PQG Seminar

Mengjie Chen

Associate Professor of Medicine
Associate Professor of Human Genetics
University of Chicago

Beyond variability: a novel gene expression stability metric to unveil homeostasis and regulation

A homeostatic cell performs regular functions to maintain internal balance by continuously responding to both internal and external stimuli, a process that often involves transcriptional regulation. Most genes within such cells exhibit transcriptional stability, while a smaller subset may enter a regulatory or compensatory state in response to stimuli. Key candidates for this type of regulation include ‘first responder’ genes, interferons, and heat shock proteins, among others. When these responses accumulate to a certain threshold, they can lead to observable phenotypic changes and, in some cases, pathological outcomes. Therefore, identifying genes with precise regulation within homeostatic cells is crucial.

Existing statistical tools have mainly focused on cells with uniform behaviors, often overlooking the nuanced regulation of genes in specific cell subsets. In this presentation, I will discuss an unexpected journey that starts with modeling zero-inflation in single-cell data and progresses to the introduction of the Gene Homeostasis Z-index—a novel metric for gene expression stability. This index reveals genes undergoing precise regulation within specific cell subsets, offering insights into their roles in cellular adaptation. For example, we discover regulatory patterns for neuropeptides like insulin and somatostatin, which exhibit extreme values in a limited number of cells. These findings highlight the limitations of conventional mean-based approaches and demonstrate how our method provides a more refined understanding of gene expression stability.

Bo Wang

Lead Scientist of the Artificial Intelligence Team for Peter Munk Cardiac Centre at University Health Network
Assistant Professor, Departments of Computer Science and Laboratory Medicine & Pathobiology
University of Toronto

Building Foundation Models for Single-cell Omics and Imaging

This talk delves into the innovative utilization of generative AI in propelling biomedical research forward. By harnessing single-cell sequencing data, we developed scGPT, a foundational model that extracts biological insights from an extensive dataset of over 33 million cells. Analogous to how words form text, genes define cells, effectively bridging the technological and biological realms. The strategic application of scGPT via transfer learning significantly boosts its efficacy in diverse applications such as cell-type annotation, multi-batch integration, and gene network inference.Additionally, the talk will spotlight MedSAM, a state-of-the-art segmentation foundational model. Designed for universal application, MedSAM excels across various medical imaging tasks and modalities. It showcased unprecedented advancements in 30 segmentation tasks, outperforming existing models considerably. Notably, MedSAM possesses the unique ability for zero-shot and few-shot segmentation, enabling it to identify previously unseen tumor types and swiftly adapt to novel imaging modalities.Collectively, these breakthroughs emphasize the importance of developing versatile and efficient foundational models. These models are poised to address the expanding needs of imaging and omics data, thus driving continuous innovation in biomedical analysis.

Qing Nie
Distinguished Professor of Mathematics and Developmental & Cell Biology
UC Irvine

Systems Learning of Single Cells

Cells make fate decisions in response to dynamic environments, and multicellular structures emerge from multiscale interplays among cells and genes in space and time. While single-cell omics data provides an unprecedented opportunity to profile cellular heterogeneity, the technology requires fixing the cells, often leading to a loss of spatiotemporal and cell interaction information. How to reconstruct temporal dynamics from single or multiple snapshots of single-cell omics data? How to recover interactions among cells, for example, cell-cell communication from single-cell gene expression data? I will present a suite of our recently developed computational methods that learn the single-cell omics data as a spatiotemporal and interactive system. Those methods are built on a strong interplay among systems biology modeling, dynamical systems approaches, machine-learning methods, and optimal transport techniques. The tools are applied to various complex biological systems in development, regeneration, and diseases to show their discovery power. Finally, I will discuss the methodology challenges in systems learning of single-cell data.

Jian Ma
Ray and Stephanie Lane Professor of Computational Biology
School of Computer Science
Carnegie Mellon University

Learning Multiscale Genome and Cellular Organization

Despite significant advancements in high-throughput data acquisition in genomics and cell biology, our understanding of the diverse cell types within the human body remains limited. The principles governing intracellular molecular spatial organization and interactions, as well as cellular spatial organization within complex tissues, are still largely unclear. A major challenge lies in developing computational methods capable of integrating heterogeneous, multiscale molecular, cellular, and tissue information. In this talk, I will discuss our work on developing machine learning approaches to advance regulatory genomics through single-cell 3D epigenomics. Additionally, I will introduce our recent efforts in creating interpretable, self-supervised models for the multiscale delineation of cellular interactions in tissues. These methods hold the potential to reveal new insights into fundamental genome structure, gene regulation, and cellular function across a wide range of biological contexts in both health and disease.

Iuliana Ionita-Laza
Professor of Biostatistics
Mailman School of Public Health, Columbia University

Quantile regression in genomics: a new lens for genetic discovery and phenotype prediction

Genome-wide association studies (GWAS) for biomarkers can lead to clinically relevant discoveries. Numerous lines of evidence from both model organisms and human studies suggest that genetic associations can be highly heterogeneous, dynamic and context dependent. Despite twenty years of GWAS, most studies are based on statistical models that fail to account for such heterogeneity. In this talk I will discuss alternative approaches based on quantile regression (QR) models that naturally extend linear regression models to the analysis of the entire conditional distribution of a phenotype of interest. I will illustrate the advantages of QR in the context of genetic discovery, showing improved power to identify heterogeneous genetic associations that may have larger effects on high-risk subgroups of individuals but with lower or no contribution overall, and increased ability to adjust for subtle population structure.
Furthermore, we investigate applications of QR to uncertainty quantification for polygenic risk score (PRS) prediction. QR shifts the focus from predicting the conditional phenotypic mean in classical PRS to predicting the conditional phenotypic quantiles, essentially allowing us to predict the entire phenotype distribution. When combined with conformal prediction, this framework offers a natural way to construct prediction intervals that give the lower and upper bounds where the phenotype may lie with high probability.
I will show applications to data from the UK biobank and the ProgeNIA/SardiNIA projects.

Mengdi Wang
Associate Professor of Electrical and Computer Engineering and the Center for Statistics and Machine Learning
Princeton University

From Genome to Theorem: Can Large Language Models Do Science?

Large Language Models (LLMs) are increasingly being explored as tools for scientific reasoning — not just in language tasks, but across disciplines such as mathematics, biology, and genomics. In this talk, I’ll discuss recent developments in AI for science, including genome language models, AI gene-editing co-scientist, and LLMs for mathematical problem solving. I’ll highlight both the capabilities and current limitations of LLMs, and discuss key gaps between model abstractions and the realities of scientific workflows. As we push toward AI systems that can assist with discovery, the question remains: can LLMs truly do science — or are we still in the early stages of bridging that divide?