T023 · What is a kinase?

Note: This talktorial is a part of TeachOpenCADD, a platform that aims to teach domain-specific skills and to provide pipeline templates as starting points for research projects.


Aim of this talktorial

In this talktorial, we will talk about kinases: why are they important in life and drug design, what do they look like, and what data resources are available? Finally, we select a set of kinases which will be analyzed in the forthcoming talktorials T024-T028 with respect to their similarity, the goal being to gain insight into potential off-target effects.

Contents in Theory

  • Kinases in a nutshell

    • The human kinome

    • Kinase structures and important motifs

  • Kinase resources

    • Kinase structures and related information

    • Bioactivity data

  • Kinase similarity: Off-targets and promiscuous binding

  • Kinase dataset compilation

Contents in Practical

  • Define the kinases of interest



Kinases in a nutshell

Kinases are established drug targets to combat cancer and inflammatory diseases (Nat. Rev. Drug Discov. (2021), 20(7), 551-569). They are involved in most aspects of cell life by phosphorylating, and thereby activating, either themselves or other proteins. They are among the most frequently mutated proteins in tumors.

As of September 2021, \(5782\) X-ray structures of human kinases have been resolved (see the KLIFS database) and \(67\) FDA-approved small molecule protein kinase inhibitors are on the market (see the list compiled by the Blue Ridge Institute for Medical Research). Most of the approved drugs bind in the ATP-binding pocket and intermediate surroundings.

Despite of decades of kinase research, there are still many open challenges:

  • A large fraction of the kinome is un-/underexplored.

  • Many kinase inhibitors are promiscuous binders causing off-target effects or enabling polypharmacology.

  • There are occurrences of drug resistance due to mutations.

The human kinome

The human kinome consists of over \(500\) protein kinases, but this number may vary depending on the data resource (see overview on Kinodata).

As reported in Science (2002), 298(5600), 1912-1934, Manning et al. clustered the human protein kinases based on their sequence similarity into eight major groups (AGC, CAMK, CK1, CMGC, RGC, STE, TK, TKL) and one “Other” group for unassigned kinases, as well as atypical kinases. The kinase clustering is visualized as the Manning kinome tree. The kinase resource KinMap enables mapping of kinase data onto that tree, e.g. the number of X-ray structures per kinase as shown in Figure 1.

Manning tree with number of structures per kinase (KinMap)

Figure 1: Number of PDB structures per kinase mapped onto the Manning kinome tree using KinMap. Check the appendix on how to generate this KinMap tree.

Kinase structures and important motifs

Kinase sequences and structures are highly conserved. Important regions in the kinase pocket include (see Figure 2):

  • Hinge region: Forms key hydrogen bonds to ligands.

  • DFG motif: Flips between phenylalanine (F) and aspartate (D), driving the active and inactive state.

  • αC-helix: Forms in the αC-in conformation a salt bridge between highly conserved lysine (K) and glutamine (Q).

  • Glycine-rich (G-rich) loop: Stabilizes ATP binding.

Kinase structure with key motifs

Figure 2: Kinase structure with important key motifs: Hinge region, DFG motif, αC-helix, and G-rich loop. The example structure shown here represents CDK2, PDB ID: 1FIN. Check the appendix on how to generate this visualization with opencadd.

Kinase resources

The focus on this protein family has led to a plethora of freely available data on compounds, bioactivity, and structures that are being used for computational drug development (Annual Reports in Medicinal Chemistry 50 (2017): 197-236).

Bioactivity data

ChEMBL is a well-known bioactivity database, which releases updated versions every now and then. In September 2021, there are over two million compounds and \(14,000\) targets that are stored. In ChEMBL29, there are over \(160,000\) measurements on kinases (see Figure 3).

As with other data types, the coverage of bioactivity data is highly unbalanced among the human kinases, depending on how much research is spent on certain kinases (Figure 3).

Manning tree with number of ChEMBL activities per kinase (KinMap)

Figure 3: Number of ChEMBL29 bioactivities per kinase mapped onto the Manning kinome tree using KinMap. Check the appendix on how to generate this KinMap tree.

However, ChEMBL is not the only available bioactivity database. Below is an non-exhaustive list of available kinase profiling data sets.

Kinase similarity: Off-targets and promiscuous binding

As described before, kinases are highly conserved, especially in their binding site. This high similarity is a challenge in drug design because ligands may form similar binding modes not only with their designated target (on-target) but also with other targets (off-targets). Such promiscuous binding can cause mild to severe side effects.

Predicting these side effects is non-trivial since some off-targets are not obvious. For example, the EGFR inhibitor Erlotinib shows affinities to other kinases in the highly sequentially-similar TK kinase group. However, it also strongly affects the off-targets GAK, LOK, and SLK, which are in more remote kinase groups (Figure 4).

Erlotinib profiling data from Karaman dataset (KinMap)

Figure 4: Profiling data for EGFR inhibitor Erlobinib from the Karaman et al. dataset (Nature Biotechnology (2008), 26, 127-132) mapped onto the Manning kinome tree using KinMap. Check the appendix of this notebook on how to generate this figure.

In the following four talktorials, namely Talktorials T024-027, we will assess kinase similarity from different perspectives, which we compare with each other in Talktorial T028:

  • Talktorial T024: Kinase similarity based on KLIFS pocket sequence

  • Talktorial T025: Kinase similarity based on KiSSim pocket structure

  • Talktorial T026: Kinase similarity based on KLIFS interaction fingerprint

  • Talktorial T027: Kinase similarity based on ligand promiscuity (ChEMBL bioactivity data)

  • Talktorial T028: Compare kinase similarity measures from Talktorials T024-T027

Kinase dataset compilation

In the course of the kinase similarity talktorials (Talktorials T024-T028), we will use nine kinases from a study published in Molecules (2021), 26(3), 629, which were selected for the following reasons:

  • Profile 1 combined EGFR and ErbB2 as targets and BRAF as a (general) anti-target.

  • Out of similar considerations, Profile 2 consisted of EGFR and PI3K as targets and BRAF as anti-target. This profile is expected to be more challenging as PI3K is an atypical kinase and thus less similar to EGFR than for example ErbB2 used in Profile 1.

  • Profile 3, comprised of EGFR and VEGFR2 as targets and BRAF as anti-target, was contrasted with the hit rate that we found with a standard docking against the single target VEGFR2 (Profile 4).

  • To broaden the comparison and obtain an estimate for the promiscuity of each compound, the kinases CDK2, LCK, MET and p38α were included in the experimental assay panel and the structure-based bioinformatics comparison as commonly used anti-targets.


from pathlib import Path

import pandas as pd
HERE = Path(_dh[-1])
DATA = HERE / "data"

Define the kinases of interest

We have collected information about these nine kinases in the CSV file kinase_selection.csv:

Note: You can run the kinase similarity Talktorials T024-T028 with your own set of kinases. To do so, please update the CSV file with your kinases; the only mandatory columns are kinase_klifs and uniprot_id.

kinase_selection_df = pd.read_csv(DATA / "kinase_selection.csv")
kinase kinase_klifs uniprot_id group full_kinase_name
0 EGFR EGFR P00533 TK Epidermal growth factor receptor
1 ErbB2 ErbB2 P04626 TK Erythroblastic leukemia viral oncogene homolog 2
2 PI3K p110a P42336 Atypical Phosphatidylinositol-3-kinase
3 VEGFR2 KDR P35968 TK Vascular endothelial growth factor receptor 2
4 BRAF BRAF P15056 TKL Rapidly accelerated fibrosarcoma isoform B
5 CDK2 CDK2 P24941 CMGC Cyclic-dependent kinase 2
6 LCK LCK P06239 TK Lymphocyte-specific protein tyrosine kinase
7 MET MET P08581 TK Mesenchymal-epithelial transition factor
8 p38a p38a Q16539 CMGC p38 mitogen activated protein kinase alpha

We will load this dataset in all downstream talktorials to assess kinase similarity from different perspectives.


KinMap data

There are some KinMap trees shown in this notebook. The code below generates the KinMap CSV files to be uploaded to KinMap: http://www.kinhub.org/kinmap.

Note: PNG downloads do not seem to work anymore, thus download as SVG and convert to PNG in your terminal (Linux) via convert -density 25 my_kinmap_figure.svg my_kinmap_figure.png (SVG cannot be included in Jupyter notebooks out-of-the-box).

def format_for_kinmap(kinase_names, kinase_values, size_min=10, size_max=50):
    Given kinase names and some associated values, generates a KinMap data file
    that will display values as circles of size [`size_min`, `size_max`].

    kinase_names : list of str
        Kinase names.
    kinase_values : list of float
        Some associated values, such as the number of bioactivites.
    size_min : int
        Minimum circle size on KinMap tree (minimum input value will be scaled to `size_min`).
    size_max : int
        Maximum circle size on KinMap tree (maximum input value will be scaled to `size_min`).

        KinMap data with columns `xName` (kinase name), `size` (circle size for KinMap tree).

    data = pd.DataFrame({"xName": kinase_names, "values": kinase_values})
    min_ = data["values"].min()
    max_ = data["values"].max()
    data["size"] = data["values"].apply(
        lambda x: ((x - min_) / (max_ - min_) * size_max) + size_min
    return data[["xName", "size"]]

Number of PDB structures per kinase

Generate the number of structures per kinase in the KinMap format to be mapped onto the kinome tree.

from opencadd.databases.klifs import setup_remote

klifs = setup_remote()
structures_df = klifs.structures.all_structures()

# Get number of structures per kinase
n_structures_per_kinase = (
    structures_df.groupby(["structure.pdb_id", "kinase.klifs_name"])

# Save in KinMap format
kinmap_n_structures_per_kinase = format_for_kinmap(
    n_structures_per_kinase.index, n_structures_per_kinase.values
kinmap_n_structures_per_kinase.to_csv(DATA / "kinmap_n_structures_per_kinase.csv", index=None)
# Some kinases will not be resolved in KinMap and will be simply dropped

Identify the kinase which has the most structures.

kinmap_n_structures_per_kinase.iloc[n_structures_per_kinase.argmax()].xName, max(
('CDK2', 426)

Number of ChEMBL bioactivities per kinase

Generate the number of ChEMBL bioactivities per kinase in the KinMap format to be mapped onto the kinome tree.

Note: The cell below takes a few seconds to execute.

from opencadd.databases.klifs import setup_remote

# Get bioactivity data
path = "https://github.com/openkinome/kinodata/releases/download/v0.3/activities-chembl29_v0.3.zip"
data = pd.read_csv(path, index_col=None)
data = data[data["activities.standard_type"] == "pIC50"]
data = data.dropna()

# Get kinase data
klifs = setup_remote()
kinases_df = klifs.kinases.all_kinases()
kinases_df = kinases_df[kinases_df["kinase.uniprot"] != "0"]
# Some UniProt IDs have several names in KLIFS, keep only first
kinases_df = kinases_df.groupby("kinase.uniprot").first()

# Map UniProt ID > kinase KLIFS name
data = pd.merge(data, kinases_df, left_on="UniprotID", right_on="kinase.uniprot", how="left")

# Get number of activities per kinase
n_activities_per_kinase = data.groupby("kinase.klifs_name").size()

# Save in KinMap format
kinmap_n_activities_per_kinase = format_for_kinmap(
    n_activities_per_kinase.index, n_activities_per_kinase.values
kinmap_n_activities_per_kinase.to_csv(DATA / "kinmap_n_activities_per_kinase.csv", index=None)
# Some kinases will not be resolved in KinMap and will be simply dropped

Erlotinib profiling data from Karaman et al. dataset

  • Go to http://www.kinhub.org/kinmap/

  • Select “Data Source”: Profiling

  • Select “Data type”: Karaman et al., 2018

  • Select “Karaman et al., 2018”: Erlotinib

  • Click “Add source”

  • Click “Apply”

Kinase structure visualization with opencadd

We are using as an example the ATP-bound CDK2 structure with the KLIFS ID 4367.

from opencadd.structure.pocket import PocketKlifs, PocketViewer

# Get structure and pocket
pocket = PocketKlifs.from_structure_klifs_id(4367)

# Show pocket
viewer = PocketViewer()
viewer.add_pocket(pocket, ligand_expo_id="ATP", show_pocket_center=False)
viewer.viewer.render_image(trim=True, factor=2);