TargetMind

Multi-agent AI for therapeutic target discovery

A virtual target review board,built from AI experts.

TargetMind convenes independent domain-expert agents to investigate, challenge and reconcile biomedical evidence, and ranks therapeutic targets only once they reach a structured consensus.

A virtual roundtable: a chief scientist and four domain-expert agents deliberate over protein structure, single-cell, genetic and membrane-target evidence

Step inside the TargetMind

Tools and databases

Every expert works from the same evidence infrastructure

A registry of 694 biomedical tools spanning six domains gives each agent direct access to primary data sources rather than recalled knowledge.

Biomedical tools registry across six domains: general, disease, target, drug, pathway and omics tools
biomedical tools
694
evidence domains
6
data sources
27

Gene and protein identity, structures, tractability, safety, homology and druggability prediction.

Where the tools come from

  • Open Targets34
  • UniProt26
  • ChEMBL20
  • RCSB PDB20
  • Ensembl17
  • +2 more4

Highlights

  • get_target_tractability_by_ensemblID

    Retrieve tractability assessments, including modality and values, for a specific target Ensembl ID

  • predict_single_gene_druggability

    Predicts the druggability of a single gene using a machine learning model

  • pdb_get_structure

    Retrieve detailed structure information for a PDB entry

  • get_target_safety_profile_by_ensemblID

    Retrieve known target safety liabilities for a specific target Ensembl ID

Explore capability hub →

Browse all 694 tools, the software environment and our datasets.

Counts come from the current tool registry; the per-domain values in the paper figure are approximate.

The platform

A working research platform, not just a paper

Start from a disease, let the review board plan and execute the investigation, and inspect every intermediate result — tables, code, reports and 3D structures — in one workspace.

TargetMind
TargetMind home page with disease search results from OpenTargets

Disease grounding

Search any disease and anchor the workspace to its OpenTargets ontology IDs (EFO / MONDO).

Live research plan

The agents publish a task plan and update its progress as they work.

Every result is a file

Tables, figures, code and reports land in the results panel — preview, filter and download.

Built-in 3D structures

PDB, SDF and other structure files open directly in an interactive Mol* viewer.

Benchmark

Stronger target reasoning, deeper into unexplored biology

Therapeutic target reasoning

Significantly outperforms state-of-the-art biomedical agents and general-purpose models across standardized therapeutic target reasoning tasks.

Novel target discovery

Navigates beyond well-studied genetic networks to prioritize highly underexplored targets with robust disease-linked support.

Target space map showing TargetMind's picks in the underexplored region versus baselines in the well-studied region, and bar charts comparing TargetMind with baselines on TRQA benchmarks
Top: TargetMind's prioritized targets fall in the underexplored region of target space, while baselines concentrate in the well-studied region. Bottom: TRQA-lit choice questions (accuracy), TRQA-lit short-answer and TRQA-db (average recall), compared with OriGene, GPT-4o-Search, o3-mini, DeepSeek-R1, DeepSeek-V3, Claude-3.7-Sonnet and Gemini-2.5-Pro.

Experimental validation

From computational reasoning to the bench

TargetMind's top-ranked candidates were taken into experimental investigation across three disease models — renal fibrosis, polycystic kidney disease and hypertrophic cardiomyopathy.

Experimental validation across three disease models: renal fibrosis, polycystic kidney disease and hypertrophic cardiomyopathy, each ending in a validated therapeutic target

Bring a review board to your next target.

byndclf@163.com