apheris AI

Apheris is a German technology company that builds and operates federated data networks for the biopharma industry. Rather than running its own drug programs, Apheris turns proprietary data contributed by network members into machine-learning models that no public dataset could match, then delivers those models to run and fine-tune inside each member's own environment. The company describes itself as operating "the industry's federated data networks" and explicitly states that it "runs no drug programs of its own." Its platform, Apheris Foundry, is the deployment layer for these networks, while ApherisFold provides co-folding predictions in the browser. The company organizes its work around several domain-specific networks. AI Structural Biology (AISB) focuses on better protein-complex prediction models trained across biopharma data. ADMET aims to design more informative DMTA (design-make-test-analyze) batches and use experimental capacity better. Antibody Developability offers purpose-built datasets and federated AI training for antibody R&D. Real-World Evidence (RWE) applies federated learning across distributed healthcare data sources. Members contribute proprietary data to a shared network, but training runs inside each member's own environment; only model weights are aggregated, so raw data never leaves its owner. Apheris reports substantial proprietary data assets behind its networks, including more than 25,000 proprietary co-folding structures, more than 100 million binding data points, over 900,000 ADME and toxicity measurements across 700 tasks, and a 10,000-item antibody developability ML-ready dataset being built with Ginkgo. The company claims its models outperform public alternatives: 52.1% of predictions reach experimental-grade interface accuracy (PL-lDDT ≥ 0.8) versus 38.4% for the average public model, and 46.8% of predictions reach an accurate ligand pose (bisyRMSD ≤ 2 Å) versus 32.2% for the comparison baseline. Apheris is headquartered in Berlin, Germany, and markets its offerings globally to pharmaceutical and biotechnology companies, healthcare data holders, and research partners. Its business model is network-based and collaborative: partners join a network, contribute data, and receive access to models trained across the pooled, privacy-preserving data. The company also maintains a Trust Center for compliance and audit reports, reflecting the sensitivity of the data it handles. Investors back the company, and its team spans engineering, science, and go-to-market roles.

Apheris is a privately held German technology company headquartered in Berlin. It was founded to address a central tension in biomedical research: the most valuable data for training drug discovery models is proprietary and siloed across pharmaceutical companies, hospitals, and research institutions, yet pooling that data into a central repository is often legally, commercially, or ethically impossible. Apheris's answer is federated learning at industry scale. The company built its business around federated data networks. In this model, members contribute proprietary data to a shared network, but training runs inside each member's own environment; only model weights are aggregated. This preserves data sovereignty while allowing models to learn from a far larger and more diverse body of data than any single organization possesses. Apheris states that it "operates the industry's federated data networks and runs no drug programs of its own," positioning itself as neutral infrastructure rather than a drug developer. Apheris's platform, Apheris Foundry, is the deployment layer for these networks. The company also offers ApherisFold, which provides co-folding predictions in the browser. Around this platform, Apheris has organized domain-specific networks. AI Structural Biology (AISB) targets better protein-complex prediction models trained across biopharma data. ADMET focuses on designing more informative DMTA batches and using experimental capacity better. Antibody Developability provides purpose-built datasets and federated AI training for antibody R&D. Real-World Evidence (RWE) applies federated learning across distributed healthcare data sources. The company reports significant proprietary data assets behind its networks: more than 25,000 proprietary co-folding structures, more than 100 million binding data points, over 900,000 ADME and toxicity measurements across 700 tasks, and a 10,000-item antibody developability ML-ready dataset being built with Ginkgo. It also publishes performance claims for its models, including 52.1% of predictions reaching experimental-grade interface accuracy (PL-lDDT ≥ 0.8) versus 38.4% for the average public model, and 46.8% of predictions reaching an accurate ligand pose (bisyRMSD ≤ 2 Å) versus 32.2% for the comparison baseline. Apheris markets its offerings globally to pharmaceutical and biotechnology companies, healthcare data holders, and research partners. It maintains a Trust Center for compliance and audit reports, reflecting the sensitivity of the data it handles. The company is backed by investors and its team spans engineering, science, and go-to-market roles. As of the available evidence, Apheris remains an active private company.

Federated data networks and models for drug discovery that run inside each member's own environment, without moving proprietary data.

• Apheris Foundry

  The deployment layer for Apheris's federated data networks; models are delivered to run and fine-tune securely where the member's data lives.

• ApherisFold

  Co-folding predictions in the browser.

• AI Structural Biology (AISB)

  Better protein-complex prediction models, trained across biopharma data.

• ADMET

  Design more informative DMTA batches and use experimental capacity better.

• Antibody Developability

  Purpose-built datasets and federated AI training for antibody R&D.

• Real-World Evidence (RWE)

  Federated real-world evidence across distributed healthcare data sources.

1. 1Superior drug discovery models | Apherisapheris.com
2. 2About | Apherisapheris.com
3. 3Wikidatawikidata.org