AI Cracks the Code to the Brain, Unlocking New Hope for CNS Drugs

BOSTON, MA – December 05, 2025 – For decades, the human brain has been a fortress, protected by a biological security system so effective it has thwarted modern medicine time and again. The blood-brain barrier (BBB), a near-impenetrable wall of cells, blocks over 98% of potential drugs from reaching their targets, leaving devastating neurological diseases like Alzheimer's, Parkinson's, and ALS with tragically few effective treatments. Now, a Boston-based biotech is claiming to have developed an AI-powered key to unlock this fortress.

In a landmark paper published in the Journal of Chemical Information and Modeling, 1910 Genetics has detailed CANDID-CNS™, an artificial intelligence model that dramatically improves the prediction of which molecules can cross the BBB. By moving beyond long-standing drug design rules and accounting for subtle molecular geometry, the technology promises to open up a vast, previously inaccessible chemical landscape for neuroscience, potentially heralding a new era in the fight against CNS disorders.

Beyond the 'Rule of 5' Barrier

At the heart of the challenge lies a set of guidelines known as Lipinski's Rule of 5. Established over two decades ago, these rules helped medicinal chemists filter for smaller, simpler molecules more likely to be absorbed and processed by the body. While effective for many drugs, this framework inadvertently sidelined a massive class of larger, more complex compounds known as 'Beyond Rule of 5' (bRo5) molecules. These compounds possess the potential to engage difficult-to-drug targets that smaller molecules cannot, but their size makes them prime candidates for rejection by the BBB.

Neuroscience has been particularly constrained by this paradigm. Existing computational tools have consistently failed to predict which of these complex bRo5 molecules might successfully breach the brain's defenses. Compounding the problem is stereochemistry—the three-dimensional arrangement of atoms in a molecule. Two molecules can be mirror images of each other (stereoisomers) but have vastly different biological effects, including their ability to cross the BBB. Most prediction tools are blind to this crucial distinction.

"For too long, the industry has been searching for keys under the same lamppost because it's where the light was brightest," explained an industry analyst following the AI-biotech space. "Models that can reliably predict complex properties like BBB penetration for bRo5 compounds don't just offer an incremental improvement; they change the fundamental economics of CNS drug discovery by illuminating a much larger search area."

Performance that Redefines the State-of-the-Art

CANDID-CNS™ appears to be that new source of light. The model, built on an attentive graph neural network (GNN) architecture, was designed to learn the subtle physicochemical principles that govern brain entry. The results presented by 1910 Genetics are striking. When tested on bRo5 molecules, CANDID-CNS™ achieved an 87% Area Under the Precision-Recall Curve (AUPRC), a measure of predictive accuracy, compared to just 56% for Pfizer’s widely used CNS MPO score. When tasked with distinguishing between effective and ineffective stereoisomers, the model succeeded 68% of the time, a significant leap over the 50% coin-flip accuracy of the industry standard.

“CANDID-CNS™ does not just classify molecules – it recovers the physicochemical principles that drive BBB transport,” said Jesse Collins, Ph.D., Senior AI Research Scientist at 1910 and lead author of the publication. “Its predictions correlate with quantum mechanical hydration free energy, indicating that the model implicitly learns the thermodynamic determinants of passive permeability. That mechanistic signal enables CANDID-CNS™ to generalize and identify brain penetrant bRo5 molecules and stereoisomers.”

This ability to understand the 'why' behind a prediction, not just the 'what,' is a critical differentiator. It suggests the AI is not merely pattern-matching but developing a foundational understanding of molecular physics, allowing it to make more reliable predictions on novel chemical structures it has never seen before.

From AI Model to Clinical Hope

The impact of this technology is not just theoretical. 1910 Genetics has already leveraged CANDID-CNS™ in its own pipeline. The model was instrumental in the discovery of 1910-102, a non-opioid, covalent small molecule inhibitor being developed for chronic pain. The promise of this program was significant enough to attract partial funding from the NIH's HEAL (Helping to End Addiction Long-term) Initiative, underscoring the real-world need for new, effective CNS therapeutics.

This initial success provides a powerful proof of concept for the company's broader ambitions. By unlocking a new chemical toolbox, the company hopes to accelerate the development of drugs for some of the most intractable brain diseases.

“Neuroscience has long been defined by what we can’t reach,” noted Jen Asher, Ph.D., Founder and CEO of 1910, in the company's announcement. “CANDID-CNS™ expands the boundaries of what’s considered druggable in the brain. By overcoming the limitations of bRo5 design and learning stereochemical effects, it opens an entirely new bRo5 chemical space for CNS drug discovery – bringing us closer to effective treatments for diseases like Alzheimer’s, Parkinson’s, and ALS.”

The AI-Native Biotech Blueprint

The development of CANDID-CNS™ offers a window into a new kind of company: the 'AI-native' biotech. Unlike traditional pharmaceutical giants that are bolting on AI capabilities, firms like 1910 Genetics are built from the ground up with AI and automation at their core. CANDID-CNS™ is just one of roughly 100 AI models within 1910's ITO™ platform, an integrated system that combines multimodal data, advanced computation, and high-throughput lab automation to design both small and large molecule drugs.

This integrated approach has attracted significant investment from tech-savvy venture firms, including Microsoft's M12 and Playground Global, who co-led a $22 million Series A round. The company has also forged a strategic partnership with Microsoft to scale its platform on the Azure cloud, signaling a deep convergence between big tech and biotech.

While 1910's performance metrics are impressive, it is not alone in this race. Competitors like Lantern Pharma are also reporting high accuracy rates for their own AI-driven BBB prediction tools. This burgeoning competition highlights a pivotal industry shift, where a company's competitive advantage may lie less in its chemical library and more in the sophistication of its predictive algorithms. For a field as challenging and costly as CNS drug development, the ability to more accurately predict success before a single molecule is synthesized in a lab could save billions of dollars and years of wasted effort, ultimately reshaping the path from digital model to patient medicine.