% IMPORTANT: The following is UTF-8 encoded. This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.
@ARTICLE{Singh:305576,
author = {G. Singh and T. Schmenger and J. C. Gonzalez-Sanchez and A.
Kutkina and N. Bremec and G. D. Diwan and P. Mozas and C.
López and R. Siebert and R. Sotillo$^*$ and R. B. Russell},
title = {{D}iscriminating activating, deactivating and resistance
variants in protein kinases.},
journal = {Genome medicine},
volume = {17},
number = {1},
issn = {1756-994X},
address = {London},
publisher = {BioMed Central},
reportid = {DKFZ-2025-02238},
pages = {133},
year = {2025},
abstract = {Distinguishing whether genetic variants in protein kinases
cause gain or loss of function is critical in clinical
genetics. In particular, gain (and not loss)-of-function
variants are often immediately amenable to treatment by
inhibitors, making their identification a potential boon to
personalised medicine. Most existing computational methods
for variant pathogenicity prediction simply distinguish
damaging from benign variants and provide no further
functional insights. Here, we present a data-driven approach
that differentiates activating, deactivating, and resistance
variants.To train and evaluate our method, we curated a
dataset of 2505 variants (375 activating, 1028 deactivating,
98 resistance and 1004 neutral) across 441 kinases from the
literature and public databases. Each variant was
represented as a vector of sequence, evolutionary and
structural features, which we then used to train machine
learning models to distinguish among the four types of
variants. The resulting predictors achieved excellent
performance (mean AUC = 0.941). We tested a selection of
variants by over-expression in T-REx-293 cells followed by
gene expression or biochemical tests.Applying the predictors
to uncharacterised variants, we observed a strong enrichment
of activating mutations in cancer genomes, deactivating
variants in hereditary disease, and few of either in
variants from healthy individuals. We experimentally
validated several predicted activating variants from cancer
samples. For p.Ser97Asn in PIM1, phosphorylation events
suggested increased activity. For p.Ala84Thr in MAP2K3, gene
expression and mitochondrial staining showed a reduction in
mitochondrial function, the opposite effect of MAP2K3
deletions. We provide an online application that enables
users to analyse any kinase-domain variant, obtain
prediction scores and explore known nearby variants in other
kinases.Our predictors, together with the rapid experimental
validations, demonstrates a feasible strategy for
identifying activating variants in kinases in a time frame
that would enable clinical decisions.},
keywords = {Humans / Protein Kinases: genetics / Protein Kinases:
metabolism / Protein Kinases: chemistry / Genetic Variation
/ Mutation / Computational Biology: methods / Machine
Learning / Cancer genomics (Other) / Gain-of-function
(Other) / Genetic variants (Other) / Loss-of-function
(Other) / Machine learning (Other) / Precision medicine
(Other) / Protein kinases (Other) / Resistance (Other) /
Variant pathogenicity prediction (Other) / Protein Kinases
(NLM Chemicals)},
cin = {B220},
ddc = {610},
cid = {I:(DE-He78)B220-20160331},
pnm = {312 - Funktionelle und strukturelle Genomforschung
(POF4-312)},
pid = {G:(DE-HGF)POF4-312},
typ = {PUB:(DE-HGF)16},
pubmed = {pmid:41152984},
pmc = {pmc:PMC12570665},
doi = {10.1186/s13073-025-01564-z},
url = {https://inrepo02.dkfz.de/record/305576},
}