7-UP:generatingin silicoCODEX from a small set of immunofluorescence markers

Abstract

Multiplex immunofluorescence (mIF) assays multiple protein biomarkers on a single tissue section. Recently, high-plex CODEX (co-detection by indexing) systems enable simultaneous imaging of 40+ protein biomarkers, unlocking more detailed molecular phenotyping, leading to richer insights into cellular interactions and disease. However, high-plex imaging can be slower and more costly to collect, limiting its applications, especially in clinical settings. We propose a machine learning framework, 7-UP, that can computationally generate in silico 40-plex CODEX at single-cell resolution from a standard 7-plex mIF panel by leveraging cellular morphology. We demonstrate the usefulness of the imputed biomarkers in accurately classifying cell types and predicting patient survival outcomes. Furthermore, 7-UP's imputations generalize well across samples from different clinical sites and cancer types. 7-UP opens the possibility of in silico CODEX, making insights from high-plex mIF more widely available.

Expanding the coverage of spatial proteomics

“…Instead of selecting a unified panel to enable cell type prediction across multiple tissues, our approach can produce biomarker panels optimized for specific tissues. Our results demonstrate that panels selected by our method outperform those produced by (20). In addition, we consider a more difficult setting in which all markers of interest cannot be imaged simultaneously.…”

“…4, and randomly cropped small patches were used to train neural networks to minimize the MSE between predicted marker intensity and observed intensity (see details in Methods). First, we compared the performance of models from our method and from (20) in terms of MSE reconstruction loss and Pearson correlation coefficients (PCC). Our method outperforms the 7-UP and 7-UP alternative on test MSE and PCC between predicted and real images, as shown in Figure 2 (a).…”

“…This concept can be extended to predict a large number of biomarkers from images of a smaller number (11,12). For example, Wu et al (13) described a method to select 7 markers out of 40 that enabled accurate prediction of cell types in a number of tissues, and showed the effectiveness of the approach by imaging only those 7. In this work, we first sought to develop a flexible approach for finding a small subset of markers and using them to predict the full-image expression pattern of the remaining markers. In contrast to the method of Wu et al (13), we consider the problem of marker selection from an optimization standpoint; our single-panel setting focuses on selecting biomarkers by explicitly modeling the spatial relationship between marker intensities with a graphical model and using a neural network to directly infer the multiplexed fluorescence image instead of predicting expression at the single-cell level.…”

“…This concept can be extended to predict a large number of biomarkers from images measuring a smaller number. For example, (20) described a method to select 7 informative markers out of 40 markers that would enable accurate pre-diction of cell types, and showed the effectiveness of the approach by performing 7-plex imaging (21). Their method began by converting multiplexed images to biomarker levels in each cell, and then learned to predict marker intensities at the single-cell level from the 7 measured marker intensities as well as cell morphology features learned separately by another machine learning model.…”

“…In this work, we introduce a new heuristic method for iteratively selecting a set of biomarkers that allow accurate prediction of full images. In contrast to the method of (20), we consider the problem of marker selection from an optimization standpoint, and our method focuses on selecting biomarkers by explicitly modeling the spatial relationship between marker intensities and uses a neural network to directly infer the multiplexed fluorescence image instead of predicting expression at the single-cell level. Instead of selecting a unified panel to enable cell type prediction across multiple tissues, our approach can produce biomarker panels optimized for specific tissues.…”