Multiplexed Imaging for Biomarker Discovery and Drug Mechanism Evaluation

Summary

• Tissue or organ-level changes in gene expression patterns often precede the onset or progression of disease
• Single-cell sequencing data overlaid on maps of tissue architecture, termed spatial transcriptomics, is a powerful method yet requires time-consuming manual annotation and the risk of user bias
• The Elemento lab has been pioneering a computational method to identify and quantify organ-specific anatomical domains from biological images without any previous knowledge of disease state
• By combining this multiplexed imaging framework with single-cell phenotypic data (e.g. cell morphology, gene expression, cell surface markers), the EIPM has a powerful tool which can yield therapeutically actionable insights into disease states and drug mechanisms

Technical Overview

• Advancements in spatial transcriptomics are providing researchers the opportunity to directly characterize cellular phenotypes and interactions in intact, heterogeneous tissues
• However, the technology has been unable to move beyond “cell-centric” insights to uncover organizing principles to tissue architecture and organ-specific tissue pathophysiology
• Dr. Elemento and the EIPM have developed an accurate, unsupervised method to discover and quantify “microanatomical tissue structures” using multiplexed histopathology images
• The deep learning model provides novel insights on cell morphology and gene expression, coupled to physical proximity data at a whole-organ scale in both healthy and diseased tissues

Market Opportunity

• Advancements in spatial transcriptomics are providing researchers the opportunity to directly characterize cellular phenotypes and interactions in intact, heterogeneous tissues
• However, the technology has been unable to move beyond “cell-centric” insights to uncover organizing principles to tissue architecture and organ-specific tissue pathophysiology
• Dr. Elemento and the EIPM have developed an accurate, unsupervised method to discover and quantify “microanatomical tissue structures” using multiplexed histopathology images
• The deep learning model provides novel insights on cell morphology and gene expression, coupled to physical proximity data at a whole-organ scale in both healthy and diseased tissues

Partnering Opportunity

Weill Cornell Medicine is seeking a partner with an interest in leveraging EIPM’s comprehensive patient biobank and deep learning model to improve biomarker strategy and clinical development insight

Supporting Data / Figures

Figure 1: Overview of UTAG, unsupervised discovery of tissue architecture with graphs, deep learning model for biomarker discovery (Kim 2024 Nature).