Computational multiplex panel reduction to maximize information retention in breast cancer tissue microarrays

Abstract

Recent state-of-the-art multiplex imaging techniques have expanded the depth of information that can be captured within a single tissue sample by allowing for panels with dozens of markers. Despite this increase in capacity, space on the panel is still limited due to technical artifacts, tissue loss, and long imaging acquisition time. As such, selecting which markers to include on a panel is important, since removing important markers will result in a loss of biologically relevant information, but identifying redundant markers will provide a room for other markers. To address this, we propose computational approaches to determine the amount of shared information between markers and select an optimally reduced panel that captures maximum amount of information with the fewest markers. Here we examine several panel selection approaches and evaluate them based on their ability to reconstruct the full panel images and information within breast cancer tissue microarray datasets using cyclic immunofluorescence as a proof of concept. We show that all methods perform adequately and can re-capture cell types using only 18 of 25 markers (72% of the original panel size). The correlation-based selection methods achieved the best single-cell marker mean intensity predictions with a Spearman correlation of 0.90 with the reduced panel. Using the proposed methods shown here, it is possible for researchers to design more efficient multiplex imaging panels that maximize the amount of information retained with the limited number of markers with respect to certain evaluation metrics and architecture biases.

Expanding the coverage of spatial proteomics

“…Examples include predicting subcellular components from unlabeled microscope images (2)(3)(4)(5), virtual histological staining of tissue images (6)(7)(8), and predicting immunofluorescence or directly inferring cell types from immunohistochemically stained images (9,10). 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.…”

MIM-CyCIF: Masked Imaging Modeling for Enhancing Cyclic Immunofluorescence (CyCIF) with Panel Reduction and Imputation

“…We evaluate different masking ratios for training by assessing the performance of different reduced panel sizes in inference on the BC TMA dataset. For testing, we choose the optimal reduced panels identi ed in Ternes et al 10 , which have sizes of 3,6,9,12,15, and 18 markers (88%, 76%, 64%, 52%, 40%, and 28% masking ratios, respectively). We train three models using a xed masking ratio of 25%, 50%, and 75%, and nd that the 50% masking ratio results in the best overall performance across different panel sizes in inference (Supplementary Fig.…”

“…Thus, the selection of markers for the panel becomes crucial, with the aim of interrogating a wide spectrum of cell states and phenotypes 5,8,9 .Previous studies computationally optimized MTI panel reduction and prediction. Ternes et al 10 pioneered CyCIF panel reduction and imputation using a two-step approach: exploring multiple strategies for marker selection and training a multi-encoder variational autoencoder (ME-VAE) 6 to reconstruct the full 25-plex CyCIF images at the single cell level. Wu et al proposed a three-step method using a concrete autoencoder and convolution neural network to reduce CODEX markers and predict intensity via a linear regression model 11 .…”

Dual-modality imaging of immunofluorescence and imaging mass cytometry for whole slide imaging with accurate single-cell segmentation

“…It results in grave difficulties in downstream image registration and introduces significant errors. Moreover, autofluorescence and experimental batch effects also introduce biases into the data and may mask the true underlying biological signals [6][7][8][9][10] .…”