AI Drug Discovery Trial Reveals Promise and Challenges of Using the Tech

As artificial intelligence (AI) promises to remake drug discovery, recent clinical trial results reveal its potential and challenges.

Recursion Pharmaceuticals, a self-described “clinical stage TechBio company,” recently announced results from its Phase 2 SYCAMORE trial for REC-994, a drug candidate targeting cerebral cavernous malformation (CCM), a rare brain disorder. The trial met its primary safety endpoint but showed mixed efficacy results, illustrating the complexities of translating AI-driven discoveries into clinical success.

The company’s Recursion OS platform uses machine-learning algorithms to analyze vast datasets, aiming to identify new drug candidates more efficiently than traditional methods. This approach represents a growing trend in the pharmaceutical industry to leverage AI in drug discovery. It also shows the challenges.

“While artificial intelligence excels at analyzing vast datasets, the scarcity of information on uncommon neurological conditions poses a major challenge,” Keaun Amani, CEO of Neurosnap, an AI platform used by labs, told PYMNTS. “Limited patient populations make it difficult to gather sufficient data for training accurate AI models.”

AI’s Growing Role in Drug Discovery

Progress in AI-driven drug discovery is evident. Alister Campbell, VP of science and technology at Dotmatics, told PYMNTS that since 2015, AI-native biotechnology companies and their partners have brought 75 candidates to clinical trials, with numbers growing yearly.

“AI use in drug discovery comes in many shapes and forms, from drug repurposing to predicting structures of anti-bodies and proteins using algorithms like AlphaFold, designing small molecule drugs using generative AI methods, using AI to mine vast OMIC datasets providing valuable insights into disease biology, druggable targets, and biomarkers,” Campbell said.

Jo Varshney, founder and CEO of AI drug discovery company VeriSIM Life, told PYMNTS: “Neurological conditions often lack clear, easily measurable indicators in lab tests or clinical assessments, resulting in a data scarcity that limits the effectiveness of AI systems.”

Recursion’s SYCAMORE clinical trial for CCM, which affects approximately 360,000 symptomatic individuals in the U.S. and EU, illustrates these challenges. Dr. Najat Khan, chief R&D officer at Recursion, noted “promising trends in exploratory efficacy endpoints,” particularly at the highest dose. However, the company acknowledged that “improvements in either patient or physician-reported outcomes were not yet seen at the 12 month time point.”

The trial’s outcome reflects broader industry challenges. According to Amani, “Mixed results in clinical studies reveal that while AI has great potential to revolutionize drug discovery, it still faces significant hurdles in accurately predicting drug efficacy. One major challenge is the complexity of biological systems, which AI models often struggle to fully capture.”

Navigating the Path Forward

Experts suggest various approaches to advance AI in drug discovery. Amani envisions developing more complex models capable of analyzing larger biological systems. He suggests “developing all-atom models capable of analyzing larger, more complex biological systems. These models, combined with a growing trend of blending machine learning and physics-based methods, offer the potential to simulate molecular interactions with unprecedented accuracy.”

Campbell proposes combining AI with traditional techniques to identify relevant biological targets and develop drug candidates more efficiently. He suggests a multi-pronged approach to identify clinically relevant biological targets, develop ideal candidates more quickly and cheaply, and reduce the chances of failure due to safety, efficacy and cost issues.

Accessibility of AI tools is also crucial. Amani notes that platforms like Neurosnap have streamlined the process, making it easier for scientists to use these tools. “Accessing AI-based tools for drug discovery can often be technically prohibitive to researchers,” Amani said. “Platforms like Neurosnap have greatly streamlined this process making it easier for scientists to efficiently utilize the tools they need.”

Varshney said developing more sophisticated “knowledge” or mechanistic systems that intricately incorporate detailed aspects of biology could yield more accurate and reliable predictions when integrated with AI.