DNA Damage Response (DDR) dysregulation is a major factor in cancer progression and therapy resistance. Such cancers are particularly vulnerable to WEE1 inhibition as it forces tumor cells with sublethal DNA damage to prematurely enter mitosis, eventually leading to mitotic crisis and cell death. Although preliminary clinical data following WEE1 inhibition have shown promising results, patient responses are diverse, and there is still a lack of robust biomarkers to guide patient selection. Our goal was to develop a clinically relevant, biology-informed machine learning predictor for response to the WEE1 inhibitor Debio 0123 in combination with carboplatin (CB).

This model was developed to predict response from bulk RNA-seq data generated from tumor biopsies collected prior to Debio 0123+CB treatment. It was built using the Genialis™ Supermodel framework, a large molecular model (LMM) that transforms raw RNA-seq data into interpretable biomodules, which are algorithmic representations of specific aspects of biology. The best performing model was configured using logistic regression with ElasticNet regularization and accounts for interactions between Genialis™ biomodules.

It achieved a high AUROC and accuracy on the Debio 0123+CB-treated cancer patient data. Our analyses show clear biological stratification, with responders and non-responders aligning with distinct Genialis™ biomodules.

This robust, interpretable predictor provides a strong foundation for biomarker-guided clinical strategies for the WEE1 inhibitor Debio 0123 in combination with CB. Future work will focus on further tuning its performance capabilities in addition to generalizability across cancer types, mono- and other combination therapies.

Published for ESMO AI 2025.