Abstract 4466: Machine-learning-driven optimization of kinase inhibitors to overcome CNS efflux and target breast cancer brain metastases

Breast cancer (BC) is the most prevalent form of malignancy and the leading cause of cancer mortality in women. The fatality of BC is highly attributed to BC brain metastases (BCBM) which represent 30%, with a 1-year survival rate of 20%. About 15-20% of BC demonstrates significant overexpression of HER2, hence the development of anti-HER2 targeted therapies such as tucatinib which have prolonged survival rates and improved patient quality of life. Despite the improvements in HER2 targeted therapy to rapidly target BCBM, tucatinib and other quinazoline HER2 inhibitors (dacomitinib and gefitinib) possess poor CNS penetration due to P-gp and BCRP efflux. In vitro, BBB permeability assays such as the BBB-PAMPA are cost-intensive prompting the need for computational methods that can rapidly predict the BBB permeability potential of small molecules in the early phase of drug discovery. Herein, we used quantitative and qualitative machine learning (ML) algorithms trained with experimental data to direct structural optimization of in-house KIs to improve BBB penetration. We employed qualitative and quantitative ML models to identify the most promising compounds from our in-house library of 120 KIs previously designed for dual HER2/VEGFR targeting. The core of our approach lies in the structural simplification of starting compounds to drastically reduce molecular weight and topological surface area and the appropriate substitution of solubilizing residues to modulate pKa. Furthermore, we developed a benchmark of optimal physiochemical properties of molecules intended as CNS therapeutics using an approved CNS drug dataset. Optimized analogs possessing physicochemical properties within the benchmark range and a favorable ML prediction of BBB permeability were synthesized and tested on a kinase panel. BBB-PAMPA was also used to evaluate BBB penetration of both starting KIs and optimized analogs. Optimized analogs demonstrated HER2 kinase inhibitory activity alongside a significant increase in BBB permeability compared to the starting compounds. Our results suggest that structural optimization to modulate the physicochemical properties of small molecules is key to identifying potential drug candidates for targeting BCBM. In vivo studies using animal models will be conducted to validate these findings. Citation Format: Eneye D. Ajayi, Mostafa M. Aref, Rokaia Abdullah, Md Emran Hossain, Hamed I. Ali. Machine-learning-driven optimization of kinase inhibitors to overcome CNS efflux and target breast cancer brain metastases [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 4466.