import z from "zod"import { Tool } from "../tool" const TIMEOUT = 30_000 async function fetchJSON(url: string, init?: RequestInit): Promise<any> {  const controller = new AbortController()  const timer = setTimeout(() => controller.abort(), TIMEOUT)  try {    const res = await fetch(url, {      ...init,      signal: controller.signal,      headers: { Accept: "application/json", "User-Agent": "openscience/biology", ...init?.headers },    })    if (!res.ok) throw new Error(`HTTP ${res.status}: ${await res.text().catch(() => res.statusText)}`)    return res.json()  } finally {    clearTimeout(timer)  }} async function fetchText(url: string): Promise<string> {  const controller = new AbortController()  const timer = setTimeout(() => controller.abort(), TIMEOUT)  try {    const res = await fetch(url, {      signal: controller.signal,      headers: { "User-Agent": "openscience/biology" },    })    if (!res.ok) throw new Error(`HTTP ${res.status}: ${await res.text().catch(() => res.statusText)}`)    return res.text()  } finally {    clearTimeout(timer)  }} // ── UniProt ────────────────────────────────────────────────────────────────── export const QueryUniprotTool = Tool.define("query_uniprot", {  description: [    "Query UniProt protein database. Returns protein function, GO terms, domains, pathways.",    "Input: gene name (TP53), UniProt accession (P04637), or search query.",    "Use for: protein annotation, function lookup, GO terms, domain identification.",  ].join("\n"),  parameters: z.object({    query: z      .string()      .describe("Gene name, UniProt ID, or search query (e.g. 'TP53', 'P04637', 'kinase AND organism_id:9606')"),    organism: z.string().default("9606").describe("NCBI taxon ID (default: 9606/human). Use '*' for all organisms."),    limit: z.number().default(5).describe("Max results (1-25)"),  }),  async execute(params, _ctx) {    const limit = Math.min(Math.max(params.limit, 1), 25)    const orgFilter = params.organism === "*" ? "" : `+AND+organism_id:${params.organism}`    const isAccession = /^[A-Z][0-9][A-Z0-9]{3}[0-9]$/i.test(params.query.trim())    const q = isAccession ? `accession:${params.query.trim()}` : params.query    const fields =      "accession,gene_names,protein_name,organism_name,cc_function,go_p,go_c,go_f,ft_domain,cc_pathway,length"    const url = `https://rest.uniprot.org/uniprotkb/search?query=${encodeURIComponent(q)}${orgFilter}&fields=${fields}&format=json&size=${limit}`     const data = await fetchJSON(url)    if (!data.results?.length) {      return { title: "UniProt", output: `No results for "${params.query}"`, metadata: {} }    }     const entries = data.results.map((e: any) => {      const lines: string[] = []      const name =        e.proteinDescription?.recommendedName?.fullName?.value ||        e.proteinDescription?.submissionNames?.[0]?.fullName?.value ||        "Unknown"      lines.push(`## ${e.primaryAccession} — ${name}`)       const genes = e.genes?.flatMap((g: any) =>        [g.geneName?.value, ...(g.synonyms?.map((s: any) => s.value) || [])].filter(Boolean),      )      if (genes?.length) lines.push(`**Genes**: ${genes.join(", ")}`)      if (e.organism) lines.push(`**Organism**: ${e.organism.scientificName}`)       const fns = e.comments        ?.filter((c: any) => c.commentType === "FUNCTION")        .flatMap((c: any) => c.texts?.map((t: any) => t.value) || [])      if (fns?.length) lines.push(`**Function**: ${fns.join(" ")}`)       const domains = e.features?.filter((f: any) => f.type === "Domain")      if (domains?.length) {        lines.push(          `**Domains**: ${domains.map((d: any) => `${d.description} (${d.location?.start?.value}-${d.location?.end?.value})`).join(", ")}`,        )      }       const pathways = e.comments        ?.filter((c: any) => c.commentType === "PATHWAY")        .flatMap((c: any) => c.texts?.map((t: any) => t.value) || [])      if (pathways?.length) lines.push(`**Pathways**: ${pathways.join("; ")}`)       for (const [label, prefix] of [        ["Biological Process", "P:"],        ["Molecular Function", "F:"],        ["Cellular Component", "C:"],      ] as const) {        const terms = e.uniProtKBCrossReferences          ?.filter((x: any) => x.database === "GO")          .filter((x: any) => x.properties?.some((p: any) => p.value?.startsWith(prefix)))          .map((x: any) => x.properties?.find((p: any) => p.value?.startsWith(prefix))?.value?.slice(2))          .filter(Boolean)        if (terms?.length)          lines.push(            `**GO ${label}**: ${terms.slice(0, 8).join("; ")}${terms.length > 8 ? ` (+${terms.length - 8} more)` : ""}`,          )      }       if (e.sequence?.length) lines.push(`**Length**: ${e.sequence.length} aa`)      return lines.join("\n")    })     return {      title: `UniProt: ${params.query}`,      output: entries.join("\n\n---\n\n"),      metadata: { count: data.results.length },    }  },}) // ── Ensembl ────────────────────────────────────────────────────────────────── export const QueryEnsemblTool = Tool.define("query_ensembl", {  description: [    "Query Ensembl for gene/transcript annotations. Returns coordinates, biotype, transcripts.",    "Input: gene symbol (TP53), Ensembl ID (ENSG00000141510), or transcript ID.",    "Use for: gene coordinates, biotype, ortholog lookup, transcript structure.",  ].join("\n"),  parameters: z.object({    query: z.string().describe("Gene symbol (TP53), Ensembl gene ID (ENSG...), or transcript ID (ENST...)"),    species: z.string().default("homo_sapiens").describe("Species (default: homo_sapiens)"),    expand: z.boolean().default(true).describe("Include transcripts in response"),  }),  async execute(params, _ctx) {    const q = params.query.trim()    const base = "https://rest.ensembl.org"    const expand = params.expand ? 1 : 0     const url = q.startsWith("ENS")      ? `${base}/lookup/id/${q}?expand=${expand}&content-type=application/json`      : `${base}/lookup/symbol/${params.species}/${q}?expand=${expand}&content-type=application/json`     const data = await fetchJSON(url)     const lines: string[] = []    lines.push(`## ${data.display_name || data.id} — ${data.description || "No description"}`)    lines.push(`**Ensembl ID**: ${data.id}`)    lines.push(`**Biotype**: ${data.biotype}`)    lines.push(`**Species**: ${data.species}`)    if (data.seq_region_name) {      lines.push(        `**Location**: chr${data.seq_region_name}:${data.start}-${data.end} (${data.strand === 1 ? "+" : "-"})`,      )    }    if (data.assembly_name) lines.push(`**Assembly**: ${data.assembly_name}`)     if (data.Transcript?.length) {      lines.push(`\n**Transcripts** (${data.Transcript.length}):`)      for (const t of data.Transcript.slice(0, 10)) {        const tag = t.is_canonical ? " [canonical]" : ""        lines.push(`- ${t.id} (${t.biotype})${tag}: ${t.start}-${t.end}, ${t.length || "?"} bp`)      }      if (data.Transcript.length > 10) {        lines.push(`  ... +${data.Transcript.length - 10} more transcripts`)      }    }     return {      title: `Ensembl: ${q}`,      output: lines.join("\n"),      metadata: { id: data.id, biotype: data.biotype },    }  },}) // ── KEGG ───────────────────────────────────────────────────────────────────── export const QueryKeggTool = Tool.define("query_kegg", {  description: [    "Query KEGG for pathway information and gene-pathway mappings.",    "Input: pathway ID (hsa04110), gene ID (hsa:7157), or keyword search.",    "Use for: pathway enrichment context, metabolic pathway lookup, gene-pathway membership.",  ].join("\n"),  parameters: z.object({    query: z.string().describe("Pathway ID (hsa04110), gene ID (hsa:7157), or search keyword (e.g. 'cell cycle')"),    operation: z      .enum(["info", "find", "genes"])      .default("find")      .describe("info: details for a pathway/entry. find: search by keyword. genes: list genes in a pathway."),    organism: z.string().default("hsa").describe("KEGG organism code (default: hsa for human)"),  }),  async execute(params, _ctx) {    const base = "https://rest.kegg.jp"    const q = params.query.trim()     if (params.operation === "info") {      const text = await fetchText(`${base}/get/${q}`)      return { title: `KEGG: ${q}`, output: text.slice(0, 8000), metadata: {} as Record<string, any> }    }     if (params.operation === "genes") {      const text = await fetchText(`${base}/link/${params.organism}/${q}`)      const genes = text        .trim()        .split("\n")        .filter(Boolean)        .map((line) => {          const parts = line.split("\t")          return parts[1] || line        })      return {        title: `KEGG genes: ${q}`,        output: `**Genes in ${q}** (${genes.length}):\n${genes.join(", ")}`,        metadata: { count: genes.length } as Record<string, any>,      }    }     // find    const text = await fetchText(`${base}/find/pathway/${encodeURIComponent(q)}`)    const results = text      .trim()      .split("\n")      .filter(Boolean)      .map((line) => {        const [id, ...rest] = line.split("\t")        return `- **${id}**: ${rest.join(" ")}`      })     if (!results.length) {      return { title: "KEGG", output: `No pathways found for "${q}"`, metadata: {} as Record<string, any> }    }     return {      title: `KEGG: ${q}`,      output: `**Pathway search: "${q}"** (${results.length} results):\n${results.join("\n")}`,      metadata: { count: results.length } as Record<string, any>,    }  },}) // ── PubMed ─────────────────────────────────────────────────────────────────── export const QueryPubmedTool = Tool.define("query_pubmed", {  description: [    "Search PubMed for biomedical literature. Returns titles, authors, abstracts, PMIDs.",    "Input: PubMed query (supports Boolean operators, [MeSH], [Author] qualifiers).",    "Use for: literature search, finding methods papers, validating findings against published work.",  ].join("\n"),  parameters: z.object({    query: z.string().describe("PubMed search query (e.g. 'TP53 AND breast cancer AND DESeq2')"),    max_results: z.number().default(5).describe("Max papers (1-20)"),    sort: z.enum(["relevance", "date"]).default("relevance").describe("Sort order"),  }),  async execute(params, _ctx) {    const limit = Math.min(Math.max(params.max_results, 1), 20)    const base = "https://eutils.ncbi.nlm.nih.gov/entrez/eutils"     // Search for IDs    const search = await fetchJSON(      `${base}/esearch.fcgi?db=pubmed&term=${encodeURIComponent(params.query)}&retmode=json&retmax=${limit}&sort=${params.sort}`,    )    const ids = search.esearchresult?.idlist     if (!ids?.length) {      return { title: "PubMed", output: `No results for "${params.query}"`, metadata: {} }    }     // Fetch summaries    const summary = await fetchJSON(`${base}/esummary.fcgi?db=pubmed&id=${ids.join(",")}&retmode=json`)     const articles = ids.map((id: string) => {      const a = summary.result?.[id]      if (!a) return `### PMID:${id}\n(no details)`       const authors = a.authors        ?.slice(0, 3)        .map((x: any) => x.name)        .join(", ")      const authorStr = a.authors?.length > 3 ? `${authors} et al.` : authors || "Unknown"      const lines: string[] = []      lines.push(`### PMID:${id}`)      lines.push(`**${a.title || "Untitled"}**`)      lines.push(`${authorStr} — *${a.source || "?"}* (${a.pubdate || "?"})`)      if (a.elocationid) lines.push(`DOI: ${a.elocationid}`)      return lines.join("\n")    })     // Fetch plain-text abstracts    const abstracts = await fetchText(`${base}/efetch.fcgi?db=pubmed&id=${ids.join(",")}&rettype=abstract&retmode=text`)     return {      title: `PubMed: ${params.query}`,      output: [        `**PubMed**: "${params.query}" (${search.esearchresult?.count || "?"} total, showing ${ids.length})`,        "",        articles.join("\n\n"),        "",        "---",        "",        "**Abstracts**:",        abstracts.slice(0, 6000),      ].join("\n"),      metadata: { count: ids.length, total: search.esearchresult?.count },    }  },}) // ── NCBI Gene ──────────────────────────────────────────────────────────────── export const QueryNcbiGeneTool = Tool.define("query_ncbi_gene", {  description: [    "Query NCBI Gene for gene information. Returns summary, aliases, genomic location, RefSeqs.",    "Input: gene symbol (TP53), NCBI Gene ID (7157), or search query.",    "Use for: gene function summary, alternative names, genomic context, cross-database IDs.",  ].join("\n"),  parameters: z.object({    query: z.string().describe("Gene symbol (TP53), NCBI Gene ID (7157), or search query"),    organism: z.string().default("human").describe("Organism filter (default: human)"),    limit: z.number().default(5).describe("Max results (1-10)"),  }),  async execute(params, _ctx) {    const limit = Math.min(Math.max(params.limit, 1), 10)    const base = "https://eutils.ncbi.nlm.nih.gov/entrez/eutils"    const q = params.query.trim()     const isId = /^\d+$/.test(q)    const ids = isId      ? [q]      : await fetchJSON(          `${base}/esearch.fcgi?db=gene&term=${encodeURIComponent(q)}[Gene]+AND+${encodeURIComponent(params.organism)}[Organism]&retmode=json&retmax=${limit}`,        ).then((d: any) => d.esearchresult?.idlist || [])     if (!ids.length) {      return { title: "NCBI Gene", output: `No genes found for "${q}"`, metadata: {} }    }     const summary = await fetchJSON(`${base}/esummary.fcgi?db=gene&id=${ids.join(",")}&retmode=json`)     const entries = ids.map((id: string) => {      const g = summary.result?.[id]      if (!g) return `- Gene ID ${id}: no details`       const lines: string[] = []      lines.push(`## ${g.name || id} — ${g.description || "?"}`)      lines.push(`**Gene ID**: ${id}`)      if (g.otheraliases) lines.push(`**Aliases**: ${g.otheraliases}`)      if (g.organism?.scientificname) lines.push(`**Organism**: ${g.organism.scientificname}`)      if (g.chromosome) lines.push(`**Chromosome**: ${g.chromosome}`)      if (g.maplocation) lines.push(`**Map location**: ${g.maplocation}`)      if (g.genomicinfo?.[0]) {        const gi = g.genomicinfo[0]        lines.push(`**Coordinates**: chr${gi.chrloc}:${gi.chrstart}-${gi.chrstop}`)      }      if (g.summary) lines.push(`**Summary**: ${g.summary}`)      if (g.nomenclaturename) lines.push(`**Full name**: ${g.nomenclaturename}`)      return lines.join("\n")    })     return {      title: `NCBI Gene: ${q}`,      output: entries.join("\n\n---\n\n"),      metadata: { count: ids.length },    }  },}) // ── STRING ─────────────────────────────────────────────────────────────────── export const QueryStringTool = Tool.define("query_string", {  description: [    "Query STRING database for protein-protein interactions and functional networks.",    "Input: protein/gene name(s). Returns interaction partners with confidence scores.",    "Use for: interaction networks, functional associations, co-expression evidence.",  ].join("\n"),  parameters: z.object({    identifiers: z      .string()      .describe("Protein/gene name(s), comma-separated for multiple (e.g. 'TP53' or 'TP53,MDM2,CDKN1A')"),    species: z.number().default(9606).describe("NCBI taxon ID (default: 9606/human)"),    limit: z.number().default(10).describe("Max interaction partners per protein (1-50)"),    score_threshold: z      .number()      .default(400)      .describe("Minimum combined score 0-1000 (default: 400 = medium confidence)"),  }),  async execute(params, _ctx) {    const ids = params.identifiers.trim()    const limit = Math.min(Math.max(params.limit, 1), 50)     const url = `https://string-db.org/api/json/interaction_partners?identifiers=${encodeURIComponent(ids)}&species=${params.species}&limit=${limit}&required_score=${params.score_threshold}`    const data = await fetchJSON(url)     if (!data?.length) {      return {        title: "STRING",        output: `No interactions found for "${ids}" (score >= ${params.score_threshold})`,        metadata: {},      }    }     // Group by query protein    const grouped: Record<string, any[]> = {}    for (const i of data) {      const key = i.preferredName_A || i.stringId_A      if (!grouped[key]) grouped[key] = []      grouped[key].push(i)    }     const sections = Object.entries(grouped).map(([protein, interactions]) => {      const lines: string[] = [`## ${protein} interactions (${interactions.length})`]      const sorted = interactions.sort((a: any, b: any) => b.score - a.score)      for (const i of sorted) {        const partner = i.preferredName_B || i.stringId_B        const scores = []        if (i.escore > 0) scores.push(`exp:${i.escore.toFixed(3)}`)        if (i.dscore > 0) scores.push(`db:${i.dscore.toFixed(3)}`)        if (i.tscore > 0) scores.push(`txt:${i.tscore.toFixed(3)}`)        lines.push(`- **${partner}** (score: ${i.score}) ${scores.length ? `[${scores.join(", ")}]` : ""}`)      }      return lines.join("\n")    })     return {      title: `STRING: ${ids}`,      output: sections.join("\n\n"),      metadata: { interactions: data.length },    }  },}) // ── PDB ────────────────────────────────────────────────────────────────────── export const QueryPdbTool = Tool.define("query_pdb", {  description: [    "Query RCSB PDB for protein 3D structure information.",    "Input: PDB ID (1TUP) for direct lookup, or protein name for search.",    "Use for: structure resolution, experimental method, polymer entities, ligands.",  ].join("\n"),  parameters: z.object({    query: z.string().describe("PDB ID (e.g. '1TUP') or protein name for search (e.g. 'p53 DNA-binding domain')"),    limit: z.number().default(5).describe("Max results for text search (1-10)"),  }),  async execute(params, _ctx) {    const q = params.query.trim()    const isPdbId = /^[0-9][A-Za-z0-9]{3}$/.test(q)     if (isPdbId) {      const id = q.toUpperCase()      const data = await fetchJSON(`https://data.rcsb.org/rest/v1/core/entry/${id}`)       const lines: string[] = []      lines.push(`## PDB: ${id}`)      if (data.struct?.title) lines.push(`**Title**: ${data.struct.title}`)      if (data.exptl?.[0]) lines.push(`**Method**: ${data.exptl[0].method}`)      if (data.rcsb_entry_info?.resolution_combined?.[0]) {        lines.push(`**Resolution**: ${data.rcsb_entry_info.resolution_combined[0]} A`)      }      if (data.rcsb_entry_info?.deposited_atom_count) {        lines.push(`**Atoms**: ${data.rcsb_entry_info.deposited_atom_count}`)      }       const entities = data.rcsb_entry_container_identifiers?.polymer_entity_ids || []      if (entities.length) {        lines.push(`\n**Polymer entities** (${entities.length}):`)        for (const entityId of entities.slice(0, 5)) {          try {            const entity = await fetchJSON(`https://data.rcsb.org/rest/v1/core/polymer_entity/${id}/${entityId}`)            if (entity.rcsb_polymer_entity?.pdbx_description) {              lines.push(`- Entity ${entityId}: ${entity.rcsb_polymer_entity.pdbx_description}`)            }          } catch {            /* skip */          }        }      }       return { title: `PDB: ${id}`, output: lines.join("\n"), metadata: { pdb_id: id } as Record<string, any> }    }     // Text search    const limit = Math.min(Math.max(params.limit, 1), 10)    const body = {      query: {        type: "terminal",        service: "full_text",        parameters: { value: q },      },      return_type: "entry",      request_options: {        results_content_type: ["experimental"],        paginate: { start: 0, rows: limit },      },    }     const data = await fetchJSON("https://search.rcsb.org/rcsbsearch/v2/query", {      method: "POST",

Lines 1–500 of 518. Load more