AI Just Designed a Drug Moving to Phase III Trials—After Screening Only 79 Molecules

AI Knowledge Graphs Meet Generative Drug Discovery: Inside the Enterprise AI Playbook Cutting R&D Cycles by Years:

How AI Just Slashed a 6-Month Research Phase Down to 3 Weeks:

From AWS GraphRAG's 87% faster pharmaceutical research to Insilico Medicine's AI-designed drug advancing to Phase III trials, two breakthroughs reveal what enterprises unlock when they unify data and act on it.

87%: Reduction in Drug R&D Cycle Time via AWS GraphRAG

79: Molecules Generated En Route to a Phase III Candidate

18 Months: From Project Start to Preclinical Nomination

Two AI breakthroughs landing in the same news cycle expose the same underlying truth: enterprises that unify fragmented data and let AI act on it are compressing timelines once measured in years down to weeks. AWS has demonstrated an 87% reduction in pharmaceutical R&D cycles using a GraphRAG architecture that connects siloed proprietary and public data.

Meanwhile, Insilico Medicine's AI-discovered drug rentosertib has advanced to Phase III human trials, proof that generative AI can move from hypothesis to clinical validation. Together, they form a blueprint every data-rich, decision-heavy organization should be studying.

1: The Data Fragmentation Problem Slowing Down Discovery:

Before any algorithm can accelerate a decision, an organization has to solve a much older problem: its own data doesn't talk to itself. In pharmaceutical R&D, AWS found that initial data gathering and screening phases historically took over six months per iteration, yielding a success rate of just five percent. The cause wasn't a lack of data — it was that clinical metrics, laboratory notes, and engineering records sat isolated across separate storage environments, invisible to one another. When key staff left, they took irreplaceable project context with them, stalling active research indefinitely.

This is not a pharma-specific problem. It's the default state of most enterprise data: valuable, abundant, and scattered across systems that were never designed to cross-reference each other. The fix AWS engineered — a GraphRAG framework built on Amazon Neptune Analytics and Bedrock — turned disconnected data points into a single, queryable knowledge network, letting staff ask natural language questions and receive answers mapped to verified sources.

2: How AWS GraphRAG Unifies Enterprise Knowledge:

The architecture pulls in unstructured files from public repositories like PubMed and blends them with internal corporate records. Amazon Comprehend Medical extracts standardized codes from the text, while Amazon Bedrock, running Claude 4.5 Sonnet, summarizes documents and determines topical relevance before AWS Lambda and S3 route the processed data into Neptune Analytics as a structured graph of entities and relationships.

The payoff is measurable. Initial discovery phases that once took six months now conclude in three weeks. Data retrieval speeds improved by 85%, and research review times dropped 70% thanks to automated citation mapping. Just as important for regulated industries, every generated answer carries an exact, traceable evidence trail — the graph traversal steps that justify the conclusion, satisfying compliance requirements that black-box AI outputs simply can't meet.

“Rentosertib was not discovered by starting from a conventional target and simply screening more compounds. It came from a biology-first, ageing-informed AI workflow.”

— Feng Ren, PhD, Co-CEO and Chief Scientific Officer, Insilico Medicine

3: From Knowledge Graphs to Generative Chemistry: Insilico's Rentosertib:

If AWS's GraphRAG shows how to organize enterprise knowledge, Insilico Medicine shows what happens when AI is trusted to act on it. Its Pharma.AI pipeline used PandaOmics to mine genomics, clinical outcomes, literature, and patent data, isolating TNIK as a novel target for idiopathic pulmonary fibrosis (IPF) — a disease with a median survival of just two to four years post-diagnosis.

From there, the Chemistry42 engine applied generative reinforcement learning to design molecules that physically fit the target protein, rather than screening existing compound libraries. The system synthesized just 79 physical molecules before the team advanced its 55th iteration, rentosertib, into preclinical testing — compressing the path to a preclinical candidate to 18 months. In a randomized trial of 71 patients, the 60 mg dose produced a mean lung capacity gain of +98.4 mL, against a 20.3 mL loss in the placebo group, clearing the way for Phase III trials now underway.

4: What These Breakthroughs Mean for Enterprise AI Adoption:

Neither result came from a single clever model. Both came from architecture: connecting fragmented data sources into a governed, queryable system, then layering AI that can reason across it and act with traceable accountability. That is the same principle behind every successful enterprise AI deployment, regardless of industry — pharma, finance, manufacturing, or professional services. The organizations pulling ahead in 2026 are not the ones with the most data. They're the ones whose data can finally talk to itself.

Your Data Deserves the Same Architecture:

AWS needed Neptune Analytics, Bedrock, and a dedicated engineering team to unify its data. Insilico needed years of proprietary AI pipeline development. Most enterprises need neither — they need Agent+.

Otherworlds AI's Agent+ Business AI Platform connects your fragmented data, documents, and workflows into a governed, queryable AI layer — powered by Google Opal automated workflows, starting at $297/month. For organizations with more complex needs, our team also builds custom enterprise AI solutions tailored to your systems.

Learn more at otherworldsai.com