Using AI to accelerate drug design and synthesis planning

Using AI to Accelerate Drug Design and Synthesis Planning

Virtually no industry has been left untouched by the influence of AI and machine learning, and drug development is no exception. Just a few years ago, pharma professionals were theorizing that AI might be able to help with some aspects of drug discovery and development, and now it seems these powerful technologies are transforming the process right before our eyes.

“Machine learning techniques are poised to impact pharmaceutical development from beginning to end,” said Dr. Eva Nittinger, associate principal scientist in computational chemistry in Respiratory and Immunology at AstraZeneca R&D, in a recent Elsevier-hosted webinar titled Drug Discovery with AI at AstraZeneca – from Generative Models to Reaction Prediction . The webinar is part of a series of AI-focused webinars called “Dear data: what compound to make next? – From SAR databases to AI/ML models.”

In her presentation, which offered great insights into how drug researchers can use AI and ML techniques in their work, Dr. Nittinger discussed drug design, including the design-make-test-analyze cycle (DMTA).

“It starts with a protein of interest, which can be a protein that is implicated in the disease. Ideally at the beginning we have a chemical starting point, which has a non-optimal dose, so it is weakly active, target unselective, has toxicity risk and so on. And what we want to achieve is a candidate drug with an optimal dose of which is highly potent, effective in in vivo models, metabolically stable, and no toxicity issues,” she explained.

“The DMTA cycle is an iterative approach so we have to repeat it multiple times, always taking into consideration what we learned from the previous cycle. However, this whole process takes, on average, more than three years. So the question is, how can we actually improve and accelerate this process, from hit to candidate drug?”

Dr. Nittinger pointed out that there are a few different opportunities throughout the DMTA cycle to accelerate the process, including multiple parts where automation, robotics and microfluids can be employed. But in this particular presentation the focus was on the use of molecular ideation tools to come up with new design ideas, and AI tools that can be used later on during the make phase with synthesis prediction.

One such design tool is called REINVENT, which is a de novo molecular design tool developed in-house by AstraZeneca. (You can learn more about how to access REINVENT in the download materials for the webinar .) “How it works is that we have a generative model which proposes new chemical structures,” said Dr. Nittinger. “And we have a scoring function, which can be very project-specific, which evaluates the proposed molecules, and thus gives feedback to the generative model, which in turn adapts and generates different molecules in the next phase.”

“Traditional approaches usually have the drawback that they rely on searching a large database for a relatively small number of suitable hits, while with a generative model we can practically encode the chemical space and thus have many hits with ideally few evaluations.”

In the webinar , Dr. Nittinger goes into greater detail about how REINVENT works, discussing molecular ideation and compound prioritization, and explaining some of the ways that they meet the challenge of how to filter and select – with a focus on the use of QSAR (qualified structure activity relationships) models. (Click here to watch the full webinar.)

Her colleague Dr. Samuel Genheden, who leads the Deep Chemistry team in Discovery Sciences at AstraZeneca R&D, followed with a peek into how AI can also help after the design process.

“We have generated a molecule that we believe in, now it’s time to see how we can synthesize that in the lab,” he said in his introduction to computer-aided synthesis planning. “I will start with the view that we have a molecule from, for instance, a REINVENT experiment, and what goes on in the mind of a chemist when they want to use the synthesis planning in order to make it. Typically, they start with a retrosynthesis analysis. This is a computational technique where you find how to make it step-by-step by breaking it down into smaller and smaller pieces until you find something that you can purchase from a vendor or grab from the shelf at the lab.”

In his presentation, Dr. Genheden details more of this process, and also talks about different types of software that researchers can use to accelerate this step. To aid in the synthesis planning, there are a few available solutions, which he outlines. Among these solutions are AstraZeneca’s own AiZynthFinder, which he showcases more closely in the presentation, as well as Elsevier’s own commercial solution, Reaxys , which has millions of reactions for machine learning sourced from journal articles.

To get a complete look at both presentations, watch the full webinar online for free. Also be sure to check out the first webinar in our “Dear data” series, Connectivity between documents, structures and bioactivity data .

#AI #DrugDesign #Pharmaceuticals

?