Multi-modal learning (e.g., integrating pathological images with genomic features) tends to improve the accuracy of cancer diagnosis and prognosis as compared to learning with a single modality. However, missing data is a common problem in clinical practice, i.e., not every patient has all modalities available. Most of the previous works directly discarded samples with missing modalities, which might lose information in these data and increase the likelihood of overfitting. In this work, we generalize the multi-modal learning in cancer diagnosis with the capacity of dealing with missing data using histological images and genomic data. Our integrated model can utilize all available data from patients with both complete and partial modalities. The experiments on the public TCGA-GBM and TCGA-LGG datasets show that the data with missing modalities can contribute to multi-modal learning, which improvesthe model performance in grade classification of glioma cancer.
Most colorectal cancers (CRCs) develop from either adenomas (ADs) or sessile serrated lesions (SSLs). The origins and molecular landscapes of these histologically distinct pre-cancerous polyps remain incompletely understood. Here, we present an atlas at single-cell resolution of sporadic conventional tubular/tubulovillous ADs, SSLs, hyperplastic polyps (HPs), microsatellite stable (MSS) and unstable (MSI-H) CRC, and normal colonic mucosa. Using single-cell transcriptomics and multiplex imaging, we studied 69 datasets from 33 participants. We also examined separate sets of 66 and 274 polyps for RNA and targeted gene sequencing, respectively. We performed multiplex imaging on a tissue microarray of 14 ADs and 15 CRCs, and we integrated pre-cancer polyp data with published single-cell and The Cancer Genome Atlas (TCGA) bulk CRC data to establish potential polyp-cancer relationships. Striking differences were observed between ADs and SSLs that extended to MSS and MSI-H CRCs, respectively, reflecting their distinct origins and trajectories. ADs arose from WNT pathway dysregulation in stem cells, which aberrantly expanded and expressed a Hippo and ASCL2 regenerative program. In marked contrast, SSLs were depleted of stem cell-like populations and instead exhibited a program of gastric metaplasia in the setting of elevated cytotoxic inflammation. Using subtype-specific gene regulatory networks and shared genetic variant analysis, we implicated serrated polyps, including some HPs conventionally considered benign, as arising from a metaplastic program in committed absorptive cells. ADs and SSLs displayed distinct patterns of immune cell infiltration that may influence their natural history. Our multi-omic atlas provides novel insights into the malignant potential of colorectal polyps and serves as a framework for precision surveillance and prevention of sporadic CRC.
Motivation: Multiplexed immunofluorescence (mIF) is an emerging assay for multichannel protein imaging that can decipher cell-level spatial features in tissues. However, existing automated cell phenotyping methods, such as clustering, face challenges in achieving consistency across experiments and often require subjective evaluation. As a result, mIF analyses often revert to marker gating based on manual thresholding of raw imaging data. Results: To address the need for an evaluable semi-automated algorithm, we developed GammaGateR, an R package for interactive marker gating designed specifically for segmented cell-level data from mIF images. Based on a novel closed-form gamma mixture model, GammaGateR provides estimates of marker-positive cell proportions and soft clustering of marker-positive cells. The model incorporates user-specified constraints that provide a consistent but slide-specific model fit. We compared GammaGateR against the newest unsupervised approach for annotating mIF data, employing two colon datasets and one ovarian cancer dataset for the evaluation. We showed that GammaGateR produces highly similar results to a silver standard established through manual annotation. Furthermore, we demonstrated its effectiveness in identifying biological signals, achieved by mapping known spatial interactions between CD68 and MUC5AC cells in the colon and by accurately predicting survival in ovarian cancer patients using the phenotype probabilities as input for machine learning methods. GammaGateR is a highly efficient tool that can improve the replicability of marker gating results, while reducing the time of manual segmentation. Availability and Implementation: The R package is available at https://github.com/JiangmeiRubyXiong/GammaGateR . Contact: Please address correspondence to jiangmei.xiong@vanderbilt.edu and simon.vandekar@vumc.org . Key Points: GammaGateR is the first semi-automated marker gating tool for mIF image, and it will help to diminish the inconsistency with manual marker gating. With novel cfGMM, GammaGateR can fit flexibly across slides with different distributions and incorporate biology priors. GammaGateR is proven to reveal credible prognostic information, and it can quantify known findings in tumor cell populations.
Spatially resolved molecular assays provide high dimensional genetic, transcriptomic, proteomic, and epigenetic information in situ and at various resolutions. Pairing these data across modalities with histological features enables powerful studies of tissue pathology in the context of an intact microenvironment and tissue structure. Increasing dimensions across molecular analytes and samples require new data science approaches to functionally annotate spatially resolved molecular data. A specific challenge is data-driven cross-sample domain detection that allows for analysis within and between consensus tissue compartments across high volumes of multiplex datasets stemming from tissue atlasing efforts. Here, we present MILWRM – multiplex image labeling with regional morphology – a Python package for rapid, multi-scale tissue domain detection and annotation. We demonstrate MILWRM’s utility in identifying histologically distinct compartments in human colonic polyps and mouse brain slices through spatially-informed clustering in two different spatial data modalities. Additionally, we used tissue domains detected in human colonic polyps to elucidate molecular distinction between polyp subtypes. We also explored the ability of MILWRM to identify anatomical regions of mouse brain and their respective distinct molecular profiles.