Paired activity learning supports systematic identification of mutant-selective EGFR inhibitors
- 1University of Helsinki
Abstract
Acquired resistance to epidermal growth factor receptor (EGFR) inhibitors remains a major clinical challenge in non-small cell lung cancer (NSCLC), where successive rounds of therapy with first- and third-generation EGFR inhibitors often lead to increasingly resistant EGFR variants. In particular, EGFR L858R/T790M/C797S triple mutation renders tumors resistant to earlier-generation inhibitors, while disrupting the covalent engagement required by third-generation inhibitors such as osimertinib. Here, we developed an integrated computational-experimental framework for identifying compounds with differential activity between the triple-mutant EGFR and wild-type EGFR. Matched bioactivity measurements were leveraged to train a mutant-aware paired-activity regression framework, and the predicted candidates meeting stringent potency and mutant-preference criteria were subsequently evaluated using a support vector machine classifier for triple-mutant activity. Application of this sequential screening approach to 12 million molecules identified 16 consensus hits, including several previously reported inhibitors of drug-resistant EGFR variants. Based on their predicted differential activity and degree of prior characterization, five hit compounds were prioritized for initial biochemical evaluation, and two were selected for further biochemical, cellular and structural characterization. Mutant EGFR crystal structures in complex with the best-performing compounds revealed two distinct noncovalent binding modes within the ATP-binding cleft, demonstrating the ability of the framework to identify potent inhibitors with diverse binding modes. Together, this work illustrates how incorporating mutant-selectivity into a predictive framework as paired activities support the identification of mutant EGFR inhibitors and provide insights into their potency and selectivity. The discovered complementary structures provide experimentally defined templates for optimizing potency, mutant preference and drug-like properties of these and future compounds combating drug resistant NSCLC.
Related
- Prime editing enables high-throughput identification and characterization of selective MutSβ inhibitors
- Personalized prediction of response to selective serotonin reuptake inhibitors (SSRIs) treatment for major depression disorder using deep learning networks from connectivity maps of EEG
- Identification of a novel NR2B-selective NMDA receptor antagonist using a virtual screening approach
- Identification of Cysteine-Reactive Early Mitotic Inhibitors
- Phenolic Compounds from Ruta chalepensis Var. angustifolia as Potential HMG-CoA Reductase Inhibitors: Bioactivity-Guided Fractionation and Preliminary Identification of Active Metabolites
BibTeX
@misc{zhu2026paired,
title = {{Paired activity learning supports systematic identification of mutant-selective EGFR inhibitors}},
author = {J. Zhu and W. Kong and L. Huang and Z. Z. Song and H. Ying and K. J. Kurppa and M. Miihkinen and S.-J. Zhu and T. Aittokallio},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.21.760684},
url = {https://www.biorxiv.org/content/10.64898/2026.10.21.760684v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Paired activity learning supports systematic identification of mutant-selective EGFR inhibitors
AU - J. Zhu
AU - W. Kong
AU - L. Huang
AU - Z. Z. Song
AU - H. Ying
AU - K. J. Kurppa
AU - M. Miihkinen
AU - S.-J. Zhu
AU - T. Aittokallio
PY - 2026
DA - 2026/10/21
PB - bioRxiv
DO - 10.64898/2026.10.21.760684
UR - https://www.biorxiv.org/content/10.64898/2026.10.21.760684v1
AB - Acquired resistance to epidermal growth factor receptor (EGFR) inhibitors remains a major clinical challenge in non-small cell lung cancer (NSCLC), where successive rounds of therapy with first- and third-generation EGFR inhibitors often lead to increasingly resistant EGFR variants. In particular, EGFR L858R/T790M/C797S triple mutation renders tumors resistant to earlier-generation inhibitors, while disrupting the covalent engagement required by third-generation inhibitors such as osimertinib. Here, we developed an integrated computational-experimental framework for identifying compounds with differential activity between the triple-mutant EGFR and wild-type EGFR. Matched bioactivity measurements were leveraged to train a mutant-aware paired-activity regression framework, and the predicted candidates meeting stringent potency and mutant-preference criteria were subsequently evaluated using a support vector machine classifier for triple-mutant activity. Application of this sequential screening approach to 12 million molecules identified 16 consensus hits, including several previously reported inhibitors of drug-resistant EGFR variants. Based on their predicted differential activity and degree of prior characterization, five hit compounds were prioritized for initial biochemical evaluation, and two were selected for further biochemical, cellular and structural characterization. Mutant EGFR crystal structures in complex with the best-performing compounds revealed two distinct noncovalent binding modes within the ATP-binding cleft, demonstrating the ability of the framework to identify potent inhibitors with diverse binding modes. Together, this work illustrates how incorporating mutant-selectivity into a predictive framework as paired activities support the identification of mutant EGFR inhibitors and provide insights into their potency and selectivity. The discovered complementary structures provide experimentally defined templates for optimizing potency, mutant preference and drug-like properties of these and future compounds combating drug resistant NSCLC.
ER -