Quris Uses AI to Speed Up Drug Development and Cut Animal Testing
The need for animal testing is unfortunate for the drug discovery process, but there seems to be no good alternative to mice, even though they are not exact human analogues. Quris claims to have created the first real alternative by combining artificial intelligence with “patient-on-a-chip” data, which enables reliable testing and automation at a significantly lower cost, with no mouse required.
The company has raised $9 million in funding to move from a pilot project to production, and a star-studded team of sponsors and advisors is a promising indicator that this approach has fundamental merit.
The core idea is logical: build a better small-scale simulation of the human body and use it to collect data that a machine learning system can easily interpret. Of course, that is easier said than done, but once the researchers came up with it, the company got to work.
The Israeli company’s approach is based on a large Harvard study on the use of so-called “organs-on-chips.” These systems, still relatively new, use a small amount of tissue derived from stem cells (“organoids”) as a testing platform for drugs or treatments, giving a good idea of how, for example, a human liver may react to a combination of certain substances.
Harvard researchers found that by linking several “organ-on-a-chip” systems together (for example, liver, kidney and heart cells), you end up with a surprisingly effective model of the human body. Of course, nothing beats the real thing, but this sequential organoid system, or “patient-on-a-chip,” could become a real alternative to testing on mice. The latter is still the most common way to see how a treatment affects the whole organ system, even though substances that pass mouse trials succeed in human tests only about 10% of the time.
The Quris CEO said that as soon as this study came out, he and his colleagues recognized its potential and began working on what needed to be done in terms of engineering and AI to turn it from an experimental system into a scalable product. This is not just a replacement for mice; it is a relatively inexpensive way to conduct limited testing on humans without humans.
“Say you are a pharmaceutical company. Do you want to wait until you are on the verge of clinical trials to find out whether a molecule that looks good on paper is actually effective? You can make all the genomics discoveries you like, but that won’t help you get past mouse experiments, which fail 90% of the time. This lets you pick the “winning horse” before heading to the races”
Given that drug candidates can cost hundreds of millions of dollars to reach the clinical stage, it is more than worth spending even a small fortune to screen out several options destined to fail. If the method is accurate, the risk is practically zero, and it pays for itself if it avoids even one costly dead end. In other words, the software mindset of “fail early, fail cheap” was never really an option.
Quris’s system uses what the company calls “chip-on-chip” technology: several organoid systems (chips) are arranged in sequence (on another chip), yet are still an order of magnitude smaller and more efficient than modern lab systems. Running a hundred simulated people, as the Harvard researchers did, would require millions of chips, but the Quris system uses less biological material. It can be automated and is supported by a well-trained machine learning model.
Another aspect Quris emphasizes is that its unique dataset will power a unique AI that understands and can help conduct and interpret experiments. The AI is already being trained on existing drugs and some drugs nearing release, learning what signals from various sensors mean for a substance’s safety. This allows effective testing on a small number of chips rather than, say, 500 mice.
The chips themselves are not all the same either. By carefully manipulating stem cells and tissues and selecting them with care, it is possible to test different types of people, different conditions or phenotypes. If a company has a drug that works well but causes side effects in 10% of cases, automated testing of various genetic predispositions or complicating factors can help identify which genetic factors lead to such side effects.
Because the AI knows all of this and catalogs all the data, it should be able to determine fairly well, based on a relatively small number of automated tests, whether a drug is a good candidate for human testing or not. Without AI interpretation, the data would suddenly become a problem for several PhD dissertations. But Bentwich quickly noted that they in no way expect to eliminate the biological side and rely on AI alone.
“This really does seem like a very big opportunity. I had similar ideas in other areas of chemistry, that you can use AI to make these predictions. It will certainly not replace testing, but it narrows the process of possibilities and, in my opinion, it will significantly speed up the task”
The business model has two options. One is offering pharmaceutical companies a screening service for their drug candidates, with payment depending on the accuracy of the results, for example, a drug approved by the system reaches a specified testing stage as predicted. The other is to work on its own drugs. Right now, the company has a drug for fragile X syndrome, which will enter clinical trials in 2022.
“This will stop being just expensive experimentation for pharmaceutical companies. In 5-10 years, hundreds of millions of people could well be doing this. When you think about it, we live quite barbarically now. You go to a pharmacist, and they list possible side effects, but you don’t know for sure. Are you a guinea pig? The answer: yes, we are all guinea pigs. But this is only the first step”
