Cancer, in any of its forms, remains a significant public health concern worldwide. Advances in early detection and treatment could lead to a decline in the overall death rate from cancer in recent decades. Therefore, tumor prediction and classification play an important role in fighting cancer. This study built computational models for a joint analysis of RNA seq, copy number variation (CNV), and DNA methylation to classify normal and tumor samples across liver cancer, breast cancer, and colon adenocarcinoma from The Cancer Genome Atlas (TCGA) dataset. Total of 18 machine learning methods were evaluated based on the AUC, precision, recall, and F-measure. Besides, five techniques were compared to ameliorate problems of class imbalance in the cancer datasets. Synthetic Minority Oversampling Technique (SMOTE) demonstrated the best performance. The results indicate that the model applying Stochastic Gradient Descent (SGD) for learning binary class SVM with hinge loss has the highest classification results on liver cancer and breast cancer datasets, with accuracy over 99% and AUC greater than or equal to 0.999. For colon adenocarcinoma dataset, both SGD and Sequential Minimal Optimization (SMO) that implements John Platt’s sequential minimal optimization algorithm for training a support vector machine shows an outstanding classification performance with accuracy of 100%, AUC, precision, recall, and F-measure all at 1.000.

Gliomas are differentiated into two major disease subtypes, astrocytoma or oligodendroglioma, which are then characterized as either IDH (isocitrate dehydrogenase)-wild type or IDH-mutant due to the dramatic differences in prognosis and overall survival. Here, we investigated the genetic background of IDH1-mutant gliomas using the Catalogue of Somatic Mutations in Cancer (COSMIC) database. In astrocytoma patients, we found that IDH1 is often co-mutated with TP53, ATRX, AMBRA1, PREX1, and NOTCH1, but not CHEK2, EGFR, PTEN, or the zinc finger transcription factor ZNF429. The majority of the mutations observed in these genes were further confirmed to be either drivers or pathogenic by the Cancer-Related Analysis of Variants Toolkit (CRAVAT). Gene expression analysis showed down-regulation of DRG2 and MSN expression, both of which promote cell proliferation and invasion. There was also significant over-expression of genes such as NDRG3 and KCNB1 in IDH1-mutant astrocytoma patients. We conclude that IDH1-mutant glioma is characterized by significant genetic changes that could contribute to a better prognosis in glioma patients.

Genetic biomarkers have played a pivotal role in the classification, prognostication, and guidance of clinical cancer therapies. Large-scale and multi-dimensional analyses of entire cancer genomes, as exemplified by projects like The Cancer Genome Atlas (TCGA), have yielded an extensive repository of data that holds the potential to unveil the underlying biology of these malignancies. Mutations stand out as the principal catalysts of cellular transformation. Nonetheless, other global genomic processes, such as alterations in gene expression and chromosomal re-arrangements, also play crucial roles in conferring cellular immortality. The incorporation of multi-omics data specific to cancer has demonstrated the capacity to enhance our comprehension of the molecular mechanisms underpinning carcinogenesis. This report elucidates how the integration of comprehensive data on methylation, gene expression, and copy number variations can effectively facilitate the unsupervised clustering of cancer samples. We have identified regressors that can effectively classify tumor and normal samples with an optimal integration of RNA sequencing, DNA methylation, and copy number variation while also achieving significant $p$-values. Further, these regressors were trained using linear and logistic regression with $k$-means clustering. For comparison, we employed autoencoder- and stacking-based omics integration and computed silhouette scores to evaluate the clusters. The proof of concept is illustrated using liver cancer data. Our analysis serves to underscore the feasibility of unsupervised cancer classification by considering genetic markers beyond mutations, thereby emphasizing the clinical relevance of additional global cellular parameters that contribute to the transformative process in cells. This work is clinically relevant because changes in gene expression and genomic re-arrangements have been shown to be signatures of cellular transformation across cancers, as well as in liver cancers.

Motivation: Cancer subtypes play a critical role in disease progression, prognosis, and treatment, making their detection essential for tailoring precision medicine. Studies have shown that multi-omics integration outperforms single-omics approaches in cancer subtyping tasks. However, due to the high-dimensionality of multi-omics data, many existing studies either fail to capture the correlation between true labels and learned features, or lack sufficient capacity to model complex biological representations. These limitations hinder the full potential of leveraging the rich and complementary information embedded in multi-omics datasets. Result: We propose a framework that leverages supervised feature learning and classification based on a graph-based learning approach with attention mechanism for cancer subtyping. More specifically, we train graph convolutional network models on each omics dataset to extract latent representations, which are then concatenated to form a comprehensive multi-omics feature embedding. We further develop sample fusion network based on the omics-specific graphs, incorporating the derived features and feeding them into a graph attention model for subtype classification. This two-stage multi-omics framework is applied to eight cancer types, with performance evaluated in terms of test accuracy, training time, macro-averaged precision, recall, and F-score. Experimental results show that the proposed method outperforms state-of-the-art approaches across various cancer types. Additionally, we provide empirical evidence supporting the hypothesis that retaining a limited number of high-confidence edges and utilizing enriched embeddings from intermediate graph neural network layers can improve predictive performance. Availability and implementation: Data and the code are available at https://github.com/YD-00/MO-GCAN-Updated.git.

Chromosomal rearrangements are generally a consequence of improperly repaired double-strand breaks in DNA. These genomic aberrations can be a driver of cancers. Here, we investigated the use of chromosomal rearrangements for classification of cancer tumors and the effect of inter- and intrachromosomal rearrangements in cancer classification. We used data from the Catalogue of Somatic Mutations in Cancer (COSMIC) for breast, pancreatic, and prostate cancers, for which the COSMIC dataset reports the highest number of chromosomal aberrations. We developed a framework known as GraphChrom for cancer classification. GraphChrom was developed using a graph neural network which models the complex structure of chromosomal aberrations (CA) and provides local connectivity between the aberrations. The proposed framework illustrates three important contributions to the field of cancers. Firstly, it successfully classifies cancer types and subtypes. Secondly, it evolved into a novel data extraction technique which can be used to extract more informative graphs (informative aberrations associated with a sample); and thirdly, it predicts that interCAs (rearrangements between two or more chromosomes) are more effective in cancer prediction than intraCAs (rearrangements within the same chromosome), although intraCAs are three times more likely to occur than intraCAs.

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder primarily impacting memory and cognitive functions. The hippocampus serves as a key biomarker associated with AD. In this study, we present an end-to-end automated approach for AD detection by introducing a reinforcement-learning-based technique to localize the hippocampus within structural MRI images. Subsequently, this localized hippocampus serves as input for a deep convolutional neural network for AD classification. We model the agent–environment interaction using a Deep Q-Network (DQN), encompassing both a convolutional Target Net and Policy Net. Furthermore, we introduce an integrated loss function that combines cross-entropy and contrastive loss to effectively train the classifier model. Our approach leverages a single optimal slice extracted from each subject’s 3D sMRI, thereby reducing computational complexity while maintaining performance comparable to volumetric data analysis methods. To evaluate the effectiveness of our proposed localization and classification framework, we compare its performance to the results achieved by supervised models directly trained on ground truth hippocampal regions as input. The proposed approach demonstrates promising performance in terms of classification accuracy, F1-score, precision, and recall. It achieves an F1-score within an error margin of $3.7\%$ and $1.1\%$ and an accuracy within an error margin of $6.6\%$ and $1.6\%$ when compared to the supervised models trained directly on ground truth masks, all while achieving the highest recall score.