The kinase structure & drug database

Every kinase structure.
Every conformation.
One database.

Kinases are enzymes that transfer a phosphate group from ATP onto target proteins — a single chemical switch that turns biological signals on or off, controlling nearly every process in a living cell. KinaDB brings together 8,626 structures, their activation states, and inhibitor drugs for the human kinome in one searchable place.

Canonical kinase fold
Canonical kinase fold — N-lobe / C-lobe, ATP pocket, activation loop
Kinases by the numbers

The database at a glance

Every figure below is computed directly from the 8,626 structures and 85 drugs currently in KinaDB.

8,626
Kinase structures (PDB chains) in the database
293
Unique kinases covered
10
Kinome groups represented
2
Species
Human 8,234 · Mouse 392
48.9%
Structures predicted in an active conformation
85.5%
Structures with a bound ligand
4,070
Distinct ligands observed across structures
85
Kinase-inhibitor drugs
82 FDA-approved · 66 target genes
The kinome

10 groups, one family tree

Every structure in KinaDB is classified into one of the standard kinome groups. The distribution below reflects the 8,626 structures currently in the database — tyrosine kinases (TYR) and the CMGC group are the most heavily represented.

TYR2,472CMGC2,378CAMK1,186TKL832STE565AGC550OTHER362CK1229NEK51RGC1
The human kinome tree
The human kinome tree — every branch is one of the 518 protein kinases
Predicted activity

Active vs. inactive across the database

Active4,216Inactive4,410
History

Milestones in kinase drug discovery

2001
Imatinib approved — the first targeted kinase inhibitor, for CML
2011
Vemurafenib approved — first BRAF inhibitor, precision oncology begins
2013
Ibrutinib approved — first BTK inhibitor, into blood cancers
2022–25
Mutation-selective and allosteric inhibitors address drug resistance
2025
100th small-molecule kinase inhibitor approved by the FDA
Methodology

How we classify kinase activation state

A geometry-based approach to Active / Inactive labeling.

Every protein kinase has an activation segment — a short, flexible stretch that acts like a molecular switch. When a kinase is "on," this segment folds into a specific shape that lines up the machinery needed to transfer a phosphate group; when it's "off," the segment collapses into a different, incompatible shape.

We represent the activation segment as a simple path through 3D space, one point per residue, and compute the vectors and angles between consecutive residues — describing how the backbone bends and twists at each step. A model trained on this geometric fingerprint, rather than manual inspection, produces the Active / Inactive call shown on every structure page.

αC-helix in
αC-helix out
DFG-in
Active
Intermediate
DFG-out
Intermediate
Inactive
Consensus DFG and αC-helix conformation together define the activation-state call shown on each structure page.
Research

Related preprint

📄

Resolving discrepancies in kinase activity labels using machine learning

EL Moumni, M. Preprints 2025, 2025060609.

A machine-learning approach to reconciling conflicting Active / Inactive activity labels across kinase structures — closely related to the activation-state classification methodology used throughout KinaDB.

doi.org/10.20944/preprints202506.0609.v1 ↗

Explore the full dataset

Browse every structure and search approved kinase inhibitor drugs.

Browse structuresView drugs