gen AI-driven protein structure prediction tools

Large language models that can understand the language of chemistry and biology will be the next revolution in AI-driven protein structure prediction. Unlike methods like AlphaFold, which rely on evolutionary information for structure prediction, protein language models require only the protein's primary sequence as its input.

HelixonBio, a Bejing-based AI-driven pharmaceutical company, was the first to announce the development of a protein language model-based structure prediction program called OmegaFold. In addition, Meta AI researchers have developed a language model-based algorithm called ESMFold. Language models would advance AI-driven protein structure prediction by improving accuracy and significantly reducing runtimes.

3D Generative AI for better and faster innovation

Drug design is engineering at the molecular level. Generative modelling designs novel, synthetically tractable, and optimized molecules via AI-driven generative modelling. The coming year will be crucial for companies built on the AI-first model as the much-anticipated clinical study readouts for the AI-designed drugs will be out in 2023.

The current panoply of AI-based generative modelling methods employs string-based or graph-based for molecule design. These methods cannot perceive the 3D structure or encode the 3D structure of the protein binding pocket. However, molecules are ultimately three-dimensional objects, and their shape and electrostatics play a crucial role in molecular recognition. Thus, 3D AI-driven generative modelling that generates molecules in the target binding pocket will be the next big thing.

Towards Explainable AI (XAI)

One of the fallbacks of AI-based approaches, and deep-learning methods, is the black-box nature of the model, which offers little or no real insight into the decision-making process. The key to unlocking this explainability crisis is using Explainable AI, abbreviated as XAI. 2023 will witness the growth of tools that make model predictions transparent, interpretable, and explainable. While no single approach works the best, methods like self-explaining neural networks could emerge to resolve the explainability crisis. It is essential to instil confidence and foster acceptance of AI for compound design and property predictions.

AI-powered target identification to take center stage

Choosing the right target is key to creating success in drug discovery. Using AI and predictive analytics will provide a better way to identify and prioritize drug targets with an increased likelihood of success. AI-powered target discovery platforms can sift through the large volume of complex, disparate multi-omics data and draw meaningful insights using knowledge graphs. Pharma majors will leverage AI-powered target identification in a big way in 2023, and more AI-identified targets will be entering the portfolio. 

AI-enable biomarker discovery the next frontier 

Translating preclinical discoveries into clinical practice is one of the biggest challenges in clinical development. Hence, translational strategies are increasingly being integrated at an early stage into drug discovery programs to de-risk clinical development. To that end, artificial intelligence-based multi-omics data analysis will power a new era of biomarker discovery for driving drug discovery. 

Disclaimer: 

The views and opinions expressed in this article are those of the author and don't necessarily reflect the opinion of the affiliated institution nor should not be construed as investment advice.

Photo by Pawel Czerwinski on Unsplash