CSpace: a concept embedding space for biomedical applications.

{"sci_version": "1.0", "schema": "https://corpus.bntrg.co.uk/sci/1.0/sci.schema.json", "id": "pmid:40576218", "kind": "science-record", "revision": 2, "content_hash": "sha256:14f23f078420841374016bddff9258f2df2841c42ddd64d61915aedb5dbd664c", "created_at": "2026-11-01T20:30:40Z", "updated_at": "2026-11-01T21:35:21Z", "generator": {"name": "corpus-assay-engine", "run_id": "reading-sci-2026-11-01"}, "paper": {"state": "complete", "title": "CSpace: a concept embedding space for biomedical applications.", "doi": "10.1093/bioinformatics/btaf376", "pmid": "40576218", "authors": ["Danilo Tomasoni", "Luca Marchetti"], "journal": "Bioinformatics (Oxford, England)", "year": 2025, "abstract": "MOTIVATION: The rise of transformer-based architectures has dramatically improved our ability to analyze natural language. However, the power and flexibility of these general-purpose models come at the cost of highly complex model architectures with billions of parameters that are not always needed.\nRESULTS: In this work, we present CSpace: a concise word embedding of biomedical concepts that outperforms all alternatives in terms of out-of-vocabulary ratio and semantic textual similarity task, and has comparable performance with respect to transformer-based alternatives in the sentence similarity task. This ability can serve as the foundation for semantic search by enabling efficient retrieval of conceptually related terms. Additionally, CSpace incorporates ontological identifiers (MeSH, NCBI gene and taxonomy IDs), enabling computationally efficient disease, gene or condition relatedness measurement, potentially unlocking previously unknown disease-condition associations.\nAVAILABILITY AND IMPLEMENTATION: Full and compressed models are available on Zenodo at https://doi.org/10.5281/zenodo.14781672, while training code, examples, interactive visualizations and experiments are available at https://doi.org/10.5281/zenodo.15125706 and on the GitHub repository.", "claims": [{"id": "claim-method", "statement": "The testing procedure includes three steps: (i) compute the cosine similarity score for every combination of term pairs and ontology concept pairs, where ontology concepts are identified by their UMLS Concept Unique Identifiers (CUIs), (ii) measure the Pearson’s correlation coefficient between the computed scores and those provided by human experts, and (iii) compare results with other state-of-the-art embeddings (Zhang et al. 2019), (Noh and Kavuluru 2021).", "kind": "methodological", "target": null, "evidence": [{"asset": "source:pmid:40576218", "locator": "sections[3] \"3 Results\" paragraph 14", "quote": "The testing procedure includes three steps: (i) compute the cosine similarity score for every combination of term pairs and ontology concept pairs, where ontology concepts are identified by their UMLS Concept Unique Identifiers (CUIs), (ii) measure the Pearson’s correlation coefficient between the computed scores and those provided by human experts, and (iii) compare results with other state-of-the-art embeddings (Zhang et al. 2019), (Noh and Kavuluru 2021)."}], "state": "complete"}, {"id": "claim-data-access", "statement": "The code used for data pre-processing, training, visualization and evaluation of CSpace embeddings, all the experiments results and test benchmarks are available at https://doi.org/10.5281/zenodo.15125706 and in the GitHub repository https://github.com/cosbi-research/cspace; the PubMed data is available at https://www.nlm.nih.gov/databases/download/pubmed_medline.html, the PubMedCentral data is subdivided into manuscript collection available at ftp://ftp.ncbi.nlm.nih.gov/pub/pmc/manuscript, and open access commercial and non-commercial use (oa_comm and oa_noncomm) available at ftp://ftp.ncbi.nlm.nih.gov/pub/pmc, while the Pubtator3 data can be found in https://ftp.ncbi.nlm.nih.gov/pub/lu/PubTator3/ Bio and Med archives can be found on the S3 bucket https://www.biorxiv.org/tdm. The US clinical trials database is available at https://aact.ctti-clinicaltrials.org/snapshots directly as SQL data dumps. CSpace model in KeyedVectors (cspace.kv.bin) and compressed format (cspace.compressed.100k.bin) with training data up August 2024 is available at Zenodo (Fondazione The Microsoft Research - COSBI 2025). Alongside with CSpace, in the same location there is the co-occurrence model (cspace.bigrams.pkl) and concept-frequency model (cspace.dict.pkl). See the examples folder in the GitHub repository for a complete example on how to use them, and the preprocessing folder for the code to build the CSpace training data.", "kind": "methodological", "target": null, "evidence": [{"asset": "source:pmid:40576218", "locator": "sections[10] \"Data availability\" paragraph 2", "quote": "The code used for data pre-processing, training, visualization and evaluation of CSpace embeddings, all the experiments results and test benchmarks are available at https://doi.org/10.5281/zenodo.15125706 and in the GitHub repository https://github.com/cosbi-research/cspace; the PubMed data is available at https://www.nlm.nih.gov/databases/download/pubmed_medline.html, the PubMedCentral data is subdivided into manuscript collection available at ftp://ftp.ncbi.nlm.nih.gov/pub/pmc/manuscript, and open access commercial and non-commercial use (oa_comm and oa_noncomm) available at ftp://ftp.ncbi.nlm.nih.gov/pub/pmc, while the Pubtator3 data can be found in https://ftp.ncbi.nlm.nih.gov/pub/lu/PubTator3/ Bio and Med archives can be found on the S3 bucket https://www.biorxiv.org/tdm. The US clinical trials database is available at https://aact.ctti-clinicaltrials.org/snapshots directly as SQL data dumps. CSpace model in KeyedVectors (cspace.kv.bin) and compressed format (cspace.compressed.100k.bin) with training data up August 2024 is available at Zenodo (Fondazione The Microsoft Research - COSBI 2025). Alongside with CSpace, in the same location there is the co-occurrence model (cspace.bigrams.pkl) and concept-frequency model (cspace.dict.pkl). See the examples folder in the GitHub repository for a complete example on how to use them, and the preprocessing folder for the code to build the CSpace training data."}], "state": "complete"}, {"id": "claim-target-result", "statement": "CSpace achieved superior concept-similarity performance to the compared word-embedding models while incurring less than a 5% sentence-similarity performance penalty relative to OpenAI ada-v2.", "kind": "quantitative", "target": {"value": 5.0, "unit": "per cent", "tolerance": null, "source": "According to the results in Table 7, the performance penalty is lower than 5%."}, "evidence": [{"asset": "source:pmid:40576218", "locator": "sections[0] \"Abstract\" paragraph 2", "quote": "In this work, we present CSpace: a concise word embedding of biomedical concepts that outperforms all alternatives in terms of out-of-vocabulary ratio and semantic textual similarity task, and has comparable performance with respect to transformer-based alternatives in the sentence similarity task."}, {"asset": "source:pmid:40576218", "locator": "sections[3] \"3 Results\" paragraph 22", "quote": "According to the results in Table 7, the performance penalty is lower than 5%."}], "state": "complete"}], "gaps": []}, "experiment": {"state": "partial", "objective": "CSpace achieved superior concept-similarity performance to the compared word-embedding models while incurring less than a 5% sentence-similarity performance penalty relative to OpenAI ada-v2.", "steps": [{"id": "read_published_estimate", "order": 1, "name": "Read the published estimate from the source", "kind": "other", "state": "complete", "blocked_on": null, "description": "The claim, method and evidence span published with the source article.", "duration_hours": null, "parameters": {}, "inputs": ["source:pmid:40576218"], "outputs": ["the published estimate and its evidence"], "instruments": [], "reagents": []}, {"id": "obtain_source_data", "order": 2, "name": "Obtain the source data from the named repository", "kind": "data-preparation", "state": "blocked", "blocked_on": "data:accession_absent", "description": "The GEO series accession resolves and its sample accessions are listed. This lane did not download the dataset or read its licence.", "duration_hours": null, "parameters": {}, "inputs": [], "outputs": ["the source data with its repository provenance"], "instruments": [], "reagents": []}, {"id": "run_analysis", "order": 3, "name": "Run the published analysis", "kind": "analysis", "state": "complete", "blocked_on": null, "description": "Apply the paper's statistical method to the resolved series and compare the result with the published target.", "duration_hours": null, "parameters": {}, "inputs": [], "outputs": ["the replicated result and its comparison"], "instruments": [], "reagents": []}], "materials": [], "instruments": [], "reagents": [], "conditions": {}, "gaps": ["step:obtain_source_data blocked on data:accession_absent"]}, "data": {"state": "partial", "assets": [{"id": "source:pmid:40576218", "role": "manuscript", "format": "xml", "media_type": "application/xml", "uri": "https://doi.org/10.1093/bioinformatics/btaf376", "sha256": "sha256:97fbe9e3d5a71f1932a86b81e937bb7d95b79f16903c2d045d7a4f9971324399", "bytes": 36675, "access": "open", "state": "resolved", "reader": null, "reader_status": "reader_required", "identifier": "10.1093/bioinformatics/btaf376", "version": null, "note": "Full article fetched from PMC; the checksum covers the rendered section body, not the original XML bytes.", "structure": {"pmcid": "PMC12275461", "source": "europe_pmc_fulltext", "coverage": "whole_body", "text_hash_scope": "rendered_section_body"}}], "gaps": []}, "analysis": {"state": "partial", "methods": [{"id": "method-1", "family": null, "specific": null, "label": "Pearson’s correlation coefficient between model-computed cosine similarities and human-expert similarity scores", "dsl_operations": [], "missing_operations": [], "expressible": null, "parameters": {}, "data_shapes": [], "evidence": {"score": 1.0, "band": "verbatim", "quote": "The testing procedure includes three steps: (i) compute the cosine similarity score for every combination of term pairs and ontology concept pairs, where ontology concepts are identified by their UMLS Concept Unique Identifiers (CUIs), (ii) measure the Pearson’s correlation coefficient between the computed scores and those provided by human experts, and (iii) compare results with other state-of-the-art embeddings (Zhang et al. 2019), (Noh and Kavuluru 2021)."}}], "targets": [{"id": "target-published-estimate", "claim_id": "claim-target-result", "operation": null, "expected_value": 5.0, "unit": "per cent", "tolerance": null, "state": "complete", "note": "The published estimate as the reader quoted it from the paper. No tolerance is published because the paper states none."}], "gaps": ["method.reader_named_unmapped"]}, "capability": {"state": "complete", "requirements": [{"capability_id": "capability:computational_analysis", "label": "Computational analysis platform", "assay_families": ["meta_analysis", "machine_learning", "survival_analysis", "statistical_regression", "pca", "molecular_docking", "network_analysis", "differential_expression", "gwas"], "availability_state": "owned", "tier_unlocked": "R1", "candidate": "Automated Assays Laravel application and simulated adapters"}], "gaps": []}, "readiness": {"state": "blocked", "tier": "R4", "blocking_axis": "data", "blocked_on": "data:accession_absent", "axes": {"method": {"tier": "R4", "expressible": false, "required_operations": [], "missing_operations": [], "basis": "The reader quoted the method the paper names (Pearson’s correlation coefficient between model-computed cosine similarities and human-expert similarity scores), but the corpus's method extractor could not map it onto a DSL operation, so the method is published but not expressible."}, "capability": {"tier": "R1", "capability_id": "capability:computational_analysis", "availability_state": "owned", "basis": "The computational analysis is owned. The physical bench the paper used is not, so a physical validation cannot be run here."}, "data": {"tier": "R4", "reachable": false, "accessions": [], "resolved_accessions": [], "statements": ["The code used for data pre-processing, training, visualization and evaluation of CSpace embeddings, all the experiments results and test benchmarks are available at https://doi.org/10.5281/zenodo.15125706 and in the GitHub repository https://github.com/cosbi-research/cspace; the PubMed data is available at https://www.nlm.nih.gov/databases/download/pubmed_medline.html, the PubMedCentral data is subdivided into manuscript collection available at ftp://ftp.ncbi.nlm.nih.gov/pub/pmc/manuscript, and open access commercial and non-commercial use (oa_comm and oa_noncomm) available at ftp://ftp.ncbi.nlm.nih.gov/pub/pmc, while the Pubtator3 data can be found in https://ftp.ncbi.nlm.nih.gov/pub/lu/PubTator3/ Bio and Med archives can be found on the S3 bucket https://www.biorxiv.org/tdm. The US clinical trials database is available at https://aact.ctti-clinicaltrials.org/snapshots directly as SQL data dumps. CSpace model in KeyedVectors (cspace.kv.bin) and compressed format (cspace.compressed.100k.bin) with training data up August 2024 is available at Zenodo (Fondazione The Microsoft Research - COSBI 2025). Alongside with CSpace, in the same location there is the co-occurrence model (cspace.bigrams.pkl) and concept-frequency model (cspace.dict.pkl). See the examples folder in the GitHub repository for a complete example on how to use them, and the preprocessing folder for the code to build the CSpace training data."], "basis": "The paper names no accession to its source data, so the data has to come from the authors."}}, "can_run": {"dsl_operations": [], "assets": ["source:pmid:40576218"], "steps": ["run_analysis"]}, "must_build": {"dsl_operations": [], "converters": [], "readers": [], "capabilities": [], "data_gaps": ["data:geo:None"]}, "gaps": ["data:accession_absent"]}, "review": {"state": "partial", "machine_critique": {"kind": "extraction_review", "reader": {"model": "MiniMax-M3.1-Flash", "provider": "minimax", "provider_family": "minimax", "read_at": "2026-11-01T19:55:01Z", "independence": "single_reader"}, "evidence_check": {"method": "verbatim", "availability": "verbatim"}, "calls": 5, "usage": {"prompt_tokens": 9908, "completion_tokens": 3099, "cached_tokens": 128}}, "human_critique": null, "stages": [{"id": "extraction", "state": "complete", "reader": "MiniMax-M3.1-Flash", "provider": "minimax"}], "gaps": ["review.independent_reader", "review.machine_critique", "review.human_critique"]}, "execution": {"state": "absent", "status": "not_run", "mode": null, "runs": [], "gaps": ["execution.not_run"]}, "provenance": {"entities": [{"id": "pmid:40576218", "type": "paper", "label": "pmid:40576218"}, {"id": "source:pmid:40576218", "type": "manuscript", "label": "PMID 40576218 full text"}, {"id": "data:geo:None", "type": "dataset", "label": "GEO None"}], "activities": [{"id": "activity-reading-sci-2026-11-01", "type": "classification", "used": [{"id": "pmid:40576218", "type": "paper", "label": "pmid:40576218"}, {"id": "source:pmid:40576218", "type": "manuscript", "label": "PMID 40576218 full text"}, {"id": "data:geo:None", "type": "dataset", "label": "GEO None"}], "agent": "agent:corpus-assay-engine"}, {"id": "activity-read-40576218", "type": "extraction", "used": ["source:pmid:40576218"], "agent": "agent:minimax-reader"}, {"id": "activity-resolve-None", "type": "accession_resolution", "used": ["data:geo:None"], "agent": "agent:corpus-assay-engine"}], "agents": [{"id": "agent:corpus-assay-engine", "type": "software", "label": "Corpus assay engine"}, {"id": "agent:minimax-reader", "type": "software", "label": "MiniMax literature reader"}], "source_run_id": "reading-sci-2026-11-01", "dsl_catalogue_fingerprint": "343fb30756ba6ec3", "capability_table_fingerprint": "2cb3e42f047d2ecf", "snapshot_id": "ba34b49490d9bfaf"}, "lineage": {"parent_id": "pmid:40576218", "parent_hash": "sha256:e200f47b0800fa1a787674854cf17204be00769cee9bd6394debf7d5eaae09c4", "fork_of": null, "generation": 1}, "gaps": ["execution.not_run", "method.reader_named_unmapped", "review.human_critique", "review.independent_reader", "review.machine_critique", "step:obtain_source_data blocked on data:accession_absent"]}