AI’s Role in Solving Rare Disease Treatment Shortages
▼ Summary
– AI is being positioned as a critical force multiplier to address a shortage of skilled labor and tackle thousands of untreated rare diseases in the pharmaceutical industry.
– Insilico Medicine is developing multi-modal AI to automate drug discovery tasks, aiming to generate therapeutic candidates and repurpose drugs with greater speed and lower cost.
– GenEditBio uses AI to design precise, tissue-specific delivery vehicles for in-vivo gene editing, aiming to make CRISPR therapies more like affordable, off-the-shelf drugs.
– A major challenge for AI in biotech is the need for more high-quality, globally balanced patient data to train accurate models and understand human biology.
– Future ambitions include building digital twins for virtual clinical trials to increase drug approval rates and enable more personalized treatment options.
The biotechnology sector possesses remarkable capabilities for genetic modification and drug design, yet a vast number of rare diseases still lack any viable treatment. A critical bottleneck has long been the scarcity of specialized scientific talent required to advance research. Artificial intelligence is now emerging as a transformative force, enabling researchers to address therapeutic challenges that have historically been neglected due to resource constraints. Industry leaders argue that AI acts as a powerful multiplier, dramatically expanding what small teams can accomplish.
At a recent industry conference, Alex Aliper, CEO of Insilico Medicine, detailed his company’s pursuit of what he terms “pharmaceutical superintelligence.” The firm has introduced an initiative called “MMAI Gym,” designed to train general-purpose large language models to match the performance of specialized AI in drug discovery. The objective is a versatile, multi-modal system capable of handling numerous discovery tasks at once with exceptional precision. Aliper emphasized the urgent need for such technology to boost productivity and counter talent shortages, pointing to the thousands of disorders that currently have no treatment options. His company’s platform integrates biological, chemical, and clinical data to propose disease targets and potential drug molecules. This automation of traditionally labor-intensive steps allows for rapid screening of enormous design spaces, identification of high-quality candidates, and exploration of existing drugs for new uses, all at a fraction of the usual time and expense. For instance, Insilico has already applied its AI to evaluate approved medications for potential repurposing against ALS.
However, the innovation challenge extends beyond initial discovery. For many conditions, effective treatment requires intervention at the most fundamental genetic level. This is where companies like GenEditBio enter the picture, operating in the advancing field of in vivo CRISPR gene editing. Their goal is to transition from editing cells outside the body to achieving precise, one-time injections directly into affected tissues. The company’s co-founder, Tian Zhu, explained their development of a proprietary engineered protein delivery vehicle (ePDV), a virus-like particle designed for targeted delivery. By employing AI and machine learning to analyze natural viral structures, they identify particles with inherent affinities for specific tissues, such as the eye or liver. This approach centers on a vast library of non-viral polymer nanoparticles. Their NanoGalaxy platform uses AI to decipher how chemical compositions correlate with tissue targeting, predicting modifications that will allow safe payload delivery without immune system activation. Laboratory results continuously feed back into the AI to refine its predictions. Zhu states this method lowers costs and standardizes a process ripe for scaling, akin to creating an off-the-shelf therapy usable for many patients. GenEditBio recently received FDA clearance to begin trials of a CRISPR-based therapy for a corneal condition.
Progress in these AI-driven fields consistently encounters a significant hurdle: data. Modeling the complexities of human biology demands vast amounts of high-quality, diverse information. Aliper notes that current biomedical data is heavily skewed toward Western populations, creating a bias. He advocates for more localized data collection efforts to build balanced datasets, which would enhance model robustness. Insilico addresses this partly through automated laboratories that generate multi-layered biological data from disease samples at scale. Zhu offers a different perspective, suggesting the necessary data is already embedded in human biology, refined by millennia of evolution. She highlights that most DNA does not code for proteins but instead regulates gene activity, a complex “instruction manual” that AI models are becoming adept at interpreting. In their labs, GenEditBio tests thousands of nanoparticle variants in parallel, creating rich datasets that Zhu describes as “gold for AI systems,” used both for internal training and external partnerships.
Looking forward, Aliper identifies the creation of digital human twins for virtual clinical trials as a major, though nascent, frontier. With annual new drug approvals stagnating and an aging global population increasing the prevalence of chronic disorders, he expresses hope that the next decade will bring a proliferation of personalized therapeutic options, driven by these intelligent systems.
(Source: TechCrunch)
