• Hongyu Zhao, Yale University, USA

Understanding cellular responses to genetic perturbations is essential for deciphering gene regulation and phenotype formation. While high-throughput single-cell RNA-sequencing has facilitated detailed profiling of heterogeneous transcriptional responses to perturbations at the single-cell level, there remains a pressing need for computational models that can decode the mechanisms driving these responses and accurately predict outcomes to prioritize target genes for experimental design. This presentation will introduce a deep generative learning framework designed to model and predict single-cell transcriptional responses to genetic perturbations, including single-gene and combinatorial multi-gene perturbations. Our method can effectively integrate prior biological knowledge and disentangle basal cell states from perturbation-specific salient representations by leveraging gene embeddings derived from large language models. Through comprehensive evaluations on multiple single-cell CRISPR Perturb-seq datasets, our method outperformed state-of-the-art methods in predicting perturbation outcomes, achieving higher prediction accuracy. Notably, it demonstrated robust generalization to unseen target genes and perturbations, and its predictions captured both average expression changes and the heterogeneity of single-cell responses. Furthermore, its predictions enable diverse downstream analyses, including identifying differentially expressed genes and exploring genetic interactions, demonstrating its utility and versatility. This is joint work with Gefei Wang, Tianyu Liu, Jia Zhao, and Youshu Cheng.

• Kai Ye, Xi'an Jiaotong University, China

Seq2Image is a computational framework that transforms sequential biological data (e.g., DNA, RNA, protein sequences) into structured 2D image representations. By encoding multidimensional sequence features into spatially resolved visual patterns, this strategy enables both automated analysis by visual AI models (e.g., Convolutional Neural Networks) and enhanced human interpretation of complex genomic information. We demonstrate its utility through two key applications:
1. Complex Structural Variant (CSV) Detection: Identifying nested rearrangements in individual genomes via CNN, leveraging sequence-depth images to resolve breakpoints with pixel-level precision.
2. Comparative Genomics: Detecting somatic or de novo variants through Difference Imaging, a method that overlays tumor-normal or parent-child genome alignments to highlight discordant regions as contrast-enhanced features.

• Liang Huang

This talk highlights some highly unexpected connections between biology and linguistics. For example, our Nature (2023) paper designed highly stable and efficient messenger RNA (mRNA) vaccines using natural language processing algorithms. Experiments on COVID and another virus show that our designs dramatically improves mRNA half-life, protein expression, and in vivo antibody response, compared to the standard method used by Pfizer and Moderna. Nature News reported our work as a “remarkable AI tool” for mRNA design. Time permitting, I will also present some other recent work on RNA design.

• Xun Xu

• Xiao-Wo Wang
• Ji-Guang Wang
• Xun Xu
• Zhang Zhang

• Jian Ma

• James Zou

Large language models (LLMs), such as ChatGPT, have read millions of papers and contains tremendous biomedical knowledge. In language-guided biology, we propose a framework using LLMs as an informative prior to integrate domain knowledge and guide downstream analyses. I will demonstrate this approach through GenePT, where we use LLM embeddings of genes to improve perturbation predictions and single-cell analysis. I will then explore extensions to spatial biology and protein annotation.

• Jian Ma
• James Zou
• Yong Wang
• Zhi-Hua Zhang
• Xing-Ming Zhao
• Guo-Liang Li