PQG Seminar Series

Luke O’Connor

Associate Member
Broad Institute

Linkage disequilibrium graphical models (LDGMs) are an efficient new approach to model LD, with promising applications in GWAS. They give rise to a very sparse precision matrix whose inverse closely approximates the LD correlation matrix. My talk will cover what LDGMs are, how they are inferred, and why they are useful. Then I will discuss two applications, to polygenic risk prediction and heritability partitioning with GWAS summary statistics.

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.

In the burgeoning field of spatial biology, the integration of multi-modal spatial omics technologies presents both a formidable challenge and a tremendous opportunity for advancing clinical research and diagnostics. I will discuss the concerted efforts of our laboratory to address the complexities inherent in large-scale multi-modal spatial data analysis, with a specific focus on making spatial biology more accessible for clinical projects. Our approach is threefold: firstly, we focus on implementing the latest spatial omics technologies with the goal to integrate their functional outputs and as such harness the full potential of spatially resolved molecular data.  Secondly, we develop robust data structures tailored for the efficient storage, retrieval, and manipulation of large volumes of multi-modal spatial data, ensuring that our solutions are scalable and adaptable to the ever-evolving landscape of spatial biology. Finally, we prioritize the usability of our analytical tools and strategies, offering a user-friendly interface that empowers clinicians and researchers with minimal computational background to engage in sophisticated spatial data analysis. By addressing these key areas, our laboratory not only aims to advance the methodological framework for spatial data analysis but also to foster the integration of spatial omics data into routine clinical practice, thereby opening new avenues for personalized medicine and biomarker discovery. Through this integrated approach, we contribute to the establishment of a more accessible, efficient, and comprehensive ecosystem for the analysis of spatial biology data, ultimately facilitating the translation of complex spatial omics data into actionable clinical insights.

Rare variants significantly impact complex diseases. This presentation will first introduce SAIGE-GENE and SAIGE-GENE+, methodologies extending SAIGE to gene/region-based rare variant tests. These methods efficiently utilize mixed effects models to adjust for sample relatedness and saddlepoint approximations to account for case-control imbalance. SAIGE-GENE+ additionally incorporates functional annotations and collapsing of ultra-rare variants that can help to improve type I error control and power. In the second part of the talk, I will introduce our recent work to estimate effect sizes of rare variants. The method, RareEffect, uses an empirical Bayesian approach that estimates gene/region-level heritability and then an effect size of each variant. We also show the effect sizes obtained from our model can be leveraged to improve the performance of polygenic scores.