Phylo’s Target Atlas Puts AI Agents on Drug Target Prioritization

Phylo launched Target Atlas on September 29, 2026, an agentic target-prioritization application built on the company’s Biomni Lab platform, with a free public edition and a first disease-specific edition for Parkinson’s disease developed with The Michael J. Fox Foundation.

How Target Atlas Works

Phylo said the application can cut the evidence-gathering behind target reviews from months of work by teams of scientists to days. The system creates a shared record that links findings to their sources and can be updated as new research appears. AI agents apply consistent analyses across targets, while scientists decide which evidence matters and how much weight it should carry. Teams can change those weights to see how their own criteria alter the rankings.

The public edition is free with a Biomni account and spans thousands of target–disease assessments, presented through source-linked target profiles and interactive 2×2 plots. According to the company, hundreds of AI agents curate and synthesize the underlying evidence. Across diseases, versioned computational skills turn evidence from databases, papers, and disease datasets into dashboards and source-linked profiles that record each item’s source, date, coverage, and missingness. Researchers can inspect the evidence behind each result and use a chat interface to explore the evidence or edit private views.

Evidence Cited Behind the Application

The announcement grounds the product in published drug-development attrition figures. It cites a 2014 analysis in Nature Biotechnology by Hay and colleagues finding that about one in ten drug programs that enter clinical development is approved, and a 2024 Nature study by Minikel and colleagues estimating that drug mechanisms with human genetic support carry a 2.6-fold higher probability of success from Phase 1 to approval. Target choice, the release states, is one of the earliest and least expensive points at which to change a drug program, and one of the most costly to get wrong, yet the evidence needed to choose well is fragmented across genetics, expression, clinical activity, safety, tractability, biomarkers, competition, and the literature, and changes continuously.

The Michael J. Fox Foundation Parkinson’s Edition

The first disease-specific application of Target Atlas is a Parkinson’s edition developed with the foundation’s Targets to Therapies (T2T) Initiative, which established a transparent Parkinson’s target-prioritization framework through community nominations, expert panels, curated scorecards, and a published methodology. Phylo said the edition makes that expert framework openly explorable and connects it to source-linked analyses that can be refreshed as new evidence emerges. The underlying methodology was published by Vaiana and colleagues in npj Parkinson’s Disease in 2025.

According to the foundation’s initiative page, T2T launched in early 2024 with a survey of the scientific community on which targets offer the greatest potential for new Parkinson’s treatments. More than 290 targets were evaluated, and 21 were nominated based on the feasibility of addressing bottlenecks over the next two to three years. Three scientific cores power the initiative: prioritization and selection, validation and toolkit development, and a public knowledgebase. The foundation funds research teams to mechanistically validate the selected targets for possible future drug development, and all learnings and tools are made freely available through the shared knowledgebase. The page names alpha-synuclein, GBA1, and LRRK2 as targets with therapies already in clinical testing for Parkinson’s, and notes the initial selection round focused on targets with potential to slow or modify disease progression.

“Expanding the number of well-characterized, druggable targets is a critical part of the puzzle,” Gaia Skibinski, VP and head of translational therapeutics at The Michael J. Fox Foundation for Parkinson’s Research, said in the launch announcement. She said the foundation looks forward to exploring how emerging AI tools like Target Atlas can help researchers move promising targets toward potential new Parkinson’s treatments.

Enterprise Edition, Expansion Plans, and Availability

The enterprise edition of Target Atlas is available by annual license for a governed Biomni deployment and supports more than 20,000 targets per selected disease, proprietary data integration, target-pair dashboards, and AI/ML-based scoring.

Phylo said it plans to expand the public edition across all major diseases over time, and it is seeking foundations, consortia, and patient organizations to sponsor additional disease editions that would be published free to their communities, with each organization’s framework represented in the scoring. Kexin Huang, the company’s co-founder and CEO, said the T2T collaboration shows that agents can assemble and score in days evidence that takes a community months to process by hand, keep it current, and do so without replacing the expert judgment that decides which evidence matters. Huang said the company’s goal is to bring the model to every major disease.

Phylo describes Biomni Lab as the first Integrated Biology Environment, a unified workspace where scientists collaborate with AI agents across literature research, experimental design, biological data analysis, and scientific decision-making, and said the platform is used by biopharmaceutical companies including Chugai Pharmaceutical and Ono Pharmaceutical. The company is based in South San Francisco and raised a $13.5 million seed round co-led by Andreessen Horowitz and Menlo Ventures’ Anthology Fund with Anthropic.

Target Atlas, including the MJFF Parkinson’s edition, is available at biomni.phylo.bio. R&D leaders can request an enterprise demonstration, and foundations and consortia can discuss sponsoring a disease edition, through contact@phylo.bio.