About

Zeyu Zhang is a Machine Learning Engineer at Meta, where he currently leads the video recommendation team at Instagram Reels. He has a rich background in data science and previously served as a data science manager at Supstat, managing the NYC Data Science Academy, an ACCET-accredited institution. Since 2015, Zeyu has been dedicated to shaping the next generation of data scientists by equipping them with essential skills and knowledge to secure data science positions across various organizations. In addition to his role as an educator, he has demonstrated his expertise by spearheading consulting projects for Fortune 500 companies, including PepsiCo, Aetna, and Barclays, delivering comprehensive end-to-end data science solutions to drive tangible results and provide actionable insights.

Research topics

• Genetics
• Computer Science
• Cancer research
• Biology
• Medicine
• Computational biology

Selected publications

• DIPG-49. A SYSTEMS BIOLOGY APPROACH TO DEFINING AND TARGETING CELL STATE-SPECIFIC MASTER REGULATOR DEPENDENCIES IN DIFFUSE MIDLINE GLIOMA

  Neuro-Oncology · 2023 · 1 citations

  • Cancer research
  • Biology
  • Computational biology

  Abstract Diffuse Midline Glioma (DMG) are fatal pediatric brain tumors. 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 a DMG regulatory network from 122 publicly available RNAseq profiles with ARACNe, and inferred sample-specific MR protein activity with VIPER. 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. OncoTreat predicted sensitivity to HDAC, MEK, CDK, PI3K, and proteosome inhibitors in subsets of patients. 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 from published scRNAseq profiles identified 6 tumor clusters with unique MR signatures representing distinct cellular states. Targetable MRs and OncoTreat-predicted drugs were distinct between these states. Bulk RNAseq analysis recapitulated predictions seen in the more prevalent Oligodendrocyte progenitor cell-like states, but failed to capture MR and drug predictions for the Astrocyte-like states. Ongoing validations of cell state-specific drug predictions in vivo in subcutaneous patient-derived xenograft and orthotopic syngeneic DMG models have already shown tumor volume and subpopulation differences (e.g. Trametinib-treated). This provides a platform to nominate much-needed novel drugs to treat DMG.

  PublisherOA PDFDOI

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

  Cancer Research · 2023 · 1 citations

  • Computer Science
  • Cancer research
  • Medicine

  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.

  PublisherDOI

Recent grants

• NIH Grant R01GM099722

  NIH · $1.2M · 2017

• Mechanism of Epigenetic Inheritance

  NIH · $8.6M · 2016–2026

• NIH Grant R01GM072719

  NIH · $2.9M · 2017

• NIH Grant R01GM081838

  NIH · $2.4M · 2017

• NIH Grant R01CA157489

  NIH · $1.6M · 2018

Frequent coauthors

• Hui Zhou

  150 shared

• Haiyun Gan

  Shenzhen Institutes of Advanced Technology

  137 shared

• Xu Hua

  94 shared

• Xu Zhang

  Peking University

  86 shared

• Shoufu Duan

  72 shared

• Albert Serra‐Cardona

  Columbia University

  70 shared

• Cheng–Chia Wu

  58 shared

• Fang Dong

  Western University of Health Sciences

  56 shared

Education

• PhD, Department of Biochemistry

  University of Utah

  1998

Similar researchers at Columbia University

• Eric R. Kandelh-248
• René Henh-177
• Peter D Kwongh-176
• Lawrence Elliot Friedh-171
• Tom Maniatish-162