Drug repositioning for human MKN45 gastric cancer mouse model using deep learning AI and experimental validation

Gastric cancer (GC) requires novel therapies. This study employed artificial neural networks (ANNs) to repurpose existing drugs against GC.

ANN models predicted candidates from DrugBank using pharmacogenomic descriptors. Top compounds, selected by favorable IC₅₀ and Z-scores, were screened in vitro (cytotoxicity in AGS/MKN-45 cells; toxicity in HDFs). Leads were validated in vivo in MKN-45 xenograft mice, assessing tumor markers (Ki67, CD44) and target genes.

Two leads, amitriptylinoxide (20313) and phytonadione (5284607), exhibited high in vitro potency and selectivity. In vivo, both significantly inhibited tumor growth, achieving final volumes of 2.0 ± 1.9 mm³ and 0.05 ± 0.02 mm³, respectively. This efficacy was comparable (0.08 ± 0.03 mm³) or superior to cisplatin. Critically, both compounds induced significant body weight gain, indicating markedly lower systemic toxicity than the weight loss observed with cisplatin. They also suppressed Ki67/CD44 expression (>50%) and stemness genes.

This ANN-driven approach successfully identified amitriptylinoxide and phytonadione as potent GC drug candidates with in vivo efficacy rivaling cisplatin and a significantly improved toxicity profile.