Guojie Zhong

New York Genome Center

Ren Lab

101 Avenue of the Americas

New York, NY 10013

I am currently a Postdoctoral Research Associate in Dr. Bing Ren’s lab at the New York Genome Center.

I build biologically-inspired machine learning models to bridge the gap between complex diseases and effective treatments, deciphering disease mechanisms from high-throughput genomics data to identify new therapeutic targets.

My current research interests include:

Building DNA sequence-to-function foundation models that predict cell-type-specific transcriptional regulatory grammar in the human brain.
Neuropsychological disease mechanism discovery utilizing a combination of high-throughput screening techniques and deep learning.
Generative deep learning models for synthetic regulatory element design.

Previously, I had the honor of completing my PhD under the guidance of Dr. Yufeng Shen at the Department of Systems Biology, Columbia University.

My past work has spanned several areas including:

Single-cell spatial transcriptomics and cancer immunology: CSOmap, Hepatocellular Carcinoma
Statistical genetics and developmental disorders: VBASS, EA/TEF
Missense variant effect predictions: PreMode, RESCVE, MisFit

Prior to Columbia, I earned my B.S. in the Integrated Science Program at Peking University with training primarily in biology, statistics, and computer science. I was privileged to work in Dr. Zemin Zhang’s Lab for my undergraduate thesis on developing computational methods to infer cellular spatial organization and cellular interactions from single-cell genomics data.

In my free time, I enjoy playing fingerstyle guitar and badminton, both of which require a balance of power and control.

selected publications

PreMode predicts mode-of-action of missense variants by deep graph representation learning of protein sequence and structural context

G. Zhong, Y. Zhao, D. Zhuang, W. K. Chung, and Y. Shen

Nat Commun 2025

Abs Bib PDF Code

Accurate prediction of the functional impact of missense variants is important for disease gene discovery, clinical genetic diagnostics, therapeutic strategies, and protein engineering. Previous efforts have focused on predicting a binary pathogenicity classification, but the functional impact of missense variants is multi-dimensional. Pathogenic missense variants in the same gene may act through different modes of action (i.e., gain/loss-of-function) by affecting different aspects of protein function. They may result in distinct clinical conditions that require different treatments. We develop a new method, PreMode, to perform gene-specific mode-of-action predictions. PreMode models effects of coding sequence variants using SE(3)-equivariant graph neural networks on protein sequences and structures. Using the largest-to-date set of missense variants with known modes of action, we show that PreMode reaches state-of-the-art performance in multiple types of mode-of-action predictions by efficient transfer-learning. Additionally, PreMode’s prediction of G/LoF variants in a kinase is consistent with inactive-active conformation transition energy changes. Finally, we show that PreMode enables efficient study design of deep mutational scans and can be expanded to fitness optimization of non-human proteins with active learning.
@article{PreMode, author = {Zhong, G. and Zhao, Y. and Zhuang, D. and Chung, W. K. and Shen, Y.}, title = {PreMode predicts mode-of-action of missense variants by deep graph representation learning of protein sequence and structural context}, journal = {Nat Commun}, volume = {16}, number = {1}, pages = {7189}, year = {2025}, issn = {2041-1723 (Electronic) 2041-1723 (Linking)}, doi = {10.1038/s41467-025-62318-4}, url = {https://www.ncbi.nlm.nih.gov/pubmed/40764308}, google_scholar_id = {3fE2CSJIrl8C}, dimensions = {true} }
A probabilistic graphical model for estimating selection coefficients of nonsynonymous variants from human population sequence data

Y. Zhao, T. Lan, G. Zhong, J. Hagen, H. Pan, W. K. Chung, and Y. Shen

Nat Commun 2025

Abs Bib PDF Code

Accurately predicting the effect of missense variants is important in discovering disease risk genes and clinical genetic diagnostics. Commonly used computational methods predict pathogenicity, which does not capture the quantitative impact on fitness in humans. We develop a method, MisFit, to estimate missense fitness effect using a graphical model. MisFit jointly models the effect at a molecular level ( d ) and a population level (selection coefficient, s ), assuming that in the same gene, missense variants with similar d have similar s . We train it by maximizing probability of observed allele counts in 236,017 individuals of European ancestry. We show that s is informative in predicting allele frequency across ancestries and consistent with the fraction of de novo mutations in sites under strong selection. Further, s outperforms previous methods in prioritizing de novo missense variants in individuals with neurodevelopmental disorders. In conclusion, MisFit accurately predicts s and yields new insights from genomic data.
@article{MisFit, author = {Zhao, Y. and Lan, T. and Zhong, G. and Hagen, J. and Pan, H. and Chung, W. K. and Shen, Y.}, title = {A probabilistic graphical model for estimating selection coefficients of nonsynonymous variants from human population sequence data}, journal = {Nat Commun}, volume = {16}, number = {1}, pages = {4670}, year = {2025}, issn = {2041-1723 (Electronic) 2041-1723 (Linking)}, doi = {10.1038/s41467-025-59937-2}, url = {https://www.ncbi.nlm.nih.gov/pubmed/40393980}, google_scholar_id = {ZHo1McVdvXMC}, dimensions = {true} }
VBASS enables integration of single cell gene expression data in Bayesian association analysis of rare variants

G. Zhong, Y. A. Choi, and Y. Shen

Commun Biol 2023

Abs Bib PDF Code

Rare or de novo variants have substantial contribution to human diseases, but the statistical power to identify risk genes by rare variants is generally low due to rarity of genotype data. Previous studies have shown that risk genes usually have high expression in relevant cell types, although for many conditions the identity of these cell types are largely unknown. Recent efforts in single cell atlas in human and model organisms produced large amount of gene expression data. Here we present VBASS, a Bayesian method that integrates single-cell expression and de novo variant (DNV) data to improve power of disease risk gene discovery. VBASS models disease risk prior as a function of expression profiles, approximated by deep neural networks. It learns the weights of neural networks and parameters of Gamma-Poisson likelihood models of DNV counts jointly from expression and genetics data. On simulated data, VBASS shows proper error rate control and better power than state-of-the-art methods. We applied VBASS to published datasets and identified more candidate risk genes with supports from literature or data from independent cohorts. VBASS can be generalized to integrate other types of functional genomics data in statistical genetics analysis.
@article{VBASS, author = {Zhong, G. and Choi, Y. A. and Shen, Y.}, title = {VBASS enables integration of single cell gene expression data in Bayesian association analysis of rare variants}, journal = {Commun Biol}, volume = {6}, number = {1}, pages = {774}, year = {2023}, note = {Commun Biol. 2023 Jul 25;6(1):774. doi: 10.1038/s42003-023-05155-9.}, issn = {2399-3642 (Electronic) 2399-3642 (Linking)}, doi = {10.1038/s42003-023-05155-9}, url = {https://www.ncbi.nlm.nih.gov/pubmed/37491581}, google_scholar_id = {Se3iqnhoufwC}, dimensions = {true} }
MLSB 2022
Representation of missense variants for predicting modes of action

G. Zhong, and Y. Shen

Machine Learning in Structural Biology, Workshop at the 36th Conference on Neural Information Processing Systems (NeurIPS) 2022

Abs Bib PDF Code

Accurate prediction of functional impact for missense variants is fundamental for genetic analysis and clinical applications. Current methods focused on generating an overall pathogenicity prediction score while overlooking the fact that variant effect should be multi-dimensional via different modes of action, such as gain or loss of function, and loss of folding stability or enzymatic activity. Recent breakthrough of high-capacity language models enabled ab initio prediction of protein structures as well as self-supervised representation learning of protein sequence and functions. Here we present RESCVE, a method to learn universal representation of sequence variation from protein context. We demonstrated the utility of the method predicting a range of modes of action for missense variants through transfer learning.
@article{RESCVE, author = {Zhong, G. and Shen, Y.}, title = {Representation of missense variants for predicting modes of action}, booktitle = {Machine Learning in Structural Biology, Workshop at the 36th Conference on Neural Information Processing Systems}, journal = {Machine Learning in Structural Biology, Workshop at the 36th Conference on Neural Information Processing Systems (NeurIPS)}, year = {2022}, google_scholar_id = {_FxGoFyzp5QC} }
Statistical models of the genetic etiology of congenital heart disease

G. Zhong, and Y. Shen

Curr Opin Genet Dev 2022

Abs Bib PDF

Congenital heart disease (CHD) is a collection of anatomically and clinically heterogeneous structure anomalies of heart at birth. Finding genetic causes of CHD can not only shed light on developmental biology of heart, but also provide basis for improving clinical care and interventions. The optimal study design and analytical approaches to identify genetic causes depend on the underlying genetic architecture. A few well-known syndromes with CHD as core conditions, such as Noonan and CHARGE, have known monogenic causes. The genetic causes of most of CHD patients, however, are unknown and likely to be complex. In this review, we highlight recent studies that assume a complex genetic architecture of CHD with two main approaches. One is genomic sequencing studies aiming for identifying rare or de novo risk variants with large genetic effect. The other is genome-wide association studies optimized for common variants with moderate genetic effect.
@article{StatsGenReview, author = {Zhong, G. and Shen, Y.}, title = {Statistical models of the genetic etiology of congenital heart disease}, journal = {Curr Opin Genet Dev}, volume = {76}, pages = {101967}, year = {2022}, issn = {1879-0380 (Electronic) 0959-437X (Linking)}, doi = {10.1016/j.gde.2022.101967}, url = {https://www.ncbi.nlm.nih.gov/pubmed/35939966}, google_scholar_id = {W7OEmFMy1HYC}, dimensions = {true} }
Identification and validation of candidate risk genes in endocytic vesicular trafficking associated with esophageal atresia and tracheoesophageal fistulas

G. Zhong*, P. Ahimaz*, N. A. Edwards*, J. J. Hagen, C. Faure, Q. Lu, P. Kingma, W. Middlesworth, J. Khlevner, M. El Fiky, D. Schindel, E. Fialkowski, A. Kashyap, S. Forlenza, A. P. Kenny, A. M. Zorn, Y. Shen, and W. K. Chung

HGG Adv 2022

Abs Bib PDF Code

Esophageal atresias/tracheoesophageal fistulas (EA/TEF) are rare congenital anomalies caused by aberrant development of the foregut. Previous studies indicate that rare or de novo genetic variants significantly contribute to EA/TEF risk, and most individuals with EA/TEF do not have pathogenic genetic variants in established risk genes. To identify novel genetic contributions to EA/TEF, we performed whole genome sequencing of 185 trios (probands and parents) with EA/TEF, including 59 isolated and 126 complex cases with additional congenital anomalies and/or neurodevelopmental disorders. There was a significant burden of protein altering de novo coding variants in complex cases (p=3.3e-4), especially in genes that are intolerant of loss of function variants in the population. We performed simulation analysis of pathway enrichment based on background mutation rate and identified a number of pathways related to endocytosis and intracellular trafficking that as a group have a significant burden of protein altering de novo variants. We assessed 18 variants for disease causality using CRISPR-Cas9 mutagenesis in Xenopus and confirmed 13 with tracheoesophageal phenotypes. Our results implicate disruption of endosome-mediated epithelial remodeling as a potential mechanism of foregut developmental defects. This research may have implications for the mechanisms of other rare congenital anomalies.
@article{EA_TEF_HGG, author = {Zhong*, G. and Ahimaz*, P. and Edwards*, N. A. and Hagen, J. J. and Faure, C. and Lu, Q. and Kingma, P. and Middlesworth, W. and Khlevner, J. and El Fiky, M. and Schindel, D. and Fialkowski, E. and Kashyap, A. and Forlenza, S. and Kenny, A. P. and Zorn, A. M. and Shen, Y. and Chung, W. K.}, title = {Identification and validation of candidate risk genes in endocytic vesicular trafficking associated with esophageal atresia and tracheoesophageal fistulas}, journal = {HGG Adv}, volume = {3}, number = {3}, pages = {100107}, year = {2022}, issn = {2666-2477 (Electronic) 2666-2477 (Linking)}, doi = {10.1016/j.xhgg.2022.100107}, url = {https://www.ncbi.nlm.nih.gov/pubmed/35519826}, google_scholar_id = {zYLM7Y9cAGgC}, dimensions = {true} }
Reconstruction of cell spatial organization from single-cell RNA sequencing data based on ligand-receptor mediated self-assembly

X. Ren*, G. Zhong*, Q. Zhang, L. Zhang, Y. Sun, and Z. Zhang

Cell Res 2020

Abs Bib PDF Code

Single-cell RNA sequencing (scRNA-seq) has revolutionized transcriptomic studies by providing unprecedented cellular and molecular throughputs, but spatial information of individual cells is lost during tissue dissociation. While imaging-based technologies such as in situ sequencing show great promise, technical difficulties currently limit their wide usage. Here we hypothesize that cellular spatial organization is inherently encoded by cell identity and can be reconstructed, at least in part, by ligand-receptor interactions, and we present CSOmap, a computational tool to infer cellular interaction de novo from scRNA-seq. We show that CSOmap can successfully recapitulate the spatial organization of multiple organs of human and mouse including tumor microenvironments for multiple cancers in pseudo-space, and reveal molecular determinants of cellular interactions. Further, CSOmap readily simulates perturbation of genes or cell types to gain novel biological insights, especially into how immune cells interact in the tumor microenvironment. CSOmap can be a widely applicable tool to interrogate cellular organizations based on scRNA-seq data for various tissues in diverse systems.
@article{CSOmap, author = {Ren*, X. and Zhong*, G. and Zhang, Q. and Zhang, L. and Sun, Y. and Zhang, Z.}, title = {Reconstruction of cell spatial organization from single-cell RNA sequencing data based on ligand-receptor mediated self-assembly}, journal = {Cell Res}, volume = {30}, number = {9}, pages = {763-778}, year = {2020}, issn = {1748-7838 (Electronic) 1001-0602 (Linking)}, doi = {10.1038/s41422-020-0353-2}, url = {https://www.ncbi.nlm.nih.gov/pubmed/32541867}, google_scholar_id = {d1gkVwhDpl0C}, dimensions = {true} }
Landscape and Dynamics of Single Immune Cells in Hepatocellular Carcinoma

Q. Zhang, Y. He, N. Luo, S. J. Patel, Y. Han, R. Gao, M. Modak, S. Carotta, C. Haslinger, D. Kind, G. W. Peet, G. Zhong, S. Lu, W. Zhu, Y. Mao, M. Xiao, M. Bergmann, X. Hu, S. P. Kerkar, A. B. Vogt, S. Pflanz, K. Liu, J. Peng, X. Ren, and Z. Zhang

Cell 2019

Abs Bib PDF

The immune microenvironment of hepatocellular carcinoma (HCC) is poorly characterized. Combining two single-cell RNA sequencing technologies, we produced transcriptomes of CD45+ immune cells for HCC patients from five immune-relevant sites: tumor, adjacent liver, hepatic lymph node (LN), blood, and ascites. A cluster of LAMP3+ dendritic cells (DCs) appeared to be the mature form of conventional DCs and possessed the potential to migrate from tumors to LNs. LAMP3+ DCs also expressed diverse immune-relevant ligands and exhibited potential to regulate multiple subtypes of lymphocytes. Of the macrophages in tumors that exhibited distinct transcriptional states, tumor-associated macrophages (TAMs) were associated with poor prognosis, and we established the inflammatory role of SLC40A1 and GPNMB in these cells. Further, myeloid and lymphoid cells in ascites were predominantly linked to tumor and blood origins, respectively. The dynamic properties of diverse CD45+ cell types revealed by this study add new dimensions to the immune landscape of HCC.
@article{HCC_SingleCell, author = {Zhang, Q. and He, Y. and Luo, N. and Patel, S. J. and Han, Y. and Gao, R. and Modak, M. and Carotta, S. and Haslinger, C. and Kind, D. and Peet, G. W. and Zhong, G. and Lu, S. and Zhu, W. and Mao, Y. and Xiao, M. and Bergmann, M. and Hu, X. and Kerkar, S. P. and Vogt, A. B. and Pflanz, S. and Liu, K. and Peng, J. and Ren, X. and Zhang, Z.}, title = {Landscape and Dynamics of Single Immune Cells in Hepatocellular Carcinoma}, journal = {Cell}, volume = {179}, number = {4}, pages = {829-845 e20}, year = {2019}, issn = {1097-4172 (Electronic) 0092-8674 (Linking)}, doi = {10.1016/j.cell.2019.10.003}, url = {https://www.ncbi.nlm.nih.gov/pubmed/31675496}, google_scholar_id = {u5HHmVD_uO8C}, dimensions = {true} }