The research agent for biology’s public data

Ask a question in plain language, or paste a gene, accession, sequence or genome coordinate. BioQ queries 15 public databases at once, from 5.7B metagenomic proteins and 149.4M UniProt entries to 5.2M binding measurements, and shows you the record behind every claim.

Free for researchers. Sign in with Google or your institutional email.

Read the API docs

Example answer from live data3 queries

Is KRAS a selective dependency in pancreatic cancer, and what has been measured binding it?

1dependency(gene=KRAS)0.2 msCRISPR dependency in 1,178 cancer cell lines
2binding_measurements(gene=KRAS)0.4 s6,796 binding measurements
3interactions(KRAS, min_score=900)0.2 ms98 high-confidence partners

Yes. 43 of 47 pancreatic cancer cell lines depend on KRAS (mean gene effect −1.79), the highest share of any lineage 1. Across all lineages, 399 of 1,178 lines are dependent 1. BioQ holds 6,796 binding measurements for KRAS; the most potent is MRTX-1133, IC50 0.2 pM 2. Its top interaction partners include RAF1, SOS1 and BRAF, each scored 999 of 1,000 by STRING 3.

Limit: dependency is measured in cell lines, not in patients.

What BioQ reads

Live record counts from the engine, refreshed every five minutes.

MGnify proteinsMVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSHGSAQVKGHGKKVADALTNAVAHVDDMPNALSALSDLHAHKLRVDPVNFKLLSHCLLVTLAAHLPAEFTPAVHASLDKFLASVSTVLTSK5.7B

Pfam-annotated proteinsVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSHGSAQVKGHGKKVADALTNAVAHVDDMPNALSALSDLHAHKLRVDPVNFKLLSHCLLVTLAAHLPAEFTPAVHASLDKFLASVSTVLTSKYRMVLSPADKTNVKAAW3B

MGnify clustersFLSFPTTKTYFPHFDLSHGSAQVKGHGKKVADALTNAVAHVDDMPNALSALSDLHAHKLRVDPVNFKLLSHCLLVTLAAHLPAEFTPAVHASLDKFLASVSTVLTSKYRMVLSPADKTNVKAAWGKVGAHAGEYGAEALE1.7B

STRING interactionsGSAQVKGHGKKVADALTNAVAHVDDMPNALSALSDLHAHKLRVDPVNFKLLSHCLLVTLAAHLPAEFTPAVHASLDKFLASVSTVLTSKYRMVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDL162.7M

UniProt TrEMBLVAHVDDMPNALSALSDLHAHKLRVDPVNFKLLSHCLLVTLAAHLPAEFTPAVHASLDKFLASVSTVLTSKYRMVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSHGSAQVKGHGKKVADALT149.4M

Binding measurementsAHKLRVDPVNFKLLSHCLLVTLAAHLPAEFTPAVHASLDKFLASVSTVLTSKYRMVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSHGSAQVKGHGKKVADALTNAVAHVDDMPNALSALSD5.2M

StructuresCLLVTLAAHLPAEFTPAVHASLDKFLASVSTVLTSKYRMVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSHGSAQVKGHGKKVADALTNAVAHVDDMPNALSALSDLHAHKLRVDPVNFKLL803K

RefSeq sequencesATGTTTGTTTTTCTTGTTTTATTGCCACTAGTCTCTAGTCAGTGTGTTAATCTTACAACCAGAACTCAATTACCCCCTGCATACACTAATTCTTTCACACGTGGTGTTTATTACCCTGACAAAGTTTTCAGATCCTCAGT544.9K

DepMap cell linesAVHASLDKFLASVSTVLTSKYRMVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSHGSAQVKGHGKKVADALTNAVAHVDDMPNALSALSDLHAHKLRVDPVNFKLLSHCLLVTLAAHLPAEF1.2K

Biology’s data is public. Using it isn’t.

Most of what you need has already been measured and published. It sits in separate portals, each with its own identifiers, file formats and download sizes. Answering one question used to mean days of downloads, ID mapping and spreadsheet joins, if your computer could open the files at all.

Task	Before	With BioQ
Read one record from UniProt TrEMBL	TrEMBL ships as a single 40 GB compressed file. To read record 100 million you decompress everything before it.	Ask for any of 149 million accessions by ID. The record comes back with the byte range it was read from.
Compare drug potency across sources	ChEMBL, DrugCentral and Guide to Pharmacology report IC50, Ki and pIC50 values in mixed units. Averaging them quietly mixes nanomolar with log values.	5.2 million measurements with molar and log units kept apart, each row linked to its source record.
Check whether a gene is a selective dependency	Download DepMap matrices, map cell lines to lineages, write the per-lineage statistics yourself.	Dependency, expression and copy number by lineage for any gene, in one request.
Trust a “not found”	A database that was never loaded and a record that does not exist look the same: an empty result.	Each database reports a hit, searched and absent, or not searched. Only the second one is absence.
Search with an unpublished sequence	Public BLAST and structure servers receive your sequence. Many labs and companies forbid that for unpublished work.	Homology search runs on our own servers. The agent works with a placeholder and never sees the residues.
Find unexplored protein families in an environment	MGnify’s 1.7 billion protein clusters come as terabytes of files. Filtering them by biome needs a compute cluster.	Browse the biome tree, list large clusters with no known domain, and check what they co-occur with.

One question, every database, the evidence attached

Ask the way you think
Type a question, a gene symbol, a UniProt or MGnify accession, a Pfam domain, a drug, a sequence or a coordinate. BioQ recognises what you entered and shows how it read it.
The agent plans and queries
For questions, the agent decides which databases to ask, runs the queries in parallel and shows each step with its parameters and timing.
Every claim cites a record
Numbers in the answer link to the query that produced them. Open any step to see the raw data, copy the API call, or download the table.

Built for the questions researchers actually ask

Target discovery

Which cancers depend on this gene, and has anything been measured binding it?

DepMap, ChEMBL, STRING, AlphaFold

KRAS

Drug–target evidence

Every measured IC50, Ki and Kd for a target and a compound, with the assay type and source.

ChEMBL, DrugCentral, Guide to Pharmacology

EGFR gefitinib

Variant interpretation

Which variants in this gene are pathogenic, and how strongly are they reviewed?

ClinVar, GRCh38

BRCA1

Protein identity

What is this sequence closest to, which domains does it carry, is there a structure?

Swiss-Prot, TrEMBL, Pfam, PDB, AlphaFold

P04637

Enzyme and metagenome mining

Which large protein families in soil have no known domain, and what do they sit next to?

MGnify proteins, clusters, biomes, assemblies

root:Environmental:Terrestrial:Soil

Genome coordinates

The exact bases at a coordinate in human, mouse or any of 544,891 RefSeq sequences.

GRCh38, GRCm39, RefSeq

NC_045512.2:21563-21600

Answers that hold up in review

A search engine for science is only useful if you can defend what it tells you. BioQ is strict about four things.

Absent and not searched are different answers

When a database was not searched, BioQ says so. It never turns a missing index into a negative result.

Units are never pooled

Molar and log-scale potencies are summarised separately, by measurement type and assay type.

Each answer says what it can support

Results carry a readiness verdict: whether there are enough independent observations for a statistical claim, or only for a description.

Every row has an address

Source file, record number and checksum travel with the data, so a reviewer can find the same bytes.

From a real responseGET /api/v1/whereis?id=P04637

{
  "trembl": {
    "status": "miss",
    "means": "the TrEMBL index was searched and does
      not carry this accession"
  },
  "pfam_domains": {
    "status": "not_applicable",
    "means": "no lookup was possible, so this is
      neither a hit nor a miss"
  }
}

GET /api/v1/affinity?gene=EGFR&drug=gefitinib
{
  "measured_type": "IC50", "value": 0.1, "unit": "nM",
  "unit_class": "molar",
  "source_key": "chembl_37.db|activity_id=13297481",
  "sha256": "4be13df3b68e…c15142"
}

Your sequences stay yours

Unpublished sequences are a lab’s most sensitive asset. BioQ runs homology search on its own servers and never forwards your sequences to third-party search services. When you ask the agent about a sequence, the language model sees a placeholder such as [SEQ1] and its length, not the residues. Your search history is kept in your browser, not in your account.

Searches run on BioQ servers
The agent never sees residues
Search history stays in your browser

Everything you can click is an API call

Create a key in your account and call the same endpoints from Python, R, Nextflow or a notebook. Every response is JSON with its provenance included.

Read the API reference

curl -H "Authorization: Bearer $BIOQ_KEY" \
  "https://bioq.bio/api/v1/depmap?gene=KRAS"

import requests

r = requests.get(
    "https://bioq.bio/api/v1/affinity",
    params={"gene": "EGFR", "drug": "gefitinib"},
    headers={"Authorization": f"Bearer {BIOQ_KEY}"},
)
r.json()["summary"]["per_type_and_class"]

library(httr2)

request("https://bioq.bio/api/v1/variant") |>
  req_url_query(gene = "BRCA1", pathogenic_only = 1) |>
  req_auth_bearer_token(Sys.getenv("BIOQ_KEY")) |>
  req_perform() |>
  resp_body_json()

What’s inside

Database	What you can ask	Records
MGnify	Metagenomic proteins and clusters, biome of origin, co-occurrence in assemblies	5.7B proteins
UniProt	Protein records by accession, organism counts, homology search against Swiss-Prot	149.4M entries
Pfam	Protein domains on metagenomic proteins, and proteins carrying a domain	27,048 domain families
PDB, AlphaFold DB	Experimental structures and AlphaFold models with confidence	803K entries
ChEMBL, DrugCentral, GtoPdb	IC50, Ki, Kd and related measurements between targets and compounds	5.2M measurements
ClinVar	Clinical significance and review status of human variants	Human genome
STRING, BioGRID	Protein–protein interactions with confidence scores or curated evidence	162.7M edges
DepMap	CRISPR dependency, expression and copy number across cancer cell lines	1,178 cell lines
GRCh38, GRCm39, RefSeq	Sequence at any coordinate in human, mouse and RefSeq	544.9K sequences

Start with one question

Free accounts include 500 queries, 25 sequence searches and 20 agent questions every day.

Start free