When you ask MIP a question, it doesn’t generate an answer from memory. It reasons about what information is needed, queries the right databases in real time, interprets the results, and synthesizes a cited response. This page explains what happens behind the scenes — the databases MIP can reach, the tools it uses, and how to get the most out of deep research.
How it works
Every question triggers a reasoning loop:
- Decomposition — MIP breaks your question into sub-problems and determines which databases and tools are relevant.
- Tool calls — MIP queries one or more databases. Each query is visible in the chat as a tool call indicator (e.g., “PubMed — Found 12 papers”).
- Interpretation — Results are parsed, filtered, and cross-referenced. MIP reads abstracts, extracts allele frequencies, maps pathways, or compares protein domains — depending on the question.
- Synthesis — A final answer is composed with inline citations linking back to original sources.
This loop can repeat multiple times within a single response. A complex question like “What is the evidence for PCSK9 as a drug target for familial hypercholesterolemia?” might trigger searches across PubMed, ClinVar, gnomAD, ChEMBL, Open Targets, and UniProt — all in one turn.
You can see exactly which databases MIP queried by looking at the tool call indicators above each response. Click any citation badge to open the original source.
Integrated databases
MIP connects to the following databases in real time. No data is pre-cached — every query hits the live API.
Literature and web
| Database | What it provides |
|---|
| PubMed / PMC | Peer-reviewed publications with full abstracts, authors, journal, DOI. Supports complex boolean queries across titles, abstracts, and MeSH terms. |
| Exa Web Search | General web search with deep and fast modes. Useful for guidelines, preprints, institutional pages, and resources not indexed in PubMed. |
Genomics and variant annotation
| Database | What it provides |
|---|
| NCBI Gene / SNP / OMIM | Gene summaries, variant rsIDs, gene-disease relationships, inheritance patterns. Unified search across multiple NCBI databases. |
| Ensembl VEP | Variant consequence prediction, SIFT/PolyPhen scores, transcript mapping, HGVS nomenclature. Accepts both HGVS notation and genomic coordinates. |
| Ensembl Lookup | Gene and transcript metadata, genomic coordinates, cross-references. Sequence retrieval for genes and transcripts. |
| gnomAD | Population allele frequencies (global and per-population), gene constraint metrics (pLI, LOEUF, missense O/E). |
| VariantValidator | HGVS validation, normalization, and coordinate conversion between assemblies and transcript versions. |
| ClinVar | Clinical significance classifications, review status, submitter information, associated conditions. Accessed via NCBI integration. |
| MyGene.info | Comprehensive gene metadata aggregation — symbols, aliases, coordinates, pathways, GO terms, cross-database IDs. |
Protein and structural biology
| Database | What it provides |
|---|
| UniProt | Protein sequence, function, domains, PTMs, active sites, disease associations. Supports lookup, search, and protein-protein comparison with sequence alignment. |
| AlphaFold Database | AI-predicted protein structures by UniProt accession. Returns PDB/CIF URLs and per-residue confidence scores (pLDDT). |
| RCSB PDB | Experimentally determined protein structures. Search by keyword, gene, or UniProt accession. Returns resolution, method, chains, and ligands. |
| STRING | Protein-protein interaction networks with confidence scores. Returns interaction partners, co-expression data, and pathway context. |
| IntAct | Curated molecular interaction data from the EBI. Binary interactions with experimental method and confidence scoring. |
Drug discovery and pharmacology
| Database | What it provides |
|---|
| ChEMBL | Bioactivity data, compound-target relationships, drug mechanism of action. Search by target, compound, or activity type. |
| PubChem | Chemical compound properties, SMILES strings, molecular weights, gene-chemical associations. |
| Open Targets | Gene-disease association scores, tractability assessments, drug information with clinical trial phases. |
| ClinicalTrials.gov | Active and completed clinical trials. Search by condition, intervention, phase, and recruitment status. |
| PharmGKB | Pharmacogenomic drug-gene interactions and clinical dosing guidelines. Accessed via NCBI integration. |
Pathways and gene expression
| Database | What it provides |
|---|
| Reactome | Curated biological pathway database. Search pathways by keyword or retrieve all pathways for a given gene. |
| Expression Atlas | Tissue and condition-specific gene expression data from the EBI. Baseline and differential expression across experiments. |
Rare diseases and phenotypes
| Database | What it provides |
|---|
| Orphanet | Rare disease classifications, prevalence data, inheritance modes, and associated genes. |
| Phen2Gene | Phenotype-driven gene prioritization. Input HPO terms, get ranked candidate genes with scores. |
| HPO (Human Phenotype Ontology) | Standardized clinical phenotype terms with definitions, synonyms, and gene associations. |
Microbiology
| Database | What it provides |
|---|
| BV-BRC | Bacterial and viral genomics — gene search, virulence factors, and antimicrobial resistance genes. |
Structure prediction
MIP can predict protein structures from amino acid sequences using NVIDIA Boltz2, a state-of-the-art structure prediction model. See Structure Prediction for details.
How to get the best results
Be specific about what you need
Vague questions get shallow answers. Specific questions trigger targeted database queries.
| Instead of | Try |
|---|
| ”Tell me about TP53" | "What loss-of-function variants in TP53 are classified as pathogenic in ClinVar, and what are their population frequencies in gnomAD?" |
| "Is this gene a drug target?" | "What compounds in ChEMBL target CDK4 with IC50 below 100nM, and are any in clinical trials?" |
| "What does this protein do?" | "What are the known domains, PTMs, and disease associations for UniProt P38398 (BRCA1)?” |
Ask follow-up questions
MIP retains context within a conversation. After an initial broad search, drill down:
- “What genes are associated with dilated cardiomyopathy?”
- “Which of those have pathogenic variants in ClinVar with at least 2-star review status?”
- “For TTN, what are the gnomAD constraint metrics and known loss-of-function variants?”
Combine databases explicitly
You can ask MIP to cross-reference multiple sources:
- “Search PubMed for recent papers on SGLT2 inhibitors in heart failure, then check ClinicalTrials.gov for ongoing Phase III trials”
- “Look up the protein structure of EGFR in PDB, then find all ChEMBL compounds targeting it with activity data”
- “Compare the UniProt entries for DR3 and DcR3, including domain alignment and sequence identity”
MIP decides which databases to query based on your question. If you want a specific database searched, mention it by name. MIP will prioritize it.
- Tool name — Which database was queried (e.g., “PubMed”, “gnomAD”, “ChEMBL”)
- Result count — How many results were returned
- Status — Green checkmark for successful queries
Click any citation badge in the response to open the original source record in a new tab. This lets you verify every claim MIP makes against the primary data. 