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.

Abstract Hematoxylin and eosin (H&E) staining is the most common type of histopathological images used for quantitative data analysis. For histopathological image analysis, nucleus segmentation represents one of the initial steps in quantitative data analysis pipelines. Recently, deep learning methods for nucleus segmentation on histopathological images become the mainstream in digital pathology. However, adopting these pre-trained models in digital pathology or clinical research remains limited. Two main factors limit the usage of deep learning methods in routine research: 1) many models required some technical background to execute the programs, and 2) the speed of returning the results to the users. To overcome these limitations, we have developed and implemented NuKit, a deep learning platform which accelerates nucleus segmentation and provides prompt results to the users. NuKit platform consists of two deep learning models coupled with an interactive graphical user interface (GUI) to provide fast and automatic nucleus segmentation “on the fly”. The two deep learning models are: 1) the whole image segmentation model and 2) the click segmentation model. Both deep learning models provide complementary tasks in nucleus segmentation in the NuKit platform. The whole image segmentation model performs whole image nucleus whereas the click segmentation model supplements the nucleus segmentation with user-driven input to edits the segmented nuclei. We used PanNuke as the training data set which contains 160,368 nuclei extracted from 7,901 image tiles. The sources of this data set were curated from The Cancer Genome Atlas (TCGA) and included a very few in-house data. For test sets, we used several popular data sets such as TCGA (not used in the training set). We used DICE coefficient as the metric for comparisons of NuKit with other deep learning methods. For the results, we observed that NuKit achieved comparable results with the state-of-the-art deep learning methods and outperforms those non-deep learning methods. For TNBC and CoNSeP data sets, NuKit achieved DICE 0.793 and 0.854 comparing to HoVer-Net 0.749 and 0.664. For CoNSeP test data set, NuKit achieved DICE 0.860 comparing to Cell Profiler 0.434 and QuPath 0.588. In addition, both pre-trained models were embedded in the NuKit GUI, which provides interactive and prompt response to users in analyzing their histopathological images. The outputs of NuKit are interoperable with other quantitative image analysis tools to facilitate the ecosystem of computational pathology to many formats to combine with other software to construct pipelines. In summary, we have introduced NuKit, an innovation platform which combines two deep learning models coupled with interactive GUI to accelerate nucleus segmentation task in histopathological images. We believe that NuKit provides a new platform to bridge the pre-trained deep learning models into digital pathology and clinical usage. Citation Format: Ching-Nung Lin, Christine H. Chung, Aik Choon Tan. NuKit: a deep learning platform for histopathological Images. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5409.

Nucleus segmentation represents the initial step for histopathological image analysis pipelines, and it remains a challenge in many quantitative analysis methods in terms of accuracy and speed. Recently, deep learning nucleus segmentation methods have demonstrated to outperform previous intensity- or pattern-based methods. However, the heavy computation of deep learning provides impression of lagging response in real time and hampered the adoptability of these models in routine research. We developed and implemented NuKit a deep learning platform, which accelerates nucleus segmentation and provides prompt results to the users. NuKit platform consists of two deep learning models coupled with an interactive graphical user interface (GUI) to provide fast and automatic nucleus segmentation “on the fly”. Both deep learning models provide complementary tasks in nucleus segmentation. The whole image segmentation model performs whole image nucleus whereas the click segmentation model supplements the nucleus segmentation with user-driven input to edits the segmented nuclei. We trained the NuKit whole image segmentation model on a large public training data set and tested its performance in seven independent public image data sets. The whole image segmentation model achieves average [Formula: see text] and [Formula: see text]. The outputs could be exported into different file formats, as well as provides seamless integration with other image analysis tools such as QuPath. NuKit can be executed on Windows, Mac, and Linux using personal computers.

Immune checkpoint inhibitors (ICI) have provided durable responses in a subset of patients in multiple solid tumor types. However, developing robust and reproducible predictive gene signatures for ICIs remains a challenge partly due to the lack of systematic comparisons across a large pan-cancer transcriptomics compendia. We recently developed Immuno-Oncology Signatures Explorer (IOSig), a user-friendly web tool, to allow users to query and explore predictive biomarkers from a large collection of pan-cancer patient samples treated with ICIs (> 2500 patient samples from 40 studies). Here, we investigated six solid tumors: kidney, melanoma, bladder, head and neck squamous cell, gastrointestinal, and non-small cell lung cancers tissue-based RNA-seq datasets in IOSig. The 2175 samples, from 45 cohorts, were Z-normalized with the genes from each signature being averaged to assign a sample a single score per signature. Area Under the Receiver Operating Characteristic (AUROC) curve was used to summarize the gene signatures’ ability to predict response to ICIs. We focus on the top 10 predictive gene signatures for each cancer type. There were 22 unique gene signatures from the top 10 of the six cancer types. Overall, 17 signatures were predictive in GI, NSCLC, melanoma and HNSCC. Many of these signatures were enriched in immune activation and inflammatory signatures. Conversely, the MYC, DNA damage repair, and proliferation signatures are only predictive in bladder cancer. No predictive signatures were associated with kidney cancer. Interestingly, the immune activation signatures were highly predictive in the GI cancer cohorts in our study, as the cohorts were enriched with patients having high tumor mutational burdens and microsatellite instability. In summary, we used IOSig to analyze the predictive ability of previously published gene signatures across 6 cancer types. Some of these signatures warrant further investigation in a cancer-specific manner. A table showing the AUROC value of the 22 selected gene signatures in each cancer type. Signature Melanoma Bladder Kidney NSCLC HNSC GI Cancer Average Chemokines 0.657 0.563 0.526 0.721 0.657 0.870 0.665 TIP Hot 0.657 0.571 0.544 0.737 0.592 0.866 0.661 ifng18 0.674 0.557 0.584 0.680 0.589 0.871 0.659 Ipi_Neo 0.660 0.567 0.563 0.675 0.602 0.859 0.654 Rooney 0.669 0.569 0.568 0.661 0.646 0.797 0.652 Chaurio 0.670 0.572 0.570 0.679 0.570 0.834 0.649 ifng6 0.655 0.593 0.553 0.627 0.635 0.823 0.648 ifng-effector 0.672 0.566 0.567 0.686 0.564 0.830 0.648 Roh 0.661 0.551 0.565 0.697 0.548 0.856 0.646 NRS 0.649 0.512 0.588 0.687 0.605 0.818 0.643 effector_t 0.679 0.567 0.579 0.670 0.524 0.835 0.642 impres 0.619 0.516 0.577 0.706 0.555 0.847 0.637 mMDSC 0.643 0.524 0.558 0.706 0.557 0.797 0.631 Ock 0.640 0.563 0.586 0.579 0.578 0.827 0.629 Davoli 0.665 0.546 0.569 0.655 0.514 0.822 0.629 MHC_1 0.644 0.593 0.550 0.574 0.553 0.806 0.620 MHC_2 0.661 0.533 0.554 0.659 0.437 0.802 0.608 gMDSC 0.582 0.477 0.526 0.706 0.458 0.695 0.574 DNA_damage_repair 0.432 0.598 0.555 0.383 0.499 0.361 0.471 Proliferation 0.434 0.655 0.530 0.370 0.420 0.372 0.463 Mitoscore 0.422 0.588 0.374 0.418 0.580 0.312 0.449 MYC 0.367 0.607 0.503 0.381 0.482 0.303 0.441 Citation Format: Samuel Coleman, Caroline Wheeler, Rebecca Hoyd, Louis Denko, Ching-Nung Lin, Muhammad Z. Fadlullah, Siwen Hu-Lieskovan, Christine Chung, Ahmad A. Tarhini, Daniel Spakowicz, Aik Choon Tan. Systematic analysis of the predictive gene expression signatures of immunotherapies across multiple cancer types using IOSig [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 6574.