Pan-cancer identification of clinically relevant genomic subtypes using outcome-weighted integrative clustering

Identification of genetic subtypes in follicular lymphoma

Follicular lymphoma (FL) exhibits considerable variability in biological features and clinical trajectories across patients. To dissect the diversity of FL, we utilized a Bernoulli mixture model to identify genetic subtypes in 713 pre-treatment tumor tissue samples. Our analysis revealed the existence of five subtypes with unique genetic profiles that correlated with clinicopathological characteristics. The clusters were enriched in specific mutations as follows: CS (CREBBP and STAT6), TT (TNFAIP3 and TP53), GM (GNA13 and MEF2B), Q (quiescent, for low mutation burden), and AR (mutations of mTOR pathway-related genes). The subtype Q was enriched for patients with stage I disease and associated with a lower proliferative history than the other subtypes. The AR subtype was unique in its enrichment for IgM-expressing FL cases and was associated with advanced-stage and more than 4 nodal sites. The existence of subtypes was validated in an independent cohort of 418 samples from the GALLIUM trial. Notably, patients assigned to the TT subtype consistently experienced inferior progression-free survival when treated with immunochemotherapy. Our findings offer insight into core pathways distinctly linked with each FL cluster and are expected to be informative in the era of targeted therapies.

wMKL: multi-omics data integration enables novel cancer subtype identification via weight-boosted multi-kernel learning

BACKGROUND

Cancer is a heterogeneous disease driven by complex molecular alterations. Cancer subtypes determined from multi-omics data can provide novel insight into personalised precision treatment. It is recognised that incorporating prior weight knowledge into multi-omics data integration can improve disease subtyping.

METHODS

We develop a weighted method, termed weight-boosted Multi-Kernel Learning (wMKL) which incorporates heterogeneous data types as well as flexible weight functions, to boost subtype identification. Given a series of weight functions, we propose an omnibus combination strategy to integrate different weight-related P-values to improve subtyping precision.

RESULTS

wMKL models each data type with multiple kernel choices, thus alleviating the sensitivity and robustness issue due to selecting kernel parameters. Furthermore, wMKL integrates different data types by learning weights of different kernels derived from each data type, recognising the heterogeneous contribution of different data types to the final subtyping performance. The proposed wMKL outperforms existing weighted and non-weighted methods. The utility and advantage of wMKL are illustrated through extensive simulations and applications to two TCGA datasets. Novel subtypes are identified followed by extensive downstream bioinformatics analysis to understand the molecular mechanisms differentiating different subtypes.

CONCLUSIONS

The proposed wMKL method provides a novel strategy for disease subtyping. The wMKL is freely available at https://github.com/biostatcao/wMKL .

Emerging applications of machine learning in genomic medicine and healthcare

The integration of artificial intelligence technologies has propelled the progress of clinical and genomic medicine in recent years. The significant increase in computing power has facilitated the ability of artificial intelligence models to analyze and extract features from extensive medical data and images, thereby contributing to the advancement of intelligent diagnostic tools. Artificial intelligence (AI) models have been utilized in the field of personalized medicine to integrate clinical data and genomic information of patients. This integration allows for the identification of customized treatment recommendations, ultimately leading to enhanced patient outcomes. Notwithstanding the notable advancements, the application of artificial intelligence (AI) in the field of medicine is impeded by various obstacles such as the limited availability of clinical and genomic data, the diversity of datasets, ethical implications, and the inconclusive interpretation of AI models' results. In this review, a comprehensive evaluation of multiple machine learning algorithms utilized in the fields of clinical and genomic medicine is conducted. Furthermore, we present an overview of the implementation of artificial intelligence (AI) in the fields of clinical medicine, drug discovery, and genomic medicine. Finally, a number of constraints pertaining to the implementation of artificial intelligence within the healthcare industry are examined.

The implication of integrative multiple RNA modification-based subtypes in gastric cancer immunotherapy and prognosis

Previous studies have focused on the impact of individual RNA modifications on tumor development. This study comprehensively investigated the effects of multiple RNA modifications, including m6A, alternative polyadenylation, pseudouridine, adenosine-to-inosine editing, and uridylation, on gastric cancer (GC). By analyzing 1,946 GC samples from eleven independent cohorts, we identified distinct clusters of RNA modification genes with varying survival rates and immunological characteristics. We assessed the chromatin activity of these RNA modification clusters through regulon enrichment analysis. A prognostic model was developed using Stepwise Regression and Random Survival Forest algorithms and validated in ten independent datasets. Notably, the low-risk group showed a more favorable prognosis and positive response to immune checkpoint blockade therapy. Single-cell RNA sequencing confirmed the abundant expression of signature genes in B cells and plasma cells. Overall, our findings shed light on the potential significance of multiple RNA modifications in GC prognosis, stemness development, and chemotherapy resistance.

Development of Tumor Markers for Breast Cancer Immunotherapy

Although breast cancer treatment has been developed remarkably in recent years, it remains the primary cause of death among women. Immune checkpoint blockade therapy has significantly altered the way breast cancer is treated, although not all patients benefit from the changes. At present, the most effective mechanism of immune checkpoint blockade application in malignant tumors is not clear and efficacy may be influenced by many factors, including host, tumor, and tumor microenvironment dynamics. Therefore, there is a pressing need for tumor immunomarkers that can be used to screen patients and help determine which of them would benefit from breast cancer immunotherapy. At present, no single tumor marker can predict treatment efficacy with sufficient accuracy. Multiple markers may be combined to more accurately pinpoint patients who will respond favorably to immune checkpoint blockade medication. In this review, we have examined the breast cancer treatments, developments in research on the role of tumor markers in maximizing the clinical efficacy of immune checkpoint inhibitors, prospects for the identification of novel therapeutic targets, and the creation of individualized treatment plans. We also discuss how tumor markers can provide guidance for clinical practice.

A comparative study of clustering methods on gene expression data for lung cancer prognosis

Lung cancer subtyping based on gene expression data is important for identifying patient subgroups with differing survival prognosis to facilitate customized treatment strategies for each subtype of patients. Unsupervised clustering methods are the traditional approach for clustering patients into subtypes. However, since those methods cluster patients based only on gene expression data, the resulting clusters may not always be relevant to the survival outcome of interest. In recent years, semi-supervised and supervised methods have been proposed, which leverage the survival outcome data to identify clusters more relevant to survival prognosis. This paper aims to compare the performance of different clustering methods for identifying clinically prognostic lung cancer subtypes based on two lung adenocarcinoma datasets. For each method, we clustered patients into two clusters and assessed the difference in patient survival time between clusters. Unsupervised methods were found to have large logrank p-values and no significant results in most cases. Semi-supervised and supervised methods had improved performance over unsupervised methods and very significant p-values. These results indicate that unsupervised methods are not capable of identifying clusters with significant differences in survival prognosis in most cases, while supervised and semi-supervised methods can better cluster patients into clinically useful subtypes.

Mutation-Attention (MuAt): deep representation learning of somatic mutations for tumour typing and subtyping

BACKGROUND

Cancer genome sequencing enables accurate classification of tumours and tumour subtypes. However, prediction performance is still limited using exome-only sequencing and for tumour types with low somatic mutation burden such as many paediatric tumours. Moreover, the ability to leverage deep representation learning in discovery of tumour entities remains unknown.

METHODS

We introduce here Mutation-Attention (MuAt), a deep neural network to learn representations of simple and complex somatic alterations for prediction of tumour types and subtypes. In contrast to many previous methods, MuAt utilizes the attention mechanism on individual mutations instead of aggregated mutation counts.

RESULTS

We trained MuAt models on 2587 whole cancer genomes (24 tumour types) from the Pan-Cancer Analysis of Whole Genomes (PCAWG) and 7352 cancer exomes (20 types) from the Cancer Genome Atlas (TCGA). MuAt achieved prediction accuracy of 89% for whole genomes and 64% for whole exomes, and a top-5 accuracy of 97% and 90%, respectively. MuAt models were found to be well-calibrated and perform well in three independent whole cancer genome cohorts with 10,361 tumours in total. We show MuAt to be able to learn clinically and biologically relevant tumour entities including acral melanoma, SHH-activated medulloblastoma, SPOP-associated prostate cancer, microsatellite instability, POLE proofreading deficiency, and MUTYH-associated pancreatic endocrine tumours without these tumour subtypes and subgroups being provided as training labels. Finally, scrunity of MuAt attention matrices revealed both ubiquitous and tumour-type specific patterns of simple and complex somatic mutations.

CONCLUSIONS

Integrated representations of somatic alterations learnt by MuAt were able to accurately identify histological tumour types and identify tumour entities, with potential to impact precision cancer medicine.

Cluster analysis of autoencoder-extracted FDG PET/CT features identifies multiple myeloma patients with poor prognosis

F-18 fluorodeoxyglucose positron emission tomography/computed tomography (FDG PET/CT) is a robust imaging modality used for staging multiple myeloma (MM) and assessing treatment responses. Herein, we extracted features from the FDG PET/CT images of MM patients using an artificial intelligence autoencoder algorithm that constructs a compressed representation of input data. We then evaluated the prognostic value of the image-feature clusters thus extracted. Conventional image parameters including metabolic tumor volume (MTV) were measured on volumes-of-interests (VOIs) covering only the bones. Features were extracted with the autoencoder algorithm on bone-covering VOIs. Supervised and unsupervised clustering were performed on image features. Survival analyses for progression-free survival (PFS) were performed for conventional parameters and clusters. In result, supervised and unsupervised clustering of the image features grouped the subjects into three clusters (A, B, and C). In multivariable Cox regression analysis, unsupervised cluster C, supervised cluster C, and high MTV were significant independent predictors of worse PFS. Supervised and unsupervised cluster analyses of image features extracted from FDG PET/CT scans of MM patients by an autoencoder allowed significant and independent prediction of worse PFS. Therefore, artificial intelligence algorithm-based cluster analyses of FDG PET/CT images could be useful for MM risk stratification.

Potential association factors for developing effective peptide-based cancer vaccines

Peptide-based cancer vaccines have been shown to boost immune systems to kill tumor cells in cancer patients. However, designing an effective T cell epitope peptide-based cancer vaccine still remains a challenge and is a major hurdle for the application of cancer vaccines. In this study, we constructed for the first time a library of peptide-based cancer vaccines and their clinical attributes, named CancerVaccine (https://peptidecancervaccine.weebly.com/). To investigate the association factors that influence the effectiveness of cancer vaccines, these peptide-based cancer vaccines were classified into high (HCR) and low (LCR) clinical responses based on their clinical efficacy. Our study highlights that modified peptides derived from artificially modified proteins are suitable as cancer vaccines, especially for melanoma. It may be possible to advance cancer vaccines by screening for HLA class II affinity peptides may be an effective therapeutic strategy. In addition, the treatment regimen has the potential to influence the clinical response of a cancer vaccine, and Montanide ISA-51 might be an effective adjuvant. Finally, we constructed a high sensitivity and specificity machine learning model to assist in designing peptide-based cancer vaccines capable of providing high clinical responses. Together, our findings illustrate that a high clinical response following peptide-based cancer vaccination is correlated with the right type of peptide, the appropriate adjuvant, and a matched HLA allele, as well as an appropriate treatment regimen. This study would allow for enhanced development of cancer vaccines.

New Genetic Technologies in Diagnosis and Treatment of Cancer of Unknown Primary

Simple Summary

The NGS and other molecular techniques creates huge hopes for effective CUP patients treatment and to select them for molecularly targeted therapies (agnostic therapies) and immunotherapy. Development of diagnostic technologies and biologically targeted therapies could make CUP’ patients access to modern therapies and change their outcome.

Abstract

Cancer of unknown primary (CUP) represents a rare oncological and heterogeneous disease in which one or more metastases are present, but the location of the primary site is unknown. Pathological diagnosis, using immunohistochemistry, of such metastatic materials is challenging and frequently does not allow for determining the tissue of origin (ToO). The selection of systemic therapy in patients with CUP is usually based on empiric grounds, and the prognosis is generally unfavourable. New molecular techniques could identify the tissue of origin and be used to select systemic agnostic therapies in various malignancies with specific molecular abnormalities. Targetable driver mutations or gene rearrangements in cancer cells may be identified using various molecular assays, of which particularly valuable are next-generation sequencing techniques. These assays may identify tumour sources and allow personalized treatments. However, current guidelines for CUP management do not recommend routine testing of gene expression and epigenetic factors. This is mainly due to the insufficient evidence supporting the improvement of CUP’s prognosis by virtue of this approach. This review summarizes the advantages and disadvantages of new genetic techniques in CUP diagnostics and proposes updating the recommendations for CUP management.

Exploring Histological Similarities Across Cancers From a Deep Learning Perspective

Histopathology image analysis is widely accepted as a gold standard for cancer diagnosis. The Cancer Genome Atlas (TCGA) contains large repositories of histopathology whole slide images spanning several organs and subtypes. However, not much work has gone into analyzing all the organs and subtypes and their similarities. Our work attempts to bridge this gap by training deep learning models to classify cancer vs. normal patches for 11 subtypes spanning seven organs (9,792 tissue slides) to achieve high classification performance. We used these models to investigate their performances in the test set of other organs (cross-organ inference). We found that every model had a good cross-organ inference accuracy when tested on breast, colorectal, and liver cancers. Further, high accuracy is observed between models trained on the cancer subtypes originating from the same organ (kidney and lung). We also validated these performances by showing the separability of cancer and normal samples in a high-dimensional feature space. We further hypothesized that the high cross-organ inferences are due to shared tumor morphologies among organs. We validated the hypothesis by showing the overlap in the Gradient-weighted Class Activation Mapping (GradCAM) visualizations and similarities in the distributions of nuclei features present within the high-attention regions.

CpG Island Methylator Phenotype-A Hope for the Future or a Road to Nowhere?

The CpG island methylator phenotype (CIMP) can be regarded as the most notable emanation of epigenetic instability in cancer. Since its discovery in the late 1990s, CIMP has been extensively studied, mainly in colorectal cancers (CRC) and gliomas. Consequently, knowledge on molecular and pathological characteristics of CIMP in CRC and other tumour types has rapidly expanded. Concordant and widespread hypermethylation of multiple CpG islands observed in CIMP in multiple cancers raised hopes for future epigenetically based diagnostics and treatments of solid tumours. However, studies on CIMP in solid tumours were hampered by a lack of generalisability and reproducibility of epigenetic markers. Moreover, CIMP was not a satisfactory marker in predicting clinical outcomes. The idea of targeting epigenetic abnormalities such as CIMP for cancer therapy has not been implemented for solid tumours, either. Twenty-one years after its discovery, we aim to cover both the fundamental and new aspects of CIMP and its future application as a diagnostic marker and target in anticancer therapies.

Identifying individualized risk subpathways reveals pan-cancer molecular classification based on multi-omics data

Cancer is a highly heterogeneous disease with different functional disorders among individuals. The initiation and progression of cancer is usually related to dysregulation of local regions within pathways. Identification of individualized risk pathways is crucial for revealing the mechanisms of tumorigenesis and heterogeneity. However, approach that focused on mining patient-specific risk subpathway regions is still lacking. Here, we developed an individualized cancer risk subpathway identification method that was referred as InCRiS by integrating multi-omics data. Then, the method was applied to nearly 3000 samples across 9 TCGA cancer types and its robustness and reliability were evaluated. Dissecting dysregulated subpathways in these tumor samples revealed several key regions which may play oncogenic roles in cancer. The construction of risk subpathway dysregulation profile of pan-cancers revealed 11 pan-cancer molecular classification (InCRiS subtypes) with significantly different clinical outcomes. Moreover, subpathway regions that tend to be disordered in individuals of specific subtypes were examined for understanding the pathogenesis of tumor and some key genes such as CTNNB1, EP300 and PRKDC were nominated in different subtypes. In summary, the proposed method and resulting data presented useful resources to study the mechanism of tumor heterogeneity and to discovery novel therapeutic targets for precise treatment of cancer.

Identification of an m6A Regulators-Mediated Prognosis Signature For Survival Prediction and Its Relevance to Immune Infiltration in Melanoma

Despite robust evidence for the role of m6A in cancer development and progression, its association with immune infiltration and survival outcomes in melanoma remains obscure. Here, we aimed to develop an m6A-related risk signature to improve prognostic and immunotherapy responder prediction performance in the context of melanoma. We comprehensively analyzed the m6A cluster and immune infiltration phenotypes of public datasets. The TCGA (n = 457) and eleven independent melanoma cohorts (n = 758) were used as the training and validation datasets, respectively. We identified two m6A clusters (m6A-clusterA and m6A-clusterB) based on the expression pattern of m6A regulators via unsupervised consensus clustering. IGF2BP1 (7.49%), KIAA1429 (7.06%), and YTHDC1 (4.28%) were the three most frequently mutated genes. There was a correlation between driver genes mutation statuses and the expression of m6A regulators. A significant difference in tumor-associated immune infiltration between two m6A clusters was detected. Compared with m6A-clusterA, the m6A-clusterB was characterized by a lower immune score and immune cell infiltration but higher mRNA expression-based stemness index (mRNAsi). An m6A-related risk signature consisting of 12 genes was determined via Cox regression analysis and divided the patients into low- and high-risk groups (IL6ST, MBNL1, NXT2, EIF2A, CSGALNACT1, C11orf58, CD14, SPI1, NCCRP1, BOK, CD74, PAEP). A nomogram was developed for the prediction of the survival rate. Compared with the high-risk group, the low-risk group was characterized by high expression of immune checkpoints and immunophenoscore (IPS), activation of immune-related pathways, and more enriched in immune cell infiltrations. The low-risk group had a favorable prognosis and contained the potential beneficiaries of the immune checkpoint blockade therapy and verified by the IMvigor210 cohort (n = 298). The m6A-related signature we have determined in melanoma highlights the relationships between m6A regulators and immune cell infiltration. The established risk signature was identified as a promising clinical biomarker of melanoma.

Subgroup Identification with Gene Expression Profiles of Adipose Tissue in Patients with Coronary Artery Disease

Among many diseases, coronary artery disease (CAD) is the primary cause of mortality and morbidity worldwide. With the aim of revealing the underlying genetic characteristics of the CAD subtypes, we recruited patients with CAD and categorized them into subgroups according to the transcriptome expression profiles of the adipose tissue.With the removal of the batch effect, consensus clustering was employed to determine the subgroup numbers. Subgroup-specific genes were determined to conduct analysis of Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG). Weighted gene co-expression network analysis (WGCNA) revealed the subgroup-specific WGCNA modules. Moreover, gene set enrichment analysis (GSEA) was conducted. Overrepresentation enrichment analysis (OEA) of subgroup-specific signatures was also conducted to reveal the significant gene module associated with the corresponding clinical characteristics.After the removal of the batch effect, 77 CAD objects were divided into three subgroups. It was observed that the patients in subgroup III tended to be fat. After analyzing the dominant pathways of each subgroup, we discovered that the protein digestion and absorption pathway was specifically upregulated in subgroup I, which might result from the lowest proportion of the epicardial adipose tissue (EAT) sample. Moreover, subgroup II patients had genetic characteristics of high expression of complement and coagulation cascades and TNF signaling pathway. Furthermore, Th17 cell differentiation was significantly upregulated in subgroup III, indicating that Th17 cell differentiation is related to the clinical characteristics of body mass index (BMI).In conclusion, the genetic classification of CAD subjects indicated that subjects from different subgroups may exhibit specific gene expression patterns, suggesting that more personalized treatment should be applied to patients in each subgroup.

LAG-3 expression on peripheral blood cells identifies patients with poorer outcomes after immune checkpoint blockade

Immune checkpoint blocking antibodies are a cornerstone in cancer treatment; however, they benefit only a subset of patients and biomarkers to guide immune checkpoint blockade (ICB) treatment choices are lacking. We designed this study to identify blood-based correlates of clinical outcome in ICB-treated patients. We performed immune profiling of 188 ICB-treated patients with melanoma using multiparametric flow cytometry to characterize immune cells in pretreatment peripheral blood. A supervised statistical learning approach was applied to a discovery cohort to classify phenotypes and determine their association with survival and treatment response. We identified three distinct immune phenotypes (immunotypes), defined in part by the presence of a LAG-3⁺CD8⁺ T cell population. Patients with melanoma with a LAG⁺ immunotype had poorer outcomes after ICB with a median survival of 22.2 months compared to 75.8 months for those with the LAG^- immunotype (P = 0.031). An independent cohort of 94 ICB-treated patients with urothelial carcinoma was used for validation where LAG⁺ immunotype was significantly associated with response (P = 0.007), survival (P < 0.001), and progression-free survival (P = 0.004). Multivariate Cox regression and stratified analyses further showed that the LAG⁺ immunotype was an independent marker of outcome when compared to known clinical prognostic markers and previously described markers for the clinical activity of ICB, PD-L1, and tumor mutation burden. The pretreatment peripheral blood LAG⁺ immunotype detects patients who are less likely to benefit from ICB and suggests a strategy for identifying actionable immune targets for further investigation.

Multi-project and Multi-profile joint Non-negative Matrix Factorization for cancer omic datasets

MOTIVATION

The integration of multi-omic data using machine learning methods has been focused on solving relevant tasks such as predicting sensitivity to a drug or subtyping patients. Recent integration methods, such as joint Non-negative Matrix Factorization (jNMF), have allowed researchers to exploit the information in the data to unravel the biological processes of multi-omic datasets.

RESULTS

We present a novel method called Multi-project and Multi-profile joint Non-negative Matrix Factorization (M&M-jNMF) capable of integrating data from different sources, such as experimental and observational multi-omic data. The method can generate co-clusters between observations, predict profiles and relate latent variables. We applied the method to integrate low-grade glioma omic profiles from The Cancer Genome Atlas (TCGA) and Cell Line Encyclopedia (CCLE) projects. The method allowed us to find gene clusters mainly enriched in cancer-associated terms. We identified groups of patients and cell lines similar to each other by comparing biological processes. We predicted the drug profile for patients, and we identified genetic signatures for resistant and sensitive tumors to a specific drug.

AVAILABILITY AND IMPLEMENTATION

Source code repository is publicly available at https://bitbucket.org/dsalazarb/mmjnmf/ - Zenodo DOI: 10.5281/zenodo.5150920.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Progress and Challenges of Predictive Biomarkers for Immune Checkpoint Blockade

Over the past decade, immune checkpoint blockade (ICB) therapy has revolutionized the outlook for oncology with significant and sustained improvement in the overall patient survival. Unlike traditional cancer therapies, which target the cancer cells directly, ICB acts on the immune system to enhance anti-tumoral immunity. However, the response rate is still far from satisfactory and most patients are refractory to such treatment. Unfortunately, the mechanisms underlying such heterogeneous responses between patients to ICB therapy remain unclear. In addition, escalating costs of cancer care and unnecessary immune-related adverse events also are pertinent considerations with applications of ICB. Given these issues, identifying explicit predictive biomarkers for patient selection is an urgent unmet need to increase the efficacy of ICB therapy. The markers can be classified as tumor related and non-tumor-related biomarkers. Although substantial efforts have been put into investigating various biomarkers, none of them has been found to be sufficient for effectively stratifying patients who may benefit from immunotherapy. The present write up is an attempt to review the various emerging clinically relevant biomarkers affecting the efficacy of immune checkpoint inhibitors, as well as the limitations associated with their clinical application.