Classification of colorectal cancer consensus molecular subtypes using attention-based multi-instance learning network on whole-slide images.

Colorectal cancer (CRC) is the third most common and second most lethal cancer globally. It is highly heterogeneous with different clinical-pathological characteristics, prognostic status, and therapy responses. Thus, the precise diagnosis of CRC subtypes is of great significance for improving the prognosis and survival of CRC patients. Nowadays, the most commonly used molecular-level CRC classification system is the Consensus Molecular Subtypes (CMSs). In this study, we applied a weakly supervised deep learning method, named attention-based multi-instance learning (MIL), on formalin-fixed paraffin-embedded (FFPE) whole-slide images (WSIs) to distinguish CMS1 subtype from CMS2, CMS3, and CMS4 subtypes, as well as distinguish CMS4 from CMS1, CMS2, and CMS3 subtypes. The advantage of MIL is training a bag of the tiled instance with bag-level labels only. Our experiment was performed on 1218 WSIs obtained from The Cancer Genome Atlas (TCGA). We constructed three convolutional neural network-based structures for model training and evaluated the ability of the max-pooling operator and mean-pooling operator on aggregating bag-level scores. The results showed that the 3-layer model achieved the best performance in both comparison groups. When compared CMS1 with CMS234, max-pooling reached the ACC of 83.86 % and the mean-pooling operator reached the AUC of 0.731. While comparing CMS4 with CMS123, mean-pooling reached the ACC of 74.26 % and max-pooling reached the AUC of 0.609. Our results implied that WSIs could be utilized to classify CMSs, and manual pixel-level annotation is not a necessity for computational pathology imaging analysis.

Identification of metabolism-associated molecular subtype in ovarian cancer.

Ovarian cancer (OC) is the most lethal gynaecological cancer with genomic complexity and extensive heterogeneity. This study aimed to characterize the molecular features of OC based on the gene expression profile of 2752 previously characterized metabolism-relevant genes and provide new strategies to improve the clinical status of patients with OC. Finally, three molecular subtypes (C1, C2 and C3) were identified. The C2 subtype displayed the worst prognosis, upregulated immune-cell infiltration status and expression level of immune checkpoint genes, lower burden of copy number gains and losses and suboptimal response to targeted drug bevacizumab. The C1 subtype showed downregulated immune-cell infiltration status and expression level of immune checkpoint genes, the lowest incidence of BRCA mutation and optimal response to targeted drug bevacizumab. The C3 subtype had an intermediate immune status, the highest incidence of BRCA mutation and a secondary optimal response to bevacizumab. Gene signatures of C1 and C2 subtypes with an opposite expression level were mainly enriched in proteolysis and immune-related biological process. The C3 subtype was mainly enriched in the T cell-related biological process. The prognostic and immune status of subtypes were validated in the Gene Expression Omnibus (GEO) dataset, which was predicted with a 45-gene classifier. These findings might improve the understanding of the diversity and therapeutic strategies for OC.

Single-Cell Transcriptome Profiling Simulation Reveals the Impact of Sequencing Parameters and Algorithms on Clustering.

Despite the scRNA-seq analytic algorithms developed, their performance for cell clustering cannot be quantified due to the unknown "true" clusters. Referencing the transcriptomic heterogeneity of cell clusters, a "true" mRNA number matrix of cell individuals was defined as ground truth. Based on the matrix and the actual data generation procedure, a simulation program (SSCRNA) for raw data was developed. Subsequently, the consistency between simulated data and real data was evaluated. Furthermore, the impact of sequencing depth and algorithms for analyses on cluster accuracy was quantified. As a result, the simulation result was highly consistent with that of the actual data. Among the clustering algorithms, the Gaussian normalization method was the more recommended. As for the clustering algorithms, the K-means clustering method was more stable than K-means plus Louvain clustering. In conclusion, the scRNA simulation algorithm developed restores the actual data generation process, discovers the impact of parameters on classification, compares the normalization/clustering algorithms, and provides novel insight into scRNA analyses.

Twenty Metabolic Genes Based Signature Predicts Survival of Glioma Patients.

Background: Glioma, caused by carcinogenesis of brain and spinal glial cells, is the most common primary malignant brain tumor. To find the important indicator for glioma prognosis is still a challenge and the metabolic alteration of glioma has been frequently reported recently. Methods: In our current work, a risk score model based on the expression of twenty metabolic genes was developed using the metabolic gene expressions in The Cancer Genome Atlas (TCGA) dataset, the methods of which included the cox multivariate regression and the random forest variable hunting, a kind of machine learning algorithm, and the risk score generated from this model is used to make predictions in the survival of glioma patients in the training dataset. Subsequently, the result was further verified in other three verification sets (GSE4271, GSE4412 and GSE16011). Risk score related pathways collected in the Kyoto Encyclopedia of Genes and Genomes (KEGG) database were identified using Gene Set Enrichment Analysis (GSEA). Results: The risk score generated from our model makes good predictions in the survival of glioma patients in the training dataset and other three verification sets. By assessing the relationships between clinical indicators and the risk score, we found that the risk score was an independent and significant indicator for the prognosis of glioma patients. Simultaneously, we conducted a survival analysis of the patients who received chemotherapy and who did not, finding that the risk score was equally valid in both cases. And signaling pathways related to the genesis and development of multiple cancers were also identified. Conclusions: In summary, our risk score model is predictive for 967 glioma patients' survival from four independent datasets, and the risk score is a meaningful and independent parameter of the clinicopathological information.

A Review of Methods to Quantify the Genomic Similarity of Topological Associating Domains.

Topologically associating domains (TADs) are the most fundamental elements and significant structures of the eukaryotic genome. Currently, algorithms have been developed to find the TADs. But few algorithms are reported to compare the similarity of TADs between genomes. In this study, mice Hi-C sequencing data of four contrasts were enrolled. Seventeen algorithms, including BPscore, Jaccard index (JI) distance, VI distance, image hash, image subtraction, image variance, and so on, were used to quantify the genomic similarity of TADs. Image subtraction, Euclidean distance, and Manhattan distance were significantly better for TAD difference detection than the others. Deferent Hash (dHash) with the best zoom size ranked the second, followed by improved Hamming distance algorithm and JI distance. Advantages and disadvantages of various algorithms for quantifying the similarity of TADs were compared. Our work could provide the fundament for TADs comparison.