The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
_meta: struct<source: string, captured_at: timestamp[s], captured_by: string, schema_version: string, purpo (... 61 chars omitted)
child 0, source: string
child 1, captured_at: timestamp[s]
child 2, captured_by: string
child 3, schema_version: string
child 4, purpose: string
child 5, refresh_cadence: string
child 6, five_bucket_ref: string
orgs: list<item: struct<slug: string, display_name: string, hf_url: string, bucket: string, subcategory: s (... 102 chars omitted)
child 0, item: struct<slug: string, display_name: string, hf_url: string, bucket: string, subcategory: string, topi (... 90 chars omitted)
child 0, slug: string
child 1, display_name: string
child 2, hf_url: string
child 3, bucket: string
child 4, subcategory: string
child 5, topics: list<item: string>
child 0, item: string
child 6, flagship: list<item: string>
child 0, item: string
child 7, bc_relevance: string
child 8, notes: string
five_bucket_summary: struct<academic_institution: int64, therapeutic_developer: int64, service_provider: int64, enterpris (... 47 chars omitted)
child 0, academic_institution: int64
child 1, therapeutic_developer: int64
child 2, service_provider: int64
child 3, enterprise_tech: int64
child 4, true_noise: int64
child 5, total: int64
high_bc_relevance_orgs_for_priority_lookup: list<item: string>
child 0, item: string
registry_routing_examples: list<item: struct<raw_name: string, canonical_domain: string, hs_org: string, action: string>>
child 0, item: struct<raw_name: string, canonical_domain: string, hs_org: string, action: string>
child 0, raw_name: string
child 1, canonical_domain: string
child 2, hs_org: string
child 3, action: string
cases: list<item: struct<raw_name: string, expect_classification: string, expect_action_one_of: list<item: (... 234 chars omitted)
child 0, item: struct<raw_name: string, expect_classification: string, expect_action_one_of: list<item: string>, ex (... 222 chars omitted)
child 0, raw_name: string
child 1, expect_classification: string
child 2, expect_action_one_of: list<item: string>
child 0, item: string
child 3, expect_target_registry: string
child 4, notes: string
child 5, expect_target_canonical_name: string
child 6, expect_subcategory: string
child 7, expect_hs_org: string
child 8, evidence_source: string
child 9, expect_institution_type: string
child 10, expect_parent_company: string
to
{'_meta': {'purpose': Value('string'), 'version': Value('int64'), 'created': Value('timestamp[s]'), 'updated': Value('timestamp[s]'), 'v2_notes': Value('string')}, 'cases': List({'raw_name': Value('string'), 'expect_classification': Value('string'), 'expect_action_one_of': List(Value('string')), 'expect_target_registry': Value('string'), 'notes': Value('string'), 'expect_target_canonical_name': Value('string'), 'expect_subcategory': Value('string'), 'expect_hs_org': Value('string'), 'evidence_source': Value('string'), 'expect_institution_type': Value('string'), 'expect_parent_company': Value('string')})}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 99, in get_rows_or_raise
return get_rows(
^^^^^^^^^
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/src/services/worker/src/worker/utils.py", line 77, in get_rows
rows_plus_one = list(itertools.islice(ds, rows_max_number + 1))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2690, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2227, in __iter__
for key, pa_table in self._iter_arrow():
^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2251, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 494, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 384, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 295, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 128, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2321, in table_cast
return cast_table_to_schema(table, schema)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2249, in cast_table_to_schema
raise CastError(
datasets.table.CastError: Couldn't cast
_meta: struct<source: string, captured_at: timestamp[s], captured_by: string, schema_version: string, purpo (... 61 chars omitted)
child 0, source: string
child 1, captured_at: timestamp[s]
child 2, captured_by: string
child 3, schema_version: string
child 4, purpose: string
child 5, refresh_cadence: string
child 6, five_bucket_ref: string
orgs: list<item: struct<slug: string, display_name: string, hf_url: string, bucket: string, subcategory: s (... 102 chars omitted)
child 0, item: struct<slug: string, display_name: string, hf_url: string, bucket: string, subcategory: string, topi (... 90 chars omitted)
child 0, slug: string
child 1, display_name: string
child 2, hf_url: string
child 3, bucket: string
child 4, subcategory: string
child 5, topics: list<item: string>
child 0, item: string
child 6, flagship: list<item: string>
child 0, item: string
child 7, bc_relevance: string
child 8, notes: string
five_bucket_summary: struct<academic_institution: int64, therapeutic_developer: int64, service_provider: int64, enterpris (... 47 chars omitted)
child 0, academic_institution: int64
child 1, therapeutic_developer: int64
child 2, service_provider: int64
child 3, enterprise_tech: int64
child 4, true_noise: int64
child 5, total: int64
high_bc_relevance_orgs_for_priority_lookup: list<item: string>
child 0, item: string
registry_routing_examples: list<item: struct<raw_name: string, canonical_domain: string, hs_org: string, action: string>>
child 0, item: struct<raw_name: string, canonical_domain: string, hs_org: string, action: string>
child 0, raw_name: string
child 1, canonical_domain: string
child 2, hs_org: string
child 3, action: string
cases: list<item: struct<raw_name: string, expect_classification: string, expect_action_one_of: list<item: (... 234 chars omitted)
child 0, item: struct<raw_name: string, expect_classification: string, expect_action_one_of: list<item: string>, ex (... 222 chars omitted)
child 0, raw_name: string
child 1, expect_classification: string
child 2, expect_action_one_of: list<item: string>
child 0, item: string
child 3, expect_target_registry: string
child 4, notes: string
child 5, expect_target_canonical_name: string
child 6, expect_subcategory: string
child 7, expect_hs_org: string
child 8, evidence_source: string
child 9, expect_institution_type: string
child 10, expect_parent_company: string
to
{'_meta': {'purpose': Value('string'), 'version': Value('int64'), 'created': Value('timestamp[s]'), 'updated': Value('timestamp[s]'), 'v2_notes': Value('string')}, 'cases': List({'raw_name': Value('string'), 'expect_classification': Value('string'), 'expect_action_one_of': List(Value('string')), 'expect_target_registry': Value('string'), 'notes': Value('string'), 'expect_target_canonical_name': Value('string'), 'expect_subcategory': Value('string'), 'expect_hs_org': Value('string'), 'evidence_source': Value('string'), 'expect_institution_type': Value('string'), 'expect_parent_company': Value('string')})}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Life Sciences Stakeholder Relationship Map (Tier A3 draft — expanded)
Target: publish as HF dataset under
biocreative-strategies/life-sciences-stakeholder-mapFormat: dataset card (this file → README.md) + JSON artifact (the relationship graph) + optional Mermaid diagram License: CC-BY-4.0 IP exposure: very low (high-level industry framing + public-domain HS-cataloged org examples only) Per-user direction (2026-04-30): "expand A to ICP and our stable understanding of the relationships between TDs, service providers, academic institutions, and financial partners. Sharing the high-level categorization is fine — the specialty stuff is contextual at the client level."
What it is
A public-data-only, high-level industry framing of the five stakeholder buckets in life-sciences commercialization, and how they relate to each other. Designed as a reference for anyone reading Hugging Science (researchers, agents, commercial teams) who needs to understand who actually buys, builds, funds, and operates in the life-sciences AI/ML space.
The 5 buckets
1. Therapeutic Developers (TDs)
The companies whose mission is to bring a therapy to market. They own the IND, run the clinical trials, and (if they make it) launch the drug.
Public sub-types we use:
- Cell & gene therapy biotech
- Antibody / biologics biotech
- Small-molecule biotech
- AI-native therapeutic developers (a newer breed — Xaira, Tahoe Bio, Eve Bio — built around AI from day 1)
- Big pharma (Pfizer, Roche, Novartis, etc.)
Public examples on Hugging Science: Xaira-Therapeutics, tahoebio, eve-bio, Aignostics
2. Service Providers
Companies that sell into therapeutic developers. They don't own the drug; they help TDs design, manufacture, test, transport, file, market, or sell it.
Public sub-types:
- CROs (Contract Research Organizations) — research and clinical-trial services
- CDMOs (Contract Development & Manufacturing Organizations) — actual drug manufacturing
- Equipment vendors — bioreactors, centrifuges, analytical instruments
- Lab supplies / consumables — reagents, media, plastics
- Software & technology vendors — LIMS, MES, AI/ML platforms, bioinformatics, MLOps
- Diagnostics partners — companion diagnostics, analytical testing
- Consulting — regulatory, quality, scientific, commercial
Public examples on Hugging Science: ginkgo-datapoints (Ginkgo Bioworks), SandboxAQ, openadmet, futurehouse, owkin, InstaDeepAI (BioNTech subsidiary), Aignostics
3. Academic Institutions
Universities, research institutes, and academic labs. They're the originating science: PIs running labs, generating IP, training future industry scientists, and increasingly publishing AI/ML models and datasets.
Public sub-types:
- R1 universities (Harvard, MIT, Stanford, etc.)
- Research institutes (Arc Institute, Allen Institute, Flatiron, etc.)
- Medical schools / academic medical centers
- Government research (NIH, NASA, ONERA, etc.)
- Academic consortia (AllTheBacteria, SAIRfoundation, etc.)
Public examples on Hugging Science: arcinstitute, wanglab, jablonkagroup, opig, polymathic-ai, imageomics, nasa-impact, EarthSpeciesProject
4. Financial Partners
Investors who fund the TDs and (sometimes) the service providers.
Public sub-types:
- Venture capital
- Private equity
- Corporate VC arms (e.g., Lilly Ventures, Pfizer Ventures, Novartis Venture Fund)
- Mutual funds / hedge funds (public-market exposure)
- Family offices
- Sovereign wealth
- Government / non-dilutive (NIH grants, BARDA, etc.)
- Incubators / accelerators
Public examples on Hugging Science: Limited direct presence (financial partners don't typically publish models/datasets), but Eric-Schmidt-funded futurehouse is an example of philanthropic-capital-backed work.
5. Other / Noise
Everything that doesn't fit cleanly into 1-4. Encoding errors, individual researchers without affiliation, test data, broken records. Flagged so it doesn't pollute commercial intelligence systems.
How they relate (the graph)
graph LR
AI[Academic Institutions] -->|spin out| TD[Therapeutic Developers]
AI -->|license IP / collaborate| TD
AI -->|publish / share data| HS[Hugging Science]
TD -->|publish / share data| HS
SP[Service Providers] -->|sell INTO| TD
SP -->|publish tools, datasets| HS
FP[Financial Partners] -->|fund| TD
FP -->|fund| SP
FP -->|sometimes fund| AI
AI -->|train future hires for| TD
AI -->|train future hires for| SP
TD -->|partner with| SP
HS -->|signal AI/ML maturity| AI
HS -->|signal AI/ML maturity| TD
HS -->|signal AI/ML maturity| SP
Why this map matters
For any commercial team in life sciences, the most important question is: "who in my universe is buying, building, funding, or operating — and how do they relate?" Without that map, outreach is generic and AI/ML resources from HS feel like noise.
Our framing:
| If you are a... | Your customers are... | Your competitors are... | HS is useful because... |
|---|---|---|---|
| Service provider (CRO, equipment, lab supplies, software, diagnostics, consulting) | Therapeutic developers + academic institutions | Other service providers | Reveals which TDs and labs are AI-active = sophisticated buyers with near-term tooling demand |
| Therapeutic developer | Patients (via clinicians + payers) | Other TDs in the same modality / indication | Reveals which competitors are building open-source AI/ML capabilities (and which academic labs they could partner with) |
| Academic institution | Funding agencies + industry partners | Other academic labs | Reveals industry partners likely to license your work, and which competitor labs are publishing similar models |
| Financial partner | TDs and service providers seeking capital | Other investors | Reveals which TDs/SPs have AI/ML maturity = a leading indicator of platform value |
Connection to AI4Science
Hugging Science makes one thing clear: the AI/ML capability gradient is no longer a private/secret thing. Companies and labs publish models and datasets publicly, and that act of publishing signals something to every other stakeholder. Service providers can use that signal to find sophisticated buyers; TDs can use it to scout partners; academic labs can use it to find industry partners; investors can use it to spot platform potential.
Reading the AI/ML posture of every account in your stakeholder graph is no longer optional. The data is public; the relationships are public; the only remaining edge is in interpretation and speed of action.
What this dataset includes
- The 5-bucket taxonomy (high-level definitions only — no proprietary scoring rubric)
- Public sub-types per bucket
- The relationship graph (Mermaid + JSON)
- Public examples drawn from Hugging Science's own catalog (orgs HS already lists are by definition fair to reference)
- The "If you are a..." reading guide
What this dataset does NOT include
- Any proprietary scoring methodology (BCAS, ICP weights, modality classification)
- Any client data, anonymized or otherwise
- Any specific account intelligence (revenue, deals, contacts, relationships)
- Any internal taxonomies beyond the 5 buckets
- Any non-public organizations
Citation
@misc{biocreative-stakeholder-map-2026,
title={Life Sciences Stakeholder Relationship Map: A 5-bucket framework for AI4Science commercialization},
author={BioCreative Strategies},
year={2026},
howpublished={\url{https://huggingface.co/datasets/biocreative-strategies/life-sciences-stakeholder-map}}
}
- Downloads last month
- 11