Colorectal cancer (CRC) is now the third most common malignancy to cause mortality worldwide, and its prognosis is of great importance. Recent CRC prognostic prediction studies mainly focused on biomarkers, radiometric images, and end-to-end deep learning methods, while only a few works paid attention to exploring the relationship between the quantitative morphological features of patients' tissue slides and their prognosis. However, existing few works in this area suffered from the drawback of choosing the cells randomly from the whole slides, which contain the non-tumor region that lakes information about prognosis. In addition, the existing works, which tried to demonstrate their biological interpretability using patients' transcriptome data, failed to show the biological meaning closely related to cancer. In this study, we proposed and evaluated a prognostic model using morphological features of cells in the tumor region. The features were first extracted by the software CellProfiler from the tumor region selected by Eff-Unet deep learning model. Features from different regions were then averaged for each patient as their representative, and the Lasso-Cox model was used to select the prognosis-related features. The prognostic prediction model was at last constructed using the selected prognosis-related features and was evaluated through KM estimate and cross-validation. In terms of biological meaning, Gene Ontology (GO) enrichment analysis of the expressed genes that correlated with the prognostically significant features was performed to show the biological interpretability of our model.With the help of tumor segmentation, our model achieved better statistical significance and better biological interpretability compared to the results without tumor segmentation. Statistically, the Kaplan Meier (KM) estimate of our model showed that the model using features in the tumor region has a higher C-index, a lower p-value, and a better performance on cross-validation than the model without tumor segmentation. In addition, revealing the pathway of the immune escape and the spread of the tumor, the model with tumor segmentation demonstrated a biological meaning much more related to cancer immunobiology than the model without tumor segmentation. Our prognostic prediction model using quantitive morphological features from tumor regions was almost as good as the TNM tumor staging system as they had a close C-index, and our model can be combined with the TNM tumor stage system to make a better prognostic prediction. And to the best of our knowledge, the biological mechanisms in our study were the most relevant to the immune mechanism of cancer compared to the previous studies.
Molecular biostatistical workflows in oncology often rely on predictive models that use multimodal data. Advances in deep learning and artificial intelligence technologies have enabled the multimodal fusion of large volumes of multimodal data. Here, we presented a decision level multimodal data fusion framework for integrating multiomics and pathological tissue slide images for prognosis prediction. Our approach established the spatial map of instances by connecting the neighboring nuclei in space and calculated the characteristic tensor via graph convolution layers for the input pathological tissue slide images. Global Average Pooling was applied to align and normalize the feature tensors from pathological images and the multiomics data, enabling seamless integration. We tested our proposed approach using Breast Invasive Carcinoma data and Non-Small Cell Lung Cancer data from the Cancer Genome Atlas, which contains paired whole-slide images, transcriptome data, genotype, epienetic, and survival information. In a 10-fold cross-validation, the comparison results demonstrated that the multimodal fusion paradigm improves outcome predictions from single modal data alone with the average C-index increasing from 0.61 to 0.52 to 0.75 and 0.67 for breast cancer and non-small cell lung cancer cohort, respectively. The proposed decision level multimodal data fusion framework is expected to provide insights and technical methodologies for the follow-up studies.
The quality control of variants from whole-genome sequencing data is vital in clinical diagnosis and human genetics research. However, current filtering methods (Frequency, Hard-Filter, VQSR, GARFIELD, and VEF) were developed to be utilized on particular variant callers and have certain limitations. Especially, the number of eliminated true variants far exceeds the number of removed false variants using these methods. Here, we present an adaptive method for quality control on genetic variants from different analysis pipelines, and validate it on the variants generated from four popular variant callers (GATK HaplotypeCaller, Mutect2, Varscan2, and DeepVariant). FVC consistently exhibited the best performance. It removed far more false variants than the current state-of-the-art filtering methods and recalled ~51-99% true variants filtered out by the other methods. Once trained, FVC can be conveniently integrated into a user-specific variant calling pipeline.
Background/Objectives: Constructing a comprehensive spatiotemporal map of tumor heterogeneity is essential for understanding tumor evolution, with copy number variation (CNV) being a significant feature. Existing studies often rely on tools originally developed for single-cell data, which fail to utilize spatial information, often leading to an incomplete map of clonal architecture. Our study aims to develop a model that fully leverages spatial omics data to elucidate spatio-temporal changes in tumor evolution. Methods: Here, we introduce SCOIGET (Spatial COpy number Inference by Graph on Evolution of Tumor), a novel framework using graph neural networks with graph attention layers to learn spatial neighborhood features of gene expression and infer copy number variations. This approach integrates spatial multi-omics features to create a comprehensive spatial map of tumor heterogeneity. Results: Notably, SCOIGET achieves a substantial reduction in error metrics (e.g., mean squared error, cosine similarity, and distance measures) and produces superior clustering performance, as indicated by higher Silhouette Scores compared to existing methods, validated by both simulated data with spot-level ground truth and patient cohorts. Our model significantly enhances the accuracy of tumor evolution depiction, capturing detailed spatial and temporal changes within the tumor microenvironment. It is versatile and applicable to various downstream tasks, demonstrating strong generalizability across different spatial omics platforms, including 10× Visium and Visium HD and various cancer types, including colorectal cancer and prostate cancer. This robust performance improves research efficiency and provides valuable insights into tumor progression. Conclusions: SCOIGET offers an innovative solution by integrating multiple features and advanced algorithms, providing a detailed and accurate representation of tumor heterogeneity and evolution, aiding in the development of personalized cancer treatment strategies.
Conventional 2D analysis of hematoxylin and eosin (H&E)‐stained images is fundamentally limited by the tissue thickness, as cellular overlap and morphological changes in the compressed perspective obscure distinct cell boundaries. To address this, it develops CellSqueeze3D, a computational framework that reconstructs the 3D spatial distribution and size of individual cells from a single H&E‐stained section. Founded on the principle that 2D cell compression preserves 3D geometry, the method employs a hybrid Particle Swarm Optimization (PSO) approach with biomechanical constraints to infer biologically plausible reconstructions. Validation shows that the nuclear‐to‐cytoplasmic (N/C) ratio distribution derived from the predicted cell radii differs significantly from random assignments (p = 1.39e‐80). By employing projected cell boundaries, the 3D‐informed cellular classifier surpassed traditional methods (AUC increases of 0.136 and 0.069). The resulting morphological metrics also revealed strong associations with key gene expression patterns, providing prognostic insights. Furthermore, cellular and nuclear size indices from CellSqueeze3D significantly predict the mutation status of 21 genes in TCGA cohorts, achieving a median AUROC above 0.65 in fivefold cross‐validation. This study demonstrates that fully utilizing the previously untapped 3D spatial information from a single slice significantly enhances computational pathology and quantitative tissue phenotyping.