Modeling results for gene expression differences between responders and non-responders

Predictive values (AUROC) of the 16 gene signatures in the ORIEN cohort (IO_NOVA_Mel) and 22 other melanoma cohorts

Model performance of 31 published gene expression signatures for predicting immunotherapy treatment response

Association of CIBERSORT cell types with the response to ICIs

Correlations between microbe abundances and MSigDB Hallmark gene sets average Z scores

16 gene signatures where high Z-scores are associated with ICI responsiveness in this cohort (FDR < 0.05)

Comparison of alternate machine learning methods

Identified cancer-associated microbial signatures in all studied cancer types.

Patient demographics stratified by response to ICIs

Gene expression profiling technologies have revolutionized cell biology, enabling researchers to identify gene signatures linked to various biological attributes of melanomas, such as pigmentation status, differentiation state, proliferative versus invasive capacity, and disease progression. Although the discovery of gene signatures has significantly enhanced our understanding of melanocytic phenotypes, reconciling the numerous signatures reported across independent studies and different profiling platforms remains a challenge. Current methods for classifying melanocytic gene signatures depend on exact gene overlap and comparison with unstandardized baseline transcriptomes. In this study, we aimed to categorize published gene signatures into clusters based on their similar patterns of expression across clinical cutaneous melanoma specimens. We analyzed nearly 800 melanoma samples from six gene expression repositories and developed a classification framework for gene signatures that is resilient against biases in gene identification across profiling platforms and inconsistencies in baseline standards. Using 39 frequently cited published gene signatures, our analysis revealed seven principal classes of gene signatures that correlate with previously identified phenotypes: Differentiated, Mitotic/MYC, AXL, Amelanotic, Neuro, Hypometabolic, and Invasive. Each class is consistent with the phenotypes that the constituent gene signatures represent, and our classification method does not rely on overlapping genes between signatures. To facilitate broader application, we created WIMMS (what is my melanocytic signature, available at ), a user-friendly web application. WIMMS allows users to categorize any gene signature, determining its relationship to predominantly cited signatures and its representation within the seven principal classes. Using 39 frequently cited published gene signatures, our analysis revealed seven principal classes of gene signatures that correlate with previously identified phenotypes: Differentiated, Mitotic/MYC, AXL, Amelanotic, Neuro, Hypometabolic, and Invasive. Each class is consistent with the phenotypes that the constituent gene signatures represent, and our classification method does not rely on overlapping genes between signatures.image

Overview of the MEGA workflow. Four main steps were included in carrying out model training and biological gene network inference. MEGA uses ORIEN datasets and two database dependencies as the data sources. Preprocessing steps are employed to generate AI-ready data for graph neural network training. After deep learning model training, the cancer-associated microbial signatures were selected on the basis of the attention scores of each species at the sample level. The final results of the identified cancer-associated microbial communities have been provided in a tabular format and are available for additional visualization.

Metabolomics data from colorectal cancer samples.

Network visualization of identified microbial communities in COAD and THCA. The cancer-type nodes were highlighted by an octagon shape, while the microbial species nodes were highlighted in a circle shape. The thickness of the edges in the network reflects the attention weight scores, indicating the strength of the relationship between the species and cancer. In addition, the metabolic compound nodes were highlighted with a yellow triangle shape, while the phylogenetic relationship edges were highlighted in gray. A, COAD-associated microbes highlighted with metabolic compound. B, THCA-associated microbes highlighted with metabolic compound. C, COAD-associated microbes highlighted with phylogenetic relationships. D, THCA-associated microbes highlighted with phylogenetic relationships.

Upset plot of overall identified species in 12 cancer types.

Identified microbial signatures with normalized attention weights.

Shared and unique microbial species identified across different cancer types.

Metabolic compounds relationships microbial signatures.

PURPOSE:Immune checkpoint inhibitors (ICIs) have significantly improved the survival of patients with cancer and provided long-term durable benefit. However, ICI-treated patients develop a range of toxicities known as immune-related adverse events (irAEs), which could compromise clinical benefits from these treatments. As the incidence and spectrum of irAEs differs across cancer types and ICI agents, it is imperative to characterize the incidence and spectrum of irAEs in a pan-cancer cohort to aid clinical management.DESIGN:We queried >400 000 trials registered at ClinicalTrials.gov and retrieved a comprehensive pan-cancer database of 71 087 ICI-treated participants from 19 cancer types and 7 ICI agents. We performed data harmonization and cleaning of these trial results into 293 harmonized adverse event categories using Medical Dictionary for Regulatory Activities.RESULTS:We developed irAExplorer (https://irae.tanlab.org), an interactive database that focuses on adverse events in patients administered with ICIs from big data mining. irAExplorer encompasses 71 087 distinct clinical trial participants from 343 clinical trials across 19 cancer types with well-annotated ICI treatment regimens and harmonized adverse event categories. We demonstrated a few of the irAE analyses through irAExplorer and highlighted some associations between treatment- or cancer-specific irAEs.CONCLUSION:The irAExplorer is a user-friendly resource that offers exploration, validation, and discovery of treatment- or cancer-specific irAEs across pan-cancer cohorts. We envision that irAExplorer can serve as a valuable resource to cross-validate users' internal datasets to increase the robustness of their findings.

16s rRNA dataset from colorectal cancer samples.

Circos plot representation of the distribution of identified species and cancer types. The segment length for each cancer type is proportional to the ratio of the total number of detected species within that cancer type, and individual ribbons are linked to their respective species. The cancer types are abbreviated as COAD (colon adenocarcinoma), LUAD (lung adenocarcinoma), LUSC (lung squamous cell carcinoma), OtherCR (other colorectal cancer types not specified), OtherLung (other lung cancer types not specified), OtherPancreatic (other pancreatic cancer types not specified), PAAD (pancreatic adenocarcinoma), READ (rectum adenocarcinoma), SARC (sarcoma), SCLC (small cell lung cancer), SKCM (skin cutaneous melanoma), and THCA (thyroid carcinoma).