Pharmaceutical research

_{Web special Fraunhofer magazine 1.2025}

Prof. Carsten Claussen has been on the hunt for more than 25 years now. “The one drug that makes it to your local pharmacy is like the proverbial needle in a haystack,” explains Claussen, head of the Hamburg office of the Fraunhofer Institute for Translational Medicine and Pharmacology ITMP. He almost made it once, he recalls. But then the drug, a candidate for treating Alzheimer’s disease, turned out to have no effect in humans. “My wife was thinking we were about to strike it rich. I was sorry to disappoint her,” he says. “I ended up having to stay at Fraunhofer − which is great, actually Claussen and his team put vast stamina and expertise into combing through libraries of substances, often comprising hundreds of thousands of molecules or even more, searching for a “hit.”

That is what experts call a promising candidate that can address a pre-identified therapeutic “target” in the body and, for example, block the growth of tumor cells. The team performs “high-throughput” screenings that automatically test the molecules to see whether they interact with the target, and if so, how. “We don’t just test any random molecules, of course. We try to select the ones with the greatest possible relevance in advance,” Claussen explains. To do that, the researchers use existing data on the chemical structures and biological properties of the various molecules, such as metabolism or protein interaction, and then train AI models that help to predict the ideal structures of the active substances. “This really helps us narrow down the group of potential substances. The search is faster, more concrete and just smarter. And the chances of a hit are twice as high.”

The team’s most recent success was the discovery of a compound that is effective at treating epilepsy in children. They found it in a “repurposing” library: a collection of substances already approved for a particular medical indication. Claussen explains: “After all, it’s not much of a stretch to think that a drug might have useful effects elsewhere in the body as well.” The benefit to this approach is that it reduces the time and costs involved in development because many tests are no longer needed and some phases of development can be skipped. It is also possible to rule out risks that might be associated with previously unknown compounds.

Only one in 10,000 substances makes the grade

Not every hit ends up being suitable for use as a drug, though. The substance could be toxic, break down quickly in the blood serum or cause dangerous side effects. This is determined by performing extensive tests on cell cultures, in animal models and later − if all goes well − in humans. “It takes about five to seven years to reach the point of administering the drug to the first human subjects. The failure rates along the way are incredibly high,” Claussen explains.

According to the German Association of Research-Based Pharmaceutical Companies (Verband Forschender Arzneimittelhersteller, vfa), out of every 10,000 substances with potential effects, about nine reach human trials and only one will eventually wind up on the market. This makes it no surprise that developing a new medication is costly and takes a long time. The average total costs come to about 2.8 billion U.S. dollars today, and it takes 12 years to go from the initial idea to approval − the highest figure so far in a trend that has continued unabated since the 1950s.

One key reason is that drugs are becoming more and more complex. The first targets for drug development were health issues involving simple mechanisms or widespread symptoms, like headache and heartburn, but these days, researchers are looking to address things like cancer or rheumatic diseases that affect the entire body. These kinds of diseases affect a large number of biological processes in the body and can only be treated effectively through combination therapies involving multiple drugs. The mechanisms of disease are multifaceted, and in many cases are not yet fully understood, which makes it even harder to identify targets. “AI can really help with this as well,” Claussen says. This is because AI makes it possible to take numerous parameters that affect a given disease into account, combine them and identify key targets.

Once a promising drug candidate has been identified, it is tested for efficacy and safety. These days, it is possible to sidestep the cost, effort and lengthy approval times involved in animal testing in many cases. Dr. Julia Neubauer, managing director at the Fraunhofer Project Center for Stem Cell Process Engineering, and her colleague Prof. Florian Groeber-Becker, head of the Fraunhofer Translational Center for Regenerative Therapies TLC-RT, are working together in Würzburg on innovative cell-based tissue models for testing active ingredients. In addition to primary cells taken from various tissues such as the skin or the eye, they also use induced pluripotent stem (iPS) cells: artificially created stem cells that can be used to grow various cell types. Among other advantages, these cells are consistent and reproducible, while primary cells can vary. Some primary cells, such as those of the heart muscle or neuronal cells, are also difficult to isolate and culture.

Another major advantage to these tissue models is that they can simulate specific illnesses or disease mechanisms. “The animals used in testing aren’t sick at first. That means you have to induce the pathology before you can perform the tests,” Groeber-Becker explains. By contrast tissue models not only come with no ethical concerns but also supply significantly better results. “It’s almost like you’re taking the human with the disease and putting them in the Petri dish,” Neubauer says.

Model hearts contract like the real thing

It takes three to six weeks for human test models to be ready to use. “For cardiomyocytes, or heart muscle cells, we have little clusters that beat and contract just like tiny hearts after only seven days,” Neubauer explains. They mature after that and are cultured under specific conditions to produce three-dimensional tissue, forming what are known as organoids. These model hearts, about the size of the head of a pin, can be used to test aspects such as how innovative cardiac medications affect the heart’s contractile force or rhythm. Different active ingredients can also be tested in series or in combination on the same model with no problems. “With our skin models, which are about the size of a fingernail, you can even use a cotton swab to massage in a formulation like a cream,” Groeber-Becker says.

The models can be used for about one to two weeks. After that, the cells grow too old and gradually lose their functionality. Animal testing still cannot be fully eliminated from drug development, Groeber-Becker explains. “But we’re working on making the occasions when it is needed rarer and rarer.”

There is a lot of interest in the tissue models and organoids, she notes. This is because they have one crucial added advantage: “They save considerable time and cost because our human models increase predictability. This allows us to predict with much greater accuracy how the drug will behave in the human body and whether clinical trials would be worthwhile.”

But even with bright prospects, only some of the most promising drug candidates ultimately end up treating patients. One of the biggest hurdles in drug research comes during these later stages of the process: recruiting participants for clinical trials.

So far, subjects are mainly sought manually. Healthcare professionals review their patient bases and compare them against the relevant trial’s list of requirements − a laborious and time-consuming process that often does not produce the desired results. At least 20 percent of all clinical trials fail because not enough candidates can be found.

A new AI-based solution developed at the Fraunhofer Institute for Intelligent Analysis and Information Systems IAIS in Sankt Augustin, near Bonn, is helping with this. It allows relevant information to be simply extracted from running text, such as medical histories or doctor’s notes, and then compared to the inclusion and exclusion criteria for clinical trials. This makes it possible to scan a hospital or clinic’s entire database in seconds, for example. It does not even matter which file format was used to store the information. “Advanced AI models can deal with multimodal data, including pictures of text or tables. We’re working to integrate these kinds of models into our solutions,” explains Sina Mackay, a data scientist at Fraunhofer IAIS. The process of comparing records against the requirements listed for current trials, which are published on centralized websites such as the EU Clinical Trials Register, could take place automatically.

App makes it faster to find study participants

But the search is not a one-sided one. Even as the drug industry looks for subjects for its trials, many patients are also on the lookout for studies that might work for them. Motivations include gaining access to innovative treatments or simply supporting medical research to help others with similar conditions. To meet this need, the Fraunhofer IAIS team worked with partners to create a prototype version of an app that could compare personal data from a person’s electronic patient record to the trial requirements published online. “The likelihood of a match increases if both parties are actively looking, by which I mean both the drug developer and patients,” Mackay explains. This could make the process of recruiting participants for clinical studies much faster, more efficient and more successful.

But before the app, called DATACARE, can access people’s electronic patient records, there are still some points to clear up, including legal questions and the issue of ensuring privacy. “Aside from that, of course, all the relevant documents should be present in the electronic patient record, and at this point, we can’t yet assume that to be the case,” Mackay says. Even so, she is confident: “We’re on our way.”