Inflammatory bowel disease (IBD), which afflicts over 1.6 million Americans, may soon have a new treatment option thanks to artificial intelligence. This past week, Insilico Medicine announced that its AI-designed drug for IBD, called ISM5411, has entered Phase I clinical trials.
If approved, this would represent a significant milestone as the first IBD medication specifically designed to promote mucosal repair and restore gut barrier function. But ISM5411 is not only noteworthy for how it works, but how it was developed: using a novel AI system called Chemistry42. This raises intriguing questions about the promise and potential risks of this new era of computer-generated drug discovery.
The Need for New IBD Treatments
IBD refers to chronic inflammatory conditions like Crohn’s disease and ulcerative colitis that damage the gastrointestinal tract. Flare-ups trigger debilitating symptoms including severe diarrhea, abdominal pain, rectal bleeding, weight loss, and fatigue.
While the exact causes remain unclear, IBD is believed to involve a dysfunctional immune response and disruption of the mucosal barrier that lines the intestines. There are limited treatment options, and no cure. Anti-inflammatory drugs and immunosuppressants can help control symptoms, but come with significant side effects given their systemic effects.
Patients often struggle to find an effective treatment regimen, resulting in poor long-term prognosis. Hospitalizations and surgery are common. The incidence of IBD has risen sharply in recent decades, especially in industrialized nations, for reasons researchers don’t fully understand. This has created an urgent need for newer, safer therapies.
A Novel Mechanism to Heal the Gut
Insilico Medicine, a pioneer in AI-driven drug discovery based in New York and Hong Kong, aimed to address this unmet need through an innovative approach. Their R&D team utilized Chemistry42, Insilico’s proprietary AI platform, to design a novel small molecule drug to block a target called prolyl hydroxylase domain (PHD).
Studies have found IBD patients have elevated levels of PHD, which regulates genes involved in gut barrier protection. Insilico hypothesized that inhibiting PHD could help heal and reinforce the intestinal lining in IBD patients.
The AI system generated a pipeline of candidate molecules tailored for optimal PHD binding and drug-like properties. After selecting the most promising drug, dubbed ISM5411, Insilico launched clinical studies to assess safety and efficacy in humans.
AI-Driven Drug Discovery
This drug development process exemplifies the power of AI techniques like deep learning and reinforcement learning. Insilico’s CEO, Alex Zhavoronkov, likens Chemistry42 to a molecular generative AI similar to systems like DALL-E 2 and ChatGPT that create images and text. But instead of generating art or prose, Chemistry42 produces novel drug-like compounds.
The AI is trained on datasets of biochemical structures, physicochemical properties, synthesis pathways, and other information predictive of pharmacology. It uses this knowledge to design virtual molecules satisfying specified design criteria.
For ISM5411, the criteria included oral bioavailability, intestinal restriction, safety, and potency against PHD. Chemistry42 outputs a list of proposed structures ranked by predicted efficacy. Insilico’s chemists then synthesize the top candidates for real-world testing.
This AI-enabled drug invention pipeline aims to shortcut the traditionally lengthy and costly process of drug discovery. It also increases the diversity of molecular structures examined, since the AI has no preconceptions about what compounds “should” work.
Promise and Unknowns of AI Drug Discovery
The ability to rapidly design promising new drug candidates with little human bias offers great promise, especially for historically challenging diseases like IBD. However, ISM5411 also represents uncharted territory as one of the first AI-invented drugs to reach human testing.
While its Phase I trial will assess short-term safety in healthy subjects, questions remain around long-term effects and efficacy. There are inherent uncertainties with any new drug platform, but perhaps especially one that derives from a black-box algorithmic system versus human intuition.
Validating the AI’s output will require extensive clinical data, which Insilico aims to gather through wide-scale Phase II/III studies if Phase I results look favorable. But early success would help instill confidence in Chemistry42, potentially accelerating future drug discovery efforts.
Rethinking the Drug Development Paradigm
If ISM5411 succeeds, it may transform how new medicines are created. AI-driven approaches could better explore chemical spaces ignored by traditional drug discovery constrained by human cognitive biases. This expands the breadth of possible treatments.
At the same time, AI drug discovery faces deep, unresolved questions. How do we ensure algorithms are trained on representative, unbiased data? How will regulatory agencies evaluate AI-designed drugs where the reasoning behind the molecule’s design is opaque? Are there unique ethical issues surrounding AI’s role in medicine development?
While the answers remain unclear, what is certain is that IBD afflicts millions desperate for better therapies. If ISM5411 can safely deliver relief by tapping into AI’s potential, it may herald a new era in rationally designed, data-driven medicines. For now, the medical community awaits with cautious optimism to see if this pioneering drug lives up to its promise.
