AI Scientist: Endometriosis Target Discovery

How many people do you personally know who have endometriosis?

Poll

012345678+

Reported people known with endometriosis

~30

Based on averages, you likely know around 30 people with endometriosis — most just haven't been diagnosed. The average person knows ~600 people; roughly half are women, and 1 in 10 women have endometriosis.

190M

Endometriosis is estimated to affect 190 million reproductive-age women and girls globally.

Target Landscape

Two views of the 178 targets: a matrix by noveltyi and druggabilityi, and a scatter plot of confidence vs. druggability colored by novelty.

Novelty vs Druggability Matrix

Click any target below to see more details

Novelty / Drug.

Low (1)

Medium (2)

High (3)

Novel (3)

Emerging (2)

Known (1)

Target Classes

Functional classification of the 178 targets

unknown

66

GPCR

35

Kinase

29

Oxidoreductase

14

Phosphatase

8

Protease

6

Nuclear Receptor

5

Transporter

3

Ion Channel

3

Ligase

3

Transcription Factor

2

Transferase

2

Lyase

2

Evidence Sources

How targets entered the discovery pipeline

Expression

77

Functional

38

Unknown

33

Pathway

20

Literature

5

Genetic

4

Drug Target

1

Confidence vs Druggability Score

Each dot is a target. Size = PDB structure count. Color = novelty (purple = novel, amber = emerging, gray = known). Hover for details.

Novel

Emerging

Known

No PDB Many PDB

How targets are scored

Every target is graded on two independent 1–3 scales. These aren't annotations — they function as hard selection gates.

Druggability

Structural evidence for small-molecule tractability.

3 · High
Ligand-bound co-crystal structure exists, or a clearly defined binding pocket with known small-molecule ligands confirmed by crystallography.
2 · Moderate
X-ray structure in the PDB but without co-crystals; known small-molecule binders or chemical probes reported in literature.
1 · Poor
No experimental crystal structure.

Gate: only targets scoring 3 satisfy the small-molecule druggability criterion.

Novelty

Clinical development stage. Hybrid phases (e.g., Phase 1/2) round down. Withdrawn, terminated, or suspended trials do not count.

2 · Dark / Novel
Most novel. No registered clinical trials at any phase; only academic tool compounds or probes exist.
3 · Early Development
Phase 0–2 trials exist for other indications, or Phase 3+ drugs hit the target only as a secondary / off-target effect. Multi-mechanism drugs with 3+ targets default here unless explicitly selective.
1 · Clinically Validated
Least novel. A selective or sole-mechanism drug is in Phase 3+ or FDA-approved for any indication, or any drug is in Phase 2+ for the disease under study. Subunits of a complex targeted by a selective Phase 3+ drug also score 1.

Gate: only targets scoring 2 or 3 are acceptable; score 1 targets are excluded.

About this research

The AI Scientist

Deep Origin's AI Scientist is an autonomous agent conducting drug discovery research for endometriosis, funded through ARIA's AI Scientist program in collaboration with Arctoris (wet lab). The agent documents its reasoning publicly on X (@DOAIScientist).

The Problem

Endometriosis affects ~190 million people worldwide, has a 7–10 year average diagnostic delay, and currently has no approved disease-modifying drug. Existing treatments manage symptoms but don't address the underlying disease biology.

Methodology

The agent built an evidence lake from 10,539 PubMed papers plus ChEMBL, PDB, ClinicalTrials.gov, GEO expression datasets, and UniProt. It extracted 323 candidate targets, filtered to 178 based on druggability and safety, then stress-tested each with standardized evaluation across four phases.

Infrastructure

Built on Deep Origin's computational drug discovery platform including Balto (AI assistant) and DO Patent (document analysis). Research logs and full methodology available on the AI Scientist's X feed.