{"id":14096,"date":"2026-06-23T22:18:37","date_gmt":"2026-06-24T05:18:37","guid":{"rendered":"https:\/\/www.solix.com\/blog\/?p=14096"},"modified":"2026-07-01T22:56:11","modified_gmt":"2026-07-02T05:56:11","slug":"signal-over-noise-how-ai-finds-the-drugs-hiding-in-plain-sight","status":"publish","type":"post","link":"https:\/\/www.solix.com\/blog\/signal-over-noise-how-ai-finds-the-drugs-hiding-in-plain-sight\/","title":{"rendered":"Signal Over Noise: How AI Finds the Drugs Hiding in Plain Sight","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
\n

A preview of the Solix EAI Pharma white paper \u2014 “Signal Over Noise: A Framework for AI-Driven Drug Repurposing and Target Identification in Pharma R&D.”<\/p>\n<\/blockquote>\n

In January 2020, as SARS-CoV-2 spread and no specific treatment existed, a research team put a single question to an AI-enhanced biomedical knowledge graph: among the drugs already approved and sitting on pharmacy shelves, was there one that could block the mechanism the virus used to enter human cells?<\/p>\n

The system surfaced an unexpected name \u2014 baricitinib<\/b>, a rheumatoid arthritis drug.<\/p>\n

The logic was not obvious, and that is the entire point. Baricitinib inhibits an enzyme called AAK1, a regulator of the cellular process SARS-CoV-2 was using to enter lung cells. AAK1’s role in viral entry had been documented across papers on dengue, Ebola, HCV, and HIV \u2014 literature that is almost never read alongside rheumatology research. The connection was not missing from science. It was scattered across domains no single researcher reads in combination.<\/p>\n

By November 2020, baricitinib had an FDA Emergency Use Authorization.<\/p>\n

The story is usually told as a COVID story. It is really a target identification story<\/b> \u2014 and it carries a lesson that applies to almost every R&D pipeline in the industry today.<\/p>\n

The connection was never the problem. The noise was.<\/h2>\n

It is tempting to read the baricitinib case as serendipity, or as a uniquely clever algorithm. It is neither. The biology had not changed. AAK1’s role was knowable from existing science. What changed was the ability to connect it \u2014 to traverse compound mechanism, viral biology, and pharmacokinetic data simultaneously, and surface one defensible hypothesis faster than manual literature review ever could.<\/p>\n

Consider what the system actually did. It found 378 AAK1 inhibitors in the knowledge graph. Of those, 47 were FDA or EMA approved. Most were eliminated on pharmacokinetic grounds \u2014 their plasma concentrations at safe doses were simply too low to inhibit AAK1 meaningfully. Baricitinib stood out because its standard dose achieved sufficient concentration, and its years of rheumatoid arthritis use provided a known safety baseline, and its JAK1\/2 anti-inflammatory action addressed the cytokine storm driving COVID mortality. A dual-mechanism candidate, surfaced and filtered from a field of hundreds.<\/p>\n

This is the central problem of modern drug discovery, and it is not a shortage of data \u2014 it is the opposite. Genomic sequencing has produced enormous target landscapes. High-throughput screening has generated billions of compound-activity data points. The scientific knowledge base has never been larger. Yet the number of drugs approved per billion dollars of R&D spend has roughly halved every nine years since the 1950s, a paradox known as Eroom’s Law.<\/p>\n

The bottleneck has quietly shifted. It is no longer generating knowledge. It is finding the signal inside it.<\/p>\n

Why this matters now<\/h2>\n

The economics make the case unavoidable. Bringing a new molecule to market costs around $2.6 billion, and fewer than 11% of Phase I candidates ever reach approval \u2014 nearly nine in ten fail.<\/p>\n

Repurposing changes the math. The average repurposing programme costs roughly $300 million \u2014 about a tenth of de novo discovery \u2014 because the compound already carries safety and pharmacokinetic data. It is no accident that an estimated 35% of transformative FDA-approved drugs emerged through repurposing pathways.<\/p>\n

The opportunity is not theoretical. It is most likely sitting inside your own pipeline right now \u2014 in a shelved asset, a Phase 2 disappointment, or a compound whose full mechanism was never fully explored. The question is whether your infrastructure is built to find it.<\/p>\n

From serendipity to system<\/h2>\n

Inspiration does not scale. A single lucky connection, however valuable, is not a strategy. The real prize is turning that kind of discovery into a repeatable process \u2014 one that can interrogate an entire asset base, again and again, and surface the connections worth pursuing.<\/p>\n

That requires moving past the idea of AI as a black box. In the baricitinib case, the reasoning was fully traceable: AAK1 inhibition blocks viral entry, combined with JAK1\/2 anti-inflammatory action, produces a dual therapeutic rationale. The system made the inference chain visible, not just the output. This is not a compliance nicety. No R&D team makes a $50 million clinical investment \u2014 and no regulator accepts a submission \u2014 on a candidate no one can explain.<\/p>\n

Signal over noise, made systematic and made explainable.<\/p>\n

What the full white paper covers<\/h2>\n

This preview makes the case for why. The white paper lays out the how \u2014 a vendor-neutral framework R&D teams can apply to their own pipelines, regardless of platform:<\/p>\n