Abstract:Abstract Diffuse Midline Glioma (DMG) are fatal pediatric brain tumors with no therapies. We leveraged network-based methodologies to dissect the heterogeneity of DMG tumors and to discover Master Regulator (MR) proteins representing pharmacologically accessible, mechanistic determinants of molecularly distinct cell states. We produced the first DMG regulatory network from 122 publicly available RNAseq profiles with ARACNe (Basso et al. Nat Genet 2005), and inferred sample-specific MR protein activity with VIPER (Alvarez et al. Nat Genet 2016) based on the differential expression of their targets. 7 of the top 25 most active MRs found comprise a well-characterized MR block (MRB2) (Paull et al.Cell 2021), frequently activated across aggressive tumors, and enriched in DMG patient MR signatures (Fisher’s Exact Test p=4.4 × 10−18). A CRISPR/Cas9 KO screen across 3 DMG patient cell lines identified a set of 73/77 essential genes that were enriched in the MR signature of 80% of patient samples (GSEA p=0.000034). FOXM1 emerged as an essential MR, significantly activated across virtually all patients. We then generated RNAseq profiles following perturbation with ~300 oncology drugs in 2 DMG cell lines most representative of patient MR signatures, and used this to identify drugs that invert patient MR activity profiles using the NYS/CA Dept. of Health approved OncoTreat algorithm (Alvarez et al. Nat Genet 2018). OncoTreat predicted sensitivity to HDAC, MEK, CDK, PI3K, and proteosome inhibitors in subsets of patients, overlapping with published DMG drug screens. Importantly, 80% of OncoTreat-predicted drugs (p<10−5) from 3 DMG patient tumor biopsies showed in vitro sensitivity in cultured tumor cells from the respective patients, with overall 68% accuracy among 223 drugs evaluated by both OncoTreat and in vitro (Fisher’s Exact Test p=0.0449). Further analysis of DMG intra-tumor heterogeneity via protein activity inference across DMG single cells from 6 published scRNAseq profiles identified 6 tumor clusters with unique MR signatures co-existing in virtually all patients representing distinct cellular states (2 astrocyte-, 1 oligodendrocyte-, and 3 oligodendrocyte precursor cell-like states). Targetable MRs and OncoTreat-predicted drugs were distinct between these states. Bulk RNAseq analysis recapitulated predictions seen in the more prevalent OPC-like states, but failed to capture MR and drug predictions for the AC-like states (e.g. JAK1/Ruxolitinib and STAT3/Napabucasin). We are currently validating cell state-specific drug predictions in vivo at single-cell resolution in subcutaneous patient-derived xenograft and orthotopic syngeneic DMG models that we have shown recapitulate patient tumor heterogeneity, including with focused ultrasound-mediated drug delivery. This provides a platform to nominate much-needed novel drugs and drug combinations to treat DMG. Citation Format: Ester Calvo Fernandez, Junqiang Wang, Xu Zhang, Hong-Jian Wei, Hanna E. Minns, Aaron T. Griffin, Lukas Vlahos, Timothy J. Martins, Pamela S. Becker, John Crawford, Robyn D. Gartrell, Luca Szalontay, Stergios Zacharoulis, Zhiguo Zhang, Robert Wechsler-Reya, Cheng-Chia Wu, Andrea Califano, Jovana Pavisic. Network-based inference identifies cell state-specific drugs targeting master regulator vulnerabilities in diffuse midline glioma. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 4304.

Diagnostic Evidence GAuge of Single cells (DEGAS): a flexible deep transfer learning framework for prioritizing cells in relation to disease

The Potential of DEirlncRNAs: A Novel Approach to Predict Glioblastoma Prognosis

Identification of cell types, states and programs by learning gene set representations

Biologically Informed Deep Learning to Infer Gene Program Activity in Single Cells

TransCell: In silico Characterization of Genomic Landscape and Cellular Responses by Deep Transfer Learning

Deep Transfer Learning of Drug Responses by Integrating Bulk and Single-cell RNA-seq data

NIMG-60. PREDICTING OVERALL SURVIVAL IN GLIOBLASTOMA USING HISTOPATHOLOGY VIA AN END-TO-END DEEP LEARNING PIPELINE: A LARGE MULTI-COHORT STUDY

GILEA: In silico phenome profiling and editing using GAN Inversion

GeneBag: training a cell foundation model for broad-spectrum cancer diagnosis and prognosis with bulk RNA-seq data

Predicting Gene Spatial Expression and Cancer Prognosis: An Integrated Graph and Image Deep Learning Approach Based on HE Slides

Abstract 5361: Deep learning models capture high dimensional features for cell morphology analysis

Deep autoencoder for interpretable tissue-adaptive deconvolution and cell-type-specific gene analysis

Abstract 4304: Network-based inference identifies cell state-specific drugs targeting master regulator vulnerabilities in diffuse midline glioma

DIPG-66. MULTIPLEXED IMMUNOFLUORESCENCE MAPPING AND AI MODEL PREDICTION OF TUMOR MICROENVIRONMENT IN DIFFUSE MIDLINE GLIOMAS

An Intergraded Bioinformatics Approach for Identifying Novel Gene Networks in Glioblastoma

Deep learning links localized digital pathology phenotypes with transcriptional subtype and patient outcome in glioblastoma

GLDADec: marker-gene guided LDA modelling for bulk gene expression deconvolution

Interpretation of Disease Evidence for Medical Images Using Adversarial Deformation Fields

DeepCG: A cell graph model for predicting prognosis in lung adenocarcinoma

Transformer for Gene Expression Modeling (T-GEM): An Interpretable Deep Learning Model for Gene Expression-Based Phenotype Predictions

Transfer learning for clustering single-cell RNA-seq data crossing-species and batch, case on uterine fibroids