How biotechs are using AI to rescue failed drugs

Despite the time and money invested by biotech and pharma companies in the drug development process, statistics show that 90% of drugs fail in clinical trials. This is usually because they do not meet their endpoints or their safety is called into question, and they are subsequently shelved by their developers. But certain companies see an opportunity here – why not try to rescue some of these failed drugs that still show potential using the help of artificial intelligence (AI)? 

In this article, we take a look at how AI allows companies like Ignota Labs and BioXcel Therapeutics to rescue drugs that have previously failed in clinical trials due to either safety or efficacy reasons, and we explore why salvaging these drugs is a prudent approach to drug development. 

The role of AI in the rescue of failed drugs 

As we all know by now, AI has already had an incredible impact on drug discovery and development, with more and more biopharma companies adopting the technology to speed up and streamline their development processes. 

It turns out that AI is also extremely adept at allowing companies to repurpose drugs for indications for which the drug in question is likely to have a high success rate, as the technology is very efficient at rapidly sifting through vast amounts of biomedical data to uncover hidden drug-disease connections and predict which existing drugs may be effective for a new therapeutic use. 

Different AI approaches can be used during this process. For example, machine learning can be applied by companies to predict how strongly a drug will interact with potential new targets, while deep learning can be applied to generate novel molecular structures optimized for a particular disease.

The integration of AI in the drug repurposing process now allows more companies to focus on trying to rescue failed drugs, where before they might not have felt able to attempt this. “Without the help of AI, it is essentially the status quo, where it is easier to simply throw away a failing drug than to fix it,” said Jordan Lane, chief scientific officer (CSO) and co-founder of Ignota Labs. “Addressing a toxicity issue could entail wet lab testing of an unfathomable number of mechanisms and endpoints. The standard safety tests would have previously been tested and come up short, so it is fishing in the dark, with very long and expensive timeframes at every turn.”

Ignota Labs: Addressing toxicity issues to rescue failed drugs with AI platform

The reason Lane pointed specifically to the ability to address the safety of drugs is because that is exactly what his company Ignota Labs is doing; it selects failing drugs to work with, solves their toxicity issues, and reenters them into clinical trials so that they can be approved and delivered to patients who need them.

Drug safety failures are one of the leading causes of clinical trial failures. According to Lane, by setting up the company, he and his co-founders wanted to address this problem. Initially, they tried to encourage customers to use AI to proactively derisk drug development, but it was difficult to get biotech companies to pay them to be told that their drug was going to be toxic.

Now, the company actively acquires distressed assets to add to its own development pipeline so that it can use its AI technology to address the root cause of the drugs’ toxicities and develop solutions that can take them successfully through the clinic the second time around.  

“Failures typically occur in complex systems such as a human,” explained Lane. “Finding the mechanism of failure allows us to recreate the toxicity in a much more basic lab test. This means we can demonstrate that our new chemistry will be safe when re-entering the clinic. Our tech stack of predictive tools also comes into play to ensure that we haven’t fixed one issue but created another.”

Ignota’s AI technology is called SAFEPATH. It uses deep learning that explains both why and how safety issues occur through a balance of cheminformatics and bioinformatics, delivering actionable insights to refine or repurpose drug candidates. Lane said that, historically, many AI companies have only focused on either cheminformatics or bioinformatics, but Ignota believes that “it is the intersection of these two techniques that will unlock true insights into the drugs that have hit issues.”

He continued to explain just how the technology works: “Our cheminformatics platform covers machine learning models from as much as the proteome as possible. We do this at different drug concentrations and across different species, ultimately creating over 15,000 models. When we have a new compound, we can recapitulate the on and (more importantly) off-target binding partners of the drug.” 

This information is then used as input to Ignota’s bioinformatics tech stack, which maps across the different pathways involved for that target and is organized in a causal knowledge graph to facilitate traversing disparate datasets across a host of different omics databases.  

“When specifically constraining the biology to a known problem, the goal is to link the compound structure to the endpoint. The mix of cheminformatics, bioinformatics, and wet-lab validation allows us to build a compelling hypothesis that gives us the belief that the problem is solvable, resulting in us in-licensing the project, developing the chemistry and bringing the drug to the clinic (or back to the clinic) as quickly as possible,” added Lane. 

Ignota recently raised $6.9 million in seed funding, which will be used to expand its pipeline and help advance its first asset, a PDE9A inhibitor, into early-stage clinical trials. Lane also told Labiotech that the company is in late-stage due diligence for a handful of “incredibly exciting” drugs that Ignota believes could be transformative if their safety issues can be overcome. “Not all of them will make it through our stringent process, but we are confident to add a roster of drugs to our pipeline very soon, and we are actively acquiring assets.”

BioXcel Therapeutics: An AI-driven drug re-innovation mission that includes repurposing shelved drugs  

Another company involved in using AI to repurpose failed drugs – as well as drugs that have already been approved – is BioXcel Therapeutics. Unlike Ignota, however, the company focuses on compounds that have already demonstrated safety in prior clinical trials, but have been discontinued by their developers for various other reasons. 

The company has a particular focus on repurposing drugs for neuroscience. Developing drugs for central nervous system (CNS) indications generally takes longer and has historically seen lower success rates than other therapeutic areas due to the complexity of the brain, difficulty in target identification, and challenges in translating research into effective treatments. 

BioXcel believes that one of the solutions for this is to repurpose already existing drugs that show promise for CNS disorders, as plenty of compounds are on the shelf that can be reexamined and redesigned for another indication. In this case, the development process will not take as long as creating an entirely new drug from scratch. As speed is of the essence here, the company applies its AI platform technology, called NovareAI, which enables it to reduce therapeutic development costs and potentially accelerate timelines, while improving the success rate of bringing new treatment options to patients.

Friso Postma, vice president (VP) of Artificial Intelligence Drug Discovery at BioXcel Therapeutics, told Inside Precision Medicine that he prefers to refer to the company’s platform as augmented intelligence rather than artificial intelligence, as he insists it is not a black box, but a composite of AI tools that requires immense human supervision. 

NovareAI works every hour of every day trying to detect failed phase two and three assets it can improve upon and re-innovate for CNS indications. The platform sorts through all available literature to pull out information connecting compounds, neural circuits, behaviors, and indications. To achieve this, it uses “knowledge graphs” to mine the literature, organizing data from various sources and capturing information about entities that can connect all of these factors. Knowledge graphs include structured information that AI systems can use for various functions, including information retrieval, recommendation systems, and question-and-answering. 

Postma explained to Inside Precision Medicine that although some of the information that is fed into the knowledge graph is well-structured, there is also a lot of unstructured data, which requires a whole layer of tools like natural language processes and large language models to make sense of it all.

BioXcel is currently working on the development of a new anti-stress drug called BXCL502, which it is studying as a feasible way to treat chronic agitation in dementia, and another new drug called BXCL503, which could potentially treat dementia-related apathy. 

BioXcel has also already received approval for a sublingual film version of the non-opioid pain drug dexmedetomidine, which it repurposed to treat agitation associated with schizophrenia or bipolar disorder. This is thanks to the fact that the drug was found to have qualities that are sympatholytic (meaning that it blocks nerves from the sympathetic nervous system connected to “fight or flight” reactions), which can treat anxiety, such as generalized anxiety disorder, panic disorder, and post-traumatic stress disorder (PTSD).

“With IGALMI, we went from first-in-human to approval in less than four years – that’s quite dramatically different from what you would see with a [new chemical entity],” said Postma in the Inside Precision Medical article. “You’d have to go through each selection, which takes 10–12 years and costs an average of $1.2 billion. There’s a tremendous upside here.”

A prudent move: The advantages of rescuing failed drugs 

When a drug fails to reach the end of its journey, patients are ultimately the ones who end up missing out. 

“So many drugs that patients are waiting for fail during the preclinical and clinical trials process, which is devastating for them and delays getting patients the treatment options they need,” expressed Lane. “Bringing the drug back to the clinic means that we recapture the hope that would have otherwise been lost.”

Plus, for companies developing a new drug, the process from discovery to market generally takes 10 to 15 years and costs, on average, around $1.1 billion. If the drug then fails, that is potentially 10 to 15 years of wasted time and $1.1 billion of wasted money. 

But repurposing failed drugs can at least offer some reprieve. “Our approach means that we are also able to recoup the otherwise lost investment spent on the drug, the years of research and people’s time, and the water, carbon, plastic, and animals used to get to their failure point. For us, it facilitates building a late-stage pipeline that otherwise would take a decade of work to build.”

Ultimately, using AI to rescue failed drugs has the potential to help get better treatments to patients faster and more affordably than ever before.