DDDR-27. Transcriptomics-based AI drug discovery in diffuse midline glioma

Abstract INTRODUCTION Diffuse midline gliomas H3K27-altered (DMGs) are the most malignant and devastating brain tumors in children. Recent advancements in tumor sample availability and next-generation sequencing we use a system-based approach that leverages gene expression as a molecular representation for therapeutic discovery. We used Open Cancer TherApeutic Discovery (OCTAD), a machine learning platform that predict drug effects by analyzing drug-induced gene expression profiles and compound structures. However, the final candidate had poor brain penetration, and we have developed a platform, Gene expression profiles Predictor on chemical Structures (GPS). This enables screening of large compound libraries and evaluation of therapeutic candidates in DMG cell lines and patient-derived xenografts (PDXs). METHODS DMG RNA-Seq data were acquired from St. Jude Children’s Research Hospital and Children’s Brain Tumor Network. We developed a deep learning autoencoder to identify reference normal tissues from Genotype-Tissue Expression. We also leveraged single-cell RNA-seq of six DMG patients to create patient signatures. A DMG meta-signature was created and fed into OCTAD to predict drug candidates that could reverse DMG gene expression. Subsequently, DMG meta-signature was fed into GPS to predict drug candidates from the Enamine CNS library, and promising compounds were tested in vitro, with the most effective validated in vivo. RESULTS OCTAD identified triptolide and mycophenolate mofetil (MMF), both of which demonstrated lower IC50 values in DMG cell lines (SF8628, SU-DIPG4) compared to normal human astrocytes. Moreover, an in vivo study using mice with SF8628 subcutaneous xenografts showed that MMF suppressed tumor growth and prolonged survival. GPS identified seven compounds, two of which exhibited IC50 values below 20 µM in three DMG cell lines (SF8628, DIPG-007, SU-DIPG36). CONCLUSION Our novel computational framework leverages deep learning to address gaps in drug discovery for DMG. Ongoing work, including in vivo studies, aims to evaluate the antitumor activity, brain penetration, and survival benefits of two most potent compounds in mice bearing DMG patient-derived xenografts (PDX).