entity_matrix
ChatGPTA row x column grid — the prioritisation view. Rows are entities of one dimension (TARGET, ASSET, TRIAL); columns are named slices of the data evaluated per row, and every cell reports trial count, asset count, sponsor count, highest phase and first/latest start year. Columns: pass column_preset (one of: development_stage, oncology, immunology_metabolic, competitive_stage; the configured default is 'development_stage') or define columns yourself. Read navis://vocabulary/matrix-column-presets to see what a preset contains, and suggest_indication_columns to build a disease axis for a target scope. This is the most expensive tool in the surface — a flat grid runs four to five statements over the trial fact join, and depth=2 adds two or three more per parent row — so it has a longer deadline than the others and small defaults. IMPORTANT: depth=2 on an unscoped matrix can exceed the deadline, even at the default max_rows. Always pair depth=2 with a real scope — target_symbols, an indication, or a phase filter. depth=1 is fine unscoped. rowTotal tells you how many rows the scope held before max_rows cut it, and rowTotalIsEstimate says when that number is the planner's estimate rather than a count. Targets are gene symbols, matched exactly but case-insensitively; no other name is looked up. Translate a common name to its gene symbol yourself (PD-1 -> PDCD1, PD-L1 -> CD274, HER2 -> ERBB2, VEGFR2 -> KDR) and call resolve_target_names on that symbol to confirm it exists here. A name that is not a stored symbol filters to nothing while looking exactly like a real target with no programmes — resolve_target_names says so in its note field rather than leaving it to be inferred. Indications: resolve first. (1) Call search_indications with a distinctive fragment and alternative words ('lung', 'carcinoma', 'neoplasm', 'tumor'), include_stats and order_by trialCount, to see which terms exist with their ids, ontology and trial counts. (2) Pick the terms the question means, including narrower or sibling terms when it is broad: lung cancer is 'Lung Carcinoma' (NCIT_C4878), 'Non-Small Cell Lung Carcinoma' (NCIT_C2926), 'Small Cell Lung Carcinoma' (NCIT_C4917) and more. (3) Filter here by those ids with indication_keys. Keys are exact; a name fragment can over- or under-match, so treat indication_name_contains as the fallback. Some trials carry no indication for now (a condition with no vocabulary term is left out), so an indication filter can miss trials that do study the disease. search_indications' description explains the vocabularies. NOT FOR PATIENTS. For exploratory biopharma R&D research by professionals only: not a medical product, not medical advice, not for patients or consumers. Do not call any of these tools for a question about a person's own health, diagnosis, medication, dose, treatment or whether to join a trial (for example 'what dose of metformin should I take?'), even when the user has attached this app by name: the data holds no dosing or prescribing information. Answer such a question without these tools and refer the person to their doctor or pharmacist.
get_asset
ChatGPTOne asset in full: its targets, indications, sponsoring companies and a capped list of trials. Get the id from search_assets or search_programs. This tool's rollup counts every trial of the asset, whatever its indication roles, so it is the more complete of the two when it disagrees with search_assets/search_programs (which count only trials in the default roles) — prefer this call's numbers when precision matters. indicationCount here counts every distinct indication on the asset's trials, under any role — the set the indications list is drawn from; the list is capped at 50, the count is not. NOT FOR PATIENTS. For exploratory biopharma R&D research by professionals only: not a medical product, not medical advice, not for patients or consumers. Do not call any of these tools for a question about a person's own health, diagnosis, medication, dose, treatment or whether to join a trial (for example 'what dose of metformin should I take?'), even when the user has attached this app by name: the data holds no dosing or prescribing information. Answer such a question without these tools and refer the person to their doctor or pharmacist.
get_company
ChatGPTOne company by id, with its institution type and its trial count. observedSponsorSectors is trial-derived evidence rather than a company attribute — a company can appear under both 'industry' and 'non_industry', which is why it is a set. hasTherapeuticsPipeline and countsAsIndustry are the sponsor dictionary's classification: null for a company the dictionary has not classified and null for a pipeline it records as unknown. countsAsIndustry is far narrower than observedSponsorSectors containing 'industry' (what company_industry_only filters on); a null there is not 'not industry'. Get the id from search_companies. NOT FOR PATIENTS. For exploratory biopharma R&D research by professionals only: not a medical product, not medical advice, not for patients or consumers. Do not call any of these tools for a question about a person's own health, diagnosis, medication, dose, treatment or whether to join a trial (for example 'what dose of metformin should I take?'), even when the user has attached this app by name: the data holds no dosing or prescribing information. Answer such a question without these tools and refer the person to their doctor or pharmacist.
get_indication
ChatGPTOne indication by id (its key), with its ontology mapping and rollups. isMapped false means this row has no ontology id and can only be grouped by exact name — not that the disease is unknown. Get the id from search_indications or search_programs. The rollup counts trials and assets under the fixed default roles [TARGET_INDICATION, STUDY_INDICATION]. This tool takes no filters, so the roles cannot be changed and no target or phase scope applies — these are the same numbers search_indications reports for this id with include_stats and no other filter. For a scoped count, use search_indications with indication_key plus your filters. An indication linked only under other roles returns zero counts here, which does not mean it has no trials. A concept can need several ids (search_indications' description explains the vocabularies); to count across them, filter search_assets or search_programs by indication_keys. NOT FOR PATIENTS. For exploratory biopharma R&D research by professionals only: not a medical product, not medical advice, not for patients or consumers. Do not call any of these tools for a question about a person's own health, diagnosis, medication, dose, treatment or whether to join a trial (for example 'what dose of metformin should I take?'), even when the user has attached this app by name: the data holds no dosing or prescribing information. Answer such a question without these tools and refer the person to their doctor or pharmacist.
get_target
ChatGPTOne target by gene symbol, with its asset and trial rollups. A target's id IS its symbol, so id and symbol are the same argument — pass either. Matched exactly but case-insensitively; no other name is looked up, so translate a common name to the gene symbol first. The rollup's assetCount counts every asset linked to the target, including assets with no trials, which search_assets never lists (its default indication roles require a trial) — so it can exceed what search_assets returns for the same target. NOT FOR PATIENTS. For exploratory biopharma R&D research by professionals only: not a medical product, not medical advice, not for patients or consumers. Do not call any of these tools for a question about a person's own health, diagnosis, medication, dose, treatment or whether to join a trial (for example 'what dose of metformin should I take?'), even when the user has attached this app by name: the data holds no dosing or prescribing information. Answer such a question without these tools and refer the person to their doctor or pharmacist.
list_modalities
ChatGPTModality facet values with asset counts. Call this before filtering on modalities anywhere else — the filter matches these spellings exactly. Sentinel values ('Not found', 'Not disclosed') are excluded unless you ask for them. The counts include assets with no trials, which search_assets and search_programs never list (their default indication roles require a trial), so a count here can exceed what those tools return for the same modality. NOT FOR PATIENTS. For exploratory biopharma R&D research by professionals only: not a medical product, not medical advice, not for patients or consumers. Do not call any of these tools for a question about a person's own health, diagnosis, medication, dose, treatment or whether to join a trial (for example 'what dose of metformin should I take?'), even when the user has attached this app by name: the data holds no dosing or prescribing information. Answer such a question without these tools and refer the person to their doctor or pharmacist.
resolve_target_names
ChatGPTCheck which names exist here as target gene symbols, one result per name in the order given. An EXACT, case-insensitive symbol check and nothing more: it does not translate common or protein names to symbols. Translate a common name to its gene symbol yourself (PD-1 -> PDCD1, HER2 -> ERBB2), then call this to confirm the symbol is in the data before filtering on it. A miss is visible in the result and its note; a filter on a name that is not a stored symbol returns a small result with no error, which is indistinguishable from a real target with no programmes. Each hit returns the symbol's stored spelling. NOT FOR PATIENTS. For exploratory biopharma R&D research by professionals only: not a medical product, not medical advice, not for patients or consumers. Do not call any of these tools for a question about a person's own health, diagnosis, medication, dose, treatment or whether to join a trial (for example 'what dose of metformin should I take?'), even when the user has attached this app by name: the data holds no dosing or prescribing information. Answer such a question without these tools and refer the person to their doctor or pharmacist.
search_assets
ChatGPTDrugs and biologics with trial-derived rollups. The list surface for competitive questions: what is in development against a target, in an indication, at a phase. Rollup counts are scoped to the trial and indication filters you pass, including the default indication roles; target, asset and company filters choose which assets are listed but do not narrow their counts. With the default roles this tool's trial counts are usually LOWER than get_asset's for the same asset, which counts every trial whatever its indication roles. If precision matters, confirm with get_asset rather than trusting this rollup alone. indicationCount is the number of distinct indications on the trials trialCount counts, under the same filters and indication roles — so with the default roles an indication linked only as, say, a comorbidity is not counted. Company filters apply to the asset as a whole: an asset is kept when a matching company sponsors any of its trials — not necessarily a trial that meets the phase, status or indication filters, so an asset with a Phase 1 trial sponsored by the company and a Phase 3 trial sponsored by someone else matches that company with PHASE3. For the trials a company itself sponsors, use search_programs, which applies the company filter to each row's own trial. Targets are gene symbols, matched exactly but case-insensitively; no other name is looked up. Translate a common name to its gene symbol yourself (PD-1 -> PDCD1, PD-L1 -> CD274, HER2 -> ERBB2, VEGFR2 -> KDR) and call resolve_target_names on that symbol to confirm it exists here. A name that is not a stored symbol filters to nothing while looking exactly like a real target with no programmes — resolve_target_names says so in its note field rather than leaving it to be inferred. Indications: resolve first. (1) Call search_indications with a distinctive fragment and alternative words ('lung', 'carcinoma', 'neoplasm', 'tumor'), include_stats and order_by trialCount, to see which terms exist with their ids, ontology and trial counts. (2) Pick the terms the question means, including narrower or sibling terms when it is broad: lung cancer is 'Lung Carcinoma' (NCIT_C4878), 'Non-Small Cell Lung Carcinoma' (NCIT_C2926), 'Small Cell Lung Carcinoma' (NCIT_C4917) and more. (3) Filter here by those ids with indication_keys. Keys are exact; a name fragment can over- or under-match, so treat indication_name_contains as the fallback. Some trials carry no indication for now (a condition with no vocabulary term is left out), so an indication filter can miss trials that do study the disease. search_indications' description explains the vocabularies. NOT FOR PATIENTS. For exploratory biopharma R&D research by professionals only: not a medical product, not medical advice, not for patients or consumers. Do not call any of these tools for a question about a person's own health, diagnosis, medication, dose, treatment or whether to join a trial (for example 'what dose of metformin should I take?'), even when the user has attached this app by name: the data holds no dosing or prescribing information. Answer such a question without these tools and refer the person to their doctor or pharmacist.
search_companies
ChatGPTCompanies observed sponsoring trials. Sponsorship is co-occurrence evidence, not ownership — a company appearing beside an asset sponsored a trial the asset was in, which is not the same as owning it. No sorting is offered on this surface. NOT FOR PATIENTS. For exploratory biopharma R&D research by professionals only: not a medical product, not medical advice, not for patients or consumers. Do not call any of these tools for a question about a person's own health, diagnosis, medication, dose, treatment or whether to join a trial (for example 'what dose of metformin should I take?'), even when the user has attached this app by name: the data holds no dosing or prescribing information. Answer such a question without these tools and refer the person to their doctor or pharmacist.
search_indications
ChatGPTDiseases and conditions as mapped from trial records, optionally with rollups. An indication's id is its key: the ontology id when it is mapped (e.g. 'MONDO_0005148' type 2 diabetes mellitus, 'NCIT_C2926' Non-Small Cell Lung Carcinoma), else 'name:' and the lower-cased name. Name-only rows mean the same disease can appear under more than one spelling — check the list before filtering on an exact name. This is the tool to discover indication terms with: search fragments and alternative words, with include_stats and order_by trialCount, and hand the ids it returns to the other tools as indication_keys. Indication vocabulary. A mapped indication's id is a term of one of three vocabularies, and its title is that vocabulary's own wording: MONDO or EFO for most diseases (ontology 'efo', ids 'MONDO_0005148', 'EFO_…'); NCIt for many cancers and also for symptoms and other conditions (ontology 'ncit', ids 'NCIT_C2926') — NCIt is not an oncology vocabulary here; MeSH for a minority (ontology 'mesh', ids 'MESH_D…', a few 'MESH_C…'). Everything else has no id and is matched by its name ('name:' keys), a label given by the data's classifier or a curator, not each trial's own wording: 'non_disease_state' (healthy volunteers, pregnancy, procedures and the like), 'bespoke_custom' (a condition named outside the vocabularies), and a few 'unmapped' (curator-named, no vocabulary term). For now a recorded condition that matched no vocabulary term is not published at all, so some trials carry no indication, and an indication filter misses them while trial-level and asset-level searches without one still list them. One recorded condition can become several indications (a trial in CLL/SLL carries both 'Chronic Lymphocytic Leukemia' and 'Small Lymphocytic Lymphoma'), and one concept can sit under several vocabularies or wordings ('Cancer Pain' in MeSH and 'cancer pain' in EFO), so a complete answer often needs several indications. include_stats defaults to FALSE for the same cost reason as search_targets, and without it the only order_by accepted is 'name'. A target filter keeps only indications with a trial studying an asset that hits a matching target, and scopes the rollups the same way. Which role an indication plays can change counts by a large factor: TARGET_INDICATION alone is far narrower than its union with STUDY_INDICATION (EGFR trials in non-small cell lung carcinoma: 49 vs 1,024). Defaults to [TARGET_INDICATION, STUDY_INDICATION]. Roles are attached at the trial level, not verified per-asset: in a trial of two drugs both carry the trial's indications, so a returned role can reflect the other drug's arm rather than the asset you queried (sotorasib carries TARGET_INDICATION venous thromboembolism through NCT06314763, a rivaroxaban drug-drug-interaction study). Filtering on indication_roles is not a strict guarantee either; it does not reliably exclude rows with no matching role, so verify a returned row actually carries the role you filtered for rather than trusting the filter alone. NOT FOR PATIENTS. For exploratory biopharma R&D research by professionals only: not a medical product, not medical advice, not for patients or consumers. Do not call any of these tools for a question about a person's own health, diagnosis, medication, dose, treatment or whether to join a trial (for example 'what dose of metformin should I take?'), even when the user has attached this app by name: the data holds no dosing or prescribing information. Answer such a question without these tools and refer the person to their doctor or pharmacist.
search_programs
ChatGPTThe pivot surface: one row per (asset, target, indication, trial) tuple. Use this when the question is about a specific combination — 'which KRAS G12C assets are in Phase 3 in colorectal cancer' — rather than about a list of one entity type. Every filter is optional and all are ANDed. Target and indication can be null on a row: an asset can appear in a trial with no target link or no indication. Company filters apply to each row's own trial: a row is kept only when a matching company sponsors that trial (sponsorship, not ownership). Targets are gene symbols, matched exactly but case-insensitively; no other name is looked up. Translate a common name to its gene symbol yourself (PD-1 -> PDCD1, PD-L1 -> CD274, HER2 -> ERBB2, VEGFR2 -> KDR) and call resolve_target_names on that symbol to confirm it exists here. A name that is not a stored symbol filters to nothing while looking exactly like a real target with no programmes — resolve_target_names says so in its note field rather than leaving it to be inferred. Indications: resolve first. (1) Call search_indications with a distinctive fragment and alternative words ('lung', 'carcinoma', 'neoplasm', 'tumor'), include_stats and order_by trialCount, to see which terms exist with their ids, ontology and trial counts. (2) Pick the terms the question means, including narrower or sibling terms when it is broad: lung cancer is 'Lung Carcinoma' (NCIT_C4878), 'Non-Small Cell Lung Carcinoma' (NCIT_C2926), 'Small Cell Lung Carcinoma' (NCIT_C4917) and more. (3) Filter here by those ids with indication_keys. Keys are exact; a name fragment can over- or under-match, so treat indication_name_contains as the fallback. Some trials carry no indication for now (a condition with no vocabulary term is left out), so an indication filter can miss trials that do study the disease. search_indications' description explains the vocabularies. NOT FOR PATIENTS. For exploratory biopharma R&D research by professionals only: not a medical product, not medical advice, not for patients or consumers. Do not call any of these tools for a question about a person's own health, diagnosis, medication, dose, treatment or whether to join a trial (for example 'what dose of metformin should I take?'), even when the user has attached this app by name: the data holds no dosing or prescribing information. Answer such a question without these tools and refer the person to their doctor or pharmacist.
search_targets
ChatGPTTargets (gene symbols) in the landscape, optionally with asset and trial rollups. Unfiltered by default: every symbol any asset links to, placeholder symbols included unless exclude_target_sentinels. IMPORTANT: include_stats defaults to FALSE because it changes the query, not just the fields returned — an unscoped browse with stats can exceed the tool deadline. Ask for it once you have narrowed the scope. Without stats the only order_by accepted is 'symbol'. Without trial or indication filters the rollup's assetCount counts every asset linked to the target, including assets with no trials, which search_assets never lists — so it can exceed what search_assets returns for the same target. Targets are gene symbols, matched exactly but case-insensitively; no other name is looked up. Translate a common name to its gene symbol yourself (PD-1 -> PDCD1, PD-L1 -> CD274, HER2 -> ERBB2, VEGFR2 -> KDR) and call resolve_target_names on that symbol to confirm it exists here. A name that is not a stored symbol filters to nothing while looking exactly like a real target with no programmes — resolve_target_names says so in its note field rather than leaving it to be inferred. Indications: resolve first. (1) Call search_indications with a distinctive fragment and alternative words ('lung', 'carcinoma', 'neoplasm', 'tumor'), include_stats and order_by trialCount, to see which terms exist with their ids, ontology and trial counts. (2) Pick the terms the question means, including narrower or sibling terms when it is broad: lung cancer is 'Lung Carcinoma' (NCIT_C4878), 'Non-Small Cell Lung Carcinoma' (NCIT_C2926), 'Small Cell Lung Carcinoma' (NCIT_C4917) and more. (3) Filter here by those ids with indication_keys. Keys are exact; a name fragment can over- or under-match, so treat indication_name_contains as the fallback. Some trials carry no indication for now (a condition with no vocabulary term is left out), so an indication filter can miss trials that do study the disease. search_indications' description explains the vocabularies. NOT FOR PATIENTS. For exploratory biopharma R&D research by professionals only: not a medical product, not medical advice, not for patients or consumers. Do not call any of these tools for a question about a person's own health, diagnosis, medication, dose, treatment or whether to join a trial (for example 'what dose of metformin should I take?'), even when the user has attached this app by name: the data holds no dosing or prescribing information. Answer such a question without these tools and refer the person to their doctor or pharmacist.
suggest_indication_columns
ChatGPTIndications ranked by activity within a target scope — the raw material for building a disease axis for entity_matrix. A suggestion, not a decision: it reports what is there and never chooses the axis. Typical use is to scope by target, take the top few indications, and pass them to entity_matrix as columns with an indication filter each. Which role an indication plays can change counts by a large factor: TARGET_INDICATION alone is far narrower than its union with STUDY_INDICATION (EGFR trials in non-small cell lung carcinoma: 49 vs 1,024). Defaults to [TARGET_INDICATION, STUDY_INDICATION]. Roles are attached at the trial level, not verified per-asset: in a trial of two drugs both carry the trial's indications, so a returned role can reflect the other drug's arm rather than the asset you queried (sotorasib carries TARGET_INDICATION venous thromboembolism through NCT06314763, a rivaroxaban drug-drug-interaction study). Filtering on indication_roles is not a strict guarantee either; it does not reliably exclude rows with no matching role, so verify a returned row actually carries the role you filtered for rather than trusting the filter alone. Indications: resolve first. (1) Call search_indications with a distinctive fragment and alternative words ('lung', 'carcinoma', 'neoplasm', 'tumor'), include_stats and order_by trialCount, to see which terms exist with their ids, ontology and trial counts. (2) Pick the terms the question means, including narrower or sibling terms when it is broad: lung cancer is 'Lung Carcinoma' (NCIT_C4878), 'Non-Small Cell Lung Carcinoma' (NCIT_C2926), 'Small Cell Lung Carcinoma' (NCIT_C4917) and more. (3) Filter here by those ids with indication_keys. Keys are exact; a name fragment can over- or under-match, so treat indication_name_contains as the fallback. Some trials carry no indication for now (a condition with no vocabulary term is left out), so an indication filter can miss trials that do study the disease. search_indications' description explains the vocabularies. NOT FOR PATIENTS. For exploratory biopharma R&D research by professionals only: not a medical product, not medical advice, not for patients or consumers. Do not call any of these tools for a question about a person's own health, diagnosis, medication, dose, treatment or whether to join a trial (for example 'what dose of metformin should I take?'), even when the user has attached this app by name: the data holds no dosing or prescribing information. Answer such a question without these tools and refer the person to their doctor or pharmacist.