SSStrategicSignal
active · legacy review required

SS-AIDT-2025-0001 · Oct 1, 2025

AbbVie → OpenFold Consortium

Several thousand experimentally determined, proprietary protein–small molecule structures contributed by pharmaceutical companies for privacy-preserving fine-tuning of OpenFold3.

Pharmaceuticals & biotechnologyFederated / Controlled Learning RightsResearch date Sep 15, 2026
PendingAI Data Asset Score
PendingTransaction Signal Score
Data ownerAbbVie, Johnson & Johnson, Bristol Myers Squibb, Takeda, Astex Pharmaceuticals
AI buyer / developerOpenFold Consortium, AlQuraishi Lab, Apheris
Asset stateEmbedded / Continuing
CorpusRenewable Corpus

1. What happened

Federated fine-tuning through Apheris; source data remains in each contributor's controlled environment.

2. Buyer / seller

OpenFold Consortium, AlQuraishi Lab, Apheris is the AI buyer or model-development side. AbbVie, Johnson & Johnson, Bristol Myers Squibb, Takeda, Astex Pharmaceuticals owns or contributes the proprietary corpus.

3. Economics

No data-specific consideration disclosed. Economics scope: not disclosed. Strategic Signal has not estimated undisclosed consideration.

4. Proprietary data

Several thousand experimentally determined, proprietary protein–small molecule structures contributed by pharmaceutical companies for privacy-preserving fine-tuning of OpenFold3. Historical depth: not_disclosed. Refreshability: periodic consortium contributions.

5. Rights and restrictions

Data stays in each owner's environment; secure aggregation prevents raw-data pooling. Ownership: no disclosed transfer. Training: yes explicit. Post-training: yes explicit fine tuning.

6. Why the data matters for AI

Post-train and validate OpenFold3 for joint protein–ligand structure prediction and drug discovery. High: experimentally validated molecular structures encode expensive real-world outcomes.

7. Asset Score — Pending/100

Legacy score withdrawn pending evidence-linked scoring and human approval. The superseded v1.0.0 release retains the original values. Score coverage: 0%.

FactorWeightRating / 5Points
Uniqueness / scarcity12Unknown
Historical depth8Unknown
Decision → outcome richness13Unknown
Decision Value Density13Unknown
Real-world grounding10Unknown
Domain economic value9Unknown
Refreshability8Unknown
Proprietary advantage9Unknown
Model-learning usefulness10Unknown
Non-replicability4Unknown
Rights usability4Unknown

8. Transaction Signal Score — Pending/100

Legacy score withdrawn pending evidence-linked scoring and human approval. The superseded v1.0.0 release retains the original values. Score coverage: 0%.

FactorWeightRating / 5Points
Disclosed economics16Unknown
Clean price discovery16Unknown
Separability of data consideration15Unknown
Explicit training / model-improvement use14Unknown
Clarity of rights purchased12Unknown
Competing bids10Unknown
Strategic buyer quality6Unknown
Repeatability / precedent value6Unknown
Independent model-uplift evidence5Unknown

9. M&A / valuation implications

Demonstrates that federated learning can monetize embedded scientific archives without separating or selling the underlying intellectual property.

10. Evidence boundary

The existence and federated training purpose are disclosed facts. Strategic-value and M&A statements are Strategic Signal analysis.