We've updated our Privacy Policy to make it clearer how we use your personal data. We use cookies to provide you with a better experience. You can read our Cookie Policy here.


Revolutionizing High-Throughput Drug Screening With AI and Organoids

Pipette tips transferring liquid to a multi-well plate.
Credit: iStock
Listen with
Register for free to listen to this article
Thank you. Listen to this article using the player above.

Want to listen to this article for FREE?

Complete the form below to unlock access to ALL audio articles.

Read time: 3 minutes

High-throughput drug screening evolved in the 1990s with the invention of multi-well plates containing up to 1,536 wells, which allowed researchers to screen multi-million-compound libraries quickly. The idea was to throw every compound in a library at an assay in the hopes of finding a few active compounds. It often worked, but the process was unwieldy and did not always yield clinically relevant results.


Now, thanks to the rise of artificial intelligence (AI), high-throughput screening (HTS) for drug discovery is taking a more intelligent, computational approach, with tools that allow scientists to quickly narrow down huge libraries of drug compounds to smaller, more focused compound libraries that will allow them to pinpoint the most promising drug candidates faster. This evolution of HTS will be further enhanced by a shift from 2D to 3D biological models such as organoids, which are miniature organs that can be derived from stem cells, primary cells or cell lines, and that reflect human biology more accurately than 2D models or lab animals can.


What AI can contribute to HTS now

Research tools are emerging that combine AI with automated tasks to streamline the development of cell models for HTS, as well as the screening itself. For example, some tools combine liquid handling, incubation and cell culture with cameras and microscopes to capture images throughout the development of the cell model and then during the drug-screening process.


The AI algorithms that drive these processes must be trained, of course, and that’s where scientists play an important role. They teach the machines how to read and interpret images at the individual well level, and how to make decisions based on those findings. But then they can step out of the process and let automation take over. With AI-enabled image analysis, scientists can quantify phenotypical changes in cells using a combination of unsupervised and supervised machine learning. So, for example, the tool can capture images of two cell types and, based on hundreds of different parameters, it can tell the scientist what the key differences are between them that are relevant to the experiment at hand. The beauty of automation is that it improves consistency and reproducibility in decision-making, as well as in the experiments themselves.

Now there’s an increasing demand for 3D models such as organoids to be used for HTS, because their capacity to mirror the complexity of human biology increases the likelihood that the drug candidates that emerge from screening assays succeed in clinical development.

But scaling organoids up has been challenging. In fact, it’s difficult to house organoids in plates that are any larger than 96 wells which limits the number of compounds that can be screened simultaneously.


How AI can spur the transition from 2D to 3D models


AI helps organoid developers improve the design of models and scale them up so that a number of candidate drugs can be screened against them. For example, to prompt the differentiation of induced pluripotent stem cells (iPSCs) into specific cell types, organoid developers typically experiment with different combinations of growth factors, media and cell-culture conditions. AI can determine which combination in a matrix format is ideal for driving the iPSCs into differentiated states quickly and precisely.


Then, when it is time to scale up the organoids for HTS, AI-enabled tools can accelerate the process by replacing manual tasks with automation. So instead of scientists looking at each well under a microscope to check that the organoids are developing correctly, they can train AI to distinguish between good and bad wells, even down to the individual organoid level to identify which to use for data gathering. That removes the risk of variability that comes from individual scientists making those assessments – and it accelerates the process of scaling up and evaluating the most predictable organoids. Armed with AI, these technologies can compute what the human brain simply cannot when using multiparametric complex cellular models at the quantities required for HTS.


What’s more, AI automates routine procedures such as feeding and passaging of cell cultures, using imaging and AI to determine when cells need to be fed or passaged, eliminating the need for scientists to be in the lab 24/7 performing hands-on tasks that are not only time-consuming and mundane, but also prone to human error.


Remaining challenges


Generating the appropriate biomass for large-scale drug screening – meaning the number of organoids needed to run 3D drug assays – remains a challenge that AI is starting to help solve.


The next step will be to improve the 3D models themselves. There are organoids now that mirror the human digestive system, brain and heart, but what the industry really needs to further advance drug discovery are “sub-organoids” that model specific regions of organs or even disease states. Researchers are developing organoids of the large and small intestine, different chambers of the heart (e.g., ventricles), different regions of the brain, and even tumors derived from individual cancer patients. AI is already making this possible.


And if we employ AI to the best of its abilities, researchers will soon be able to perform HTS on organoids that accurately reflect the diversity of human biology and disease, uncovering better therapeutic candidates faster than ever before.

About the author:

Daniel DiSepio, PhD, is an automation and customization commercial manager at Molecular Devices.