Abstract:Background Digital whole-slide images are a unique way to assess the spatial context of the career microenvironment. Exploring these spatial characteristics will enable us to better identify cross-level molecular markers that could deepen our understanding of cancer biology and related patient outcomes. Methods We proposed a graph neural network approach that emphasises spatialisation of tumour tiles towards a comprehensive evaluation of predicting cross-level molecular profiles of genetic mutations, copy number alterations, and functional protein expressions from whole-slide images. We introduced a transformation strategy that converts whole-slide image scans into graph-structured data to address the spatial heterogeneity of colon cancer. We developed and assessed the performance of the model on The Cancer Genome Atlas colon adenocarcinoma (TCGA-COAD) and validated it on two external datasets (ie, The Cancer Genome Atlas rectum adenocarcinoma [TCGA-READ] and Clinical Proteomic Tumor Analysis Consortium colon adenocarcinoma [CPTAC-COAD]). We also predicted microsatellite instability and result interpretability. Findings The model was developed on 459 colon tumour whole-slide images from TCGA-COAD, and externally validated on 165 rectum tumour whole-slide images from TCGA-READ and 161 colon tumour whole-slide images from CPTAC-COAD. For TCGA cohorts, our method accurately predicted the molecular classes of the gene mutations (area under the curve [AUCs] from 82.54 [95% CI 77.41-87.14] to 87.08 [83-28-90.82] on TCGA-COAD, and AUCs from 70.46 [61.37-79.61] to 81.80 [72.20-89.70] on TCGA-READ), along with genes with copy number alterations (AUCs from 81.98 [73.3449-68] to 90.55 [86.02-94.89] on TCGA-COAD, and AUCs from 62.05 [48.94-73-46] to 76.48 [64.78-86.71] on TCGA-READ), microsatellite instability (MSI) status classification (AUC 83.92 [77.41-87.59] on TCGA-COAD, and AUC 61.28 [53.28-67.93] on TCGA-READ), and protein expressions (AUCs from 85.57 [81.16-89.44] to 89.64 [86.29-93.19] on TCGA-COAD, and AUCs from 51.77 [42.53-61.83] to 59.79 [50.79-68.57] on TCGA-READ). For the CPTAC-COAD cohort, our model predicted a panel of gene mutations with AUC values from 63.74 (95% CI 52.92-75.37) to 82.90 (73.69-90.71), genes with copy number alterations with AUC values from 62.39 (51.37-73.76) to 86.08 (79.67-91.74), and MSI status prediction with AUC value of 73.15 (63.21-83.13). Interpretation We showed that spatially connected graph models enable molecular profile predictions in colon cancer and are generalised to rectum cancer. After further validation, our method could be used to infer the prognostic value of multiscale molecular biomarkers and identify targeted therapies for patients with colon cancer. Copyright (C) 2022 The Author(s). Published by Elsevier Ltd.

Spatially Aware Graph Neural Networks and Cross-Level Molecular Profile Prediction in Colon Cancer Histopathology: a Retrospective Multi-Cohort Study.