Deep learning-based cross-classifications reveal conserved spatial behaviors within tumor histological images.

Histopathological images are a rich but incompletely explored data type for studying cancer. Manual inspection is time consuming, making it challenging to use for image data mining. Here we show that convolutional neural networks (CNNs) can be systematically applied across cancer types, enabling comparisons to reveal shared spatial behaviors. We develop CNN architectures to analyze 27,815 hematoxylin and eosin scanned images from The Cancer Genome Atlas for tumor/normal, cancer subtype, and mutation classification. Our CNNs are able to classify TCGA pathologist-annotated tumor/normal status of whole slide images (WSIs) in 19 cancer types with consistently high AUCs (0.995 ± 0.008), as well as subtypes with lower but significant accuracy (AUC 0.87 ± 0.1). Remarkably, tumor/normal CNNs trained on one tissue are effective in others (AUC 0.88 ± 0.11), with classifier relationships also recapitulating known adenocarcinoma, carcinoma, and developmental biology. Moreover, classifier comparisons reveal intra-slide spatial similarities, with an average tile-level correlation of 0.45 ± 0.16 between classifier pairs. Breast cancers, bladder cancers, and uterine cancers have spatial patterns that are particularly easy to detect, suggesting these cancers can be canonical types for image analysis. Patterns for TP53 mutations can also be detected, with WSI self- and cross-tissue AUCs ranging from 0.65-0.80. Finally, we comparatively evaluate CNNs on 170 breast and colon cancer images with pathologist-annotated nuclei, finding that both cellular and intercellular regions contribute to CNN accuracy. These results demonstrate the power of CNNs not only for histopathological classification, but also for cross-comparisons to reveal conserved spatial behaviors across tumors.

CUP-AI-Dx: A tool for inferring cancer tissue of origin and molecular subtype using RNA gene-expression data and artificial intelligence.

BACKGROUND: Cancer of unknown primary (CUP), representing approximately 3-5% of all malignancies, is defined as metastatic cancer where a primary site of origin cannot be found despite a standard diagnostic workup. Because knowledge of a patient's primary cancer remains fundamental to their treatment, CUP patients are significantly disadvantaged and most have a poor survival outcome. Developing robust and accessible diagnostic methods for resolving cancer tissue of origin, therefore, has significant value for CUP patients. METHODS: We developed an RNA-based classifier called CUP-AI-Dx that utilizes a 1D Inception convolutional neural network (1D-Inception) model to infer a tumor's primary tissue of origin. CUP-AI-Dx was trained using the transcriptional profiles of 18,217 primary tumours representing 32 cancer types from The Cancer Genome Atlas project (TCGA) and International Cancer Genome Consortium (ICGC). Gene expression data was ordered by gene chromosomal coordinates as input to the 1D-CNN model, and the model utilizes multiple convolutional kernels with different configurations simultaneously to improve generality. The model was optimized through extensive hyperparameter tuning, including different max-pooling layers and dropout settings. For 11 tumour types, we also developed a random forest model that can classify the tumour's molecular subtype according to prior TCGA studies. The optimised CUP-AI-Dx tissue of origin classifier was tested on 394 metastatic samples from 11 tumour types from TCGA and 92 formalin-fixed paraffin-embedded (FFPE) samples representing 18 cancer types from two clinical laboratories. The CUP-AI-Dx molecular subtype was also independently tested on independent ovarian and breast cancer microarray datasets FINDINGS: CUP-AI-Dx identifies the primary site with an overall top-1-accuracy of 98.54% in cross-validation and 96.70% on a test dataset. When applied to two independent clinical-grade RNA-seq datasets generated from two different institutes from the US and Australia, our model predicted the primary site with a top-1-accuracy of 86.96% and 72.46% respectively. INTERPRETATION: The CUP-AI-Dx predicts tumour primary site and molecular subtype with high accuracy and therefore can be used to assist the diagnostic work-up of cancers of unknown primary or uncertain origin using a common and accessible genomics platform. FUNDING: NIH R35 GM133562, NCI P30 CA034196, Victorian Cancer Agency Australia.

Extrachromosomal DNA is associated with oncogene amplification and poor outcome across multiple cancers.

Extrachromosomal DNA (ecDNA) amplification promotes intratumoral genetic heterogeneity and accelerated tumor evolution1-3; however, its frequency and clinical impact are unclear. Using computational analysis of whole-genome sequencing data from 3,212 cancer patients, we show that ecDNA amplification frequently occurs in most cancer types but not in blood or normal tissue. Oncogenes were highly enriched on amplified ecDNA, and the most common recurrent oncogene amplifications arose on ecDNA. EcDNA amplifications resulted in higher levels of oncogene transcription compared to copy number-matched linear DNA, coupled with enhanced chromatin accessibility, and more frequently resulted in transcript fusions. Patients whose cancers carried ecDNA had significantly shorter survival, even when controlled for tissue type, than patients whose cancers were not driven by ecDNA-based oncogene amplification. The results presented here demonstrate that ecDNA-based oncogene amplification is common in cancer, is different from chromosomal amplification and drives poor outcome for patients across many cancer types.

ChIA-PIPE: A fully automated pipeline for comprehensive ChIA-PET data analysis and visualization.

ChIA-PET (chromatin interaction analysis with paired-end tags) enables genome-wide discovery of chromatin interactions involving specific protein factors, with base pair resolution. Interpretation of ChIA-PET data requires a robust analytic pipeline. Here, we introduce ChIA-PIPE, a fully automated pipeline for ChIA-PET data processing, quality assessment, visualization, and analysis. ChIA-PIPE performs linker filtering, read mapping, peak calling, and loop calling and automates quality control assessment for each dataset. To enable visualization, ChIA-PIPE generates input files for two-dimensional contact map viewing with Juicebox and HiGlass and provides a new dockerized visualization tool for high-resolution, browser-based exploration of peaks and loops. To enable structural interpretation, ChIA-PIPE calls chromatin contact domains, resolves allele-specific peaks and loops, and annotates enhancer-promoter loops. ChIA-PIPE also supports the analysis of other related chromatin-mapping data types.

Fostering bioinformatics education through skill development of professors: Big Genomic Data Skills Training for Professors.

Bioinformatics has become an indispensable part of life science over the past 2 decades. However, bioinformatics education is not well integrated at the undergraduate level, especially in liberal arts colleges and regional universities in the United States. One significant obstacle pointed out by the Network for Integrating Bioinformatics into Life Sciences Education is the lack of faculty in the bioinformatics area. Most current life science professors did not acquire bioinformatics analysis skills during their own training. Consequently, a great number of undergraduate and graduate students do not get the chance to learn bioinformatics or computational biology skills within a structured curriculum during their education. To address this gap, we developed a module-based, week-long short course to train small college and regional university professors with essential bioinformatics skills. The bioinformatics modules were built to be adapted by the professor-trainees afterward and used in their own classes. All the course materials can be accessed at https://github.com/TheJacksonLaboratory/JAXBD2K-ShortCourse.

Mutations in DNA repair genes are associated with increased neoantigen burden and a distinct immunophenotype in lung squamous cell carcinoma.

Deficiencies in DNA repair pathways, including mismatch repair (MMR), have been linked to higher tumor mutation burden and improved response to immune checkpoint inhibitors. However, the significance of MMR mutations in lung cancer has not been well characterized, and the relevance of other processes, including homologous recombination (HR) and polymerase epsilon (POLE) activity, remains unclear. Here, we analyzed a dataset of lung squamous cell carcinoma samples from The Cancer Genome Atlas. Variants in DNA repair genes were associated with increased tumor mutation and neoantigen burden, which in turn were linked with greater tumor infiltration by activated T cells. The subset of tumors with DNA repair gene variants but without T cell infiltration exhibited upregulation of TGF-ß and Wnt pathway genes, and a combined score incorporating these genes and DNA repair status accurately predicted immune cell infiltration. Finally, high neoantigen burden was positively associated with genes related to cytolytic activity and immune checkpoints. These findings provide evidence that DNA repair pathway defects and immunomodulatory genes together lead to specific immunophenotypes in lung squamous cell carcinoma and could potentially serve as biomarkers for immunotherapy.

Distribution-based measures of tumor heterogeneity are sensitive to mutation calling and lack strong clinical predictive power.

Mutant allele frequency distributions in cancer samples have been used to estimate intratumoral heterogeneity and its implications for patient survival. However, mutation calls are sensitive to the calling algorithm. It remains unknown whether the relationship of heterogeneity and clinical outcome is robust to these variations. To resolve this question, we studied the robustness of allele frequency distributions to the mutation callers MuTect, SomaticSniper, and VarScan in 4722 cancer samples from The Cancer Genome Atlas. We observed discrepancies among the results, particularly a pronounced difference between allele frequency distributions called by VarScan and SomaticSniper. Survival analysis showed little robust predictive power for heterogeneity as measured by Mutant-Allele Tumor Heterogeneity (MATH) score, with the exception of uterine corpus endometrial carcinoma. However, we found that variations in mutant allele frequencies were mediated by variations in copy number. Our results indicate that the clinical predictions associated with MATH score are primarily caused by copy number aberrations that alter mutant allele frequencies. Finally, we present a mathematical model of linear tumor evolution demonstrating why MATH score is insufficient for distinguishing different scenarios of tumor growth. Our findings elucidate the importance of allele frequency distributions as a measure for tumor heterogeneity and their prognostic role.

Mutations in DNA repair genes are associated with increased neo-antigen load and activated T cell infiltration in lung adenocarcinoma.

Mutations in DNA repair genes lead to increased genomic instability and mutation frequency. These mutations represent potential biomarkers for cancer immunotherapy efficacy, as high tumor mutational burden has been associated with increased neo-antigens and tumor infiltrating lymphocytes. While mismatch repair mutations have successfully predicted response to anti-PD-1 therapy in colorectal and other cancers, they have not yet been tested for lung cancer, and few have investigated genes from other DNA repair pathways. We utilized TCGA samples to comprehensively immunophenotype lung tumors and analyze the links between DNA repair mutations, neo-antigen and total mutational burden, and tumor immune infiltration. Overall, 73% of lung tumors contained infiltration by at least one T cell subset, with high mutational burden tumors containing significantly increased infiltration by activated CD4 and CD8 T cells. Further, mutations in mismatch repair genes, homologous recombination genes, or POLE accurately predicted increased tumor mutational burden, neo-antigen load, and T cell infiltration. Finally, neo-antigen load correlated with expression of M1-polarized macrophage genes, PD-1, PD-L1, IFNγ, GZMB, and FASLG, among other immune-related genes. Overall, after defining the immune infiltrate in lung tumors, we demonstrate the potential value of utilizing gene mutations from multiple DNA repair pathways as biomarkers for lung cancer immunotherapy.

CloudNeo: a cloud pipeline for identifying patient-specific tumor neoantigens.

SUMMARY: We present CloudNeo, a cloud-based computational workflow for identifying patient-specific tumor neoantigens from next generation sequencing data. Tumor-specific mutant peptides can be detected by the immune system through their interactions with the human leukocyte antigen complex, and neoantigen presence has recently been shown to correlate with anti T-cell immunity and efficacy of checkpoint inhibitor therapy. However computing capabilities to identify neoantigens from genomic sequencing data are a limiting factor for understanding their role. This challenge has grown as cancer datasets become increasingly abundant, making them cumbersome to store and analyze on local servers. Our cloud-based pipeline provides scalable computation capabilities for neoantigen identification while eliminating the need to invest in local infrastructure for data transfer, storage or compute. The pipeline is a Common Workflow Language (CWL) implementation of human leukocyte antigen (HLA) typing using Polysolver or HLAminer combined with custom scripts for mutant peptide identification and NetMHCpan for neoantigen prediction. We have demonstrated the efficacy of these pipelines on Amazon cloud instances through the Seven Bridges Genomics implementation of the NCI Cancer Genomics Cloud, which provides graphical interfaces for running and editing, infrastructure for workflow sharing and version tracking, and access to TCGA data. AVAILABILITY AND IMPLEMENTATION: The CWL implementation is at: https://github.com/TheJacksonLaboratory/CloudNeo. For users who have obtained licenses for all internal software, integrated versions in CWL and on the Seven Bridges Cancer Genomics Cloud platform (https://cgc.sbgenomics.com/, recommended version) can be obtained by contacting the authors. CONTACT: jeff.chuang@jax.org. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Integrating 5-hydroxymethylcytosine into the epigenomic landscape of human embryonic stem cells.

Covalent modification of DNA distinguishes cellular identities and is crucial for regulating the pluripotency and differentiation of embryonic stem (ES) cells. The recent demonstration that 5-methylcytosine (5-mC) may be further modified to 5-hydroxymethylcytosine (5-hmC) in ES cells has revealed a novel regulatory paradigm to modulate the epigenetic landscape of pluripotency. To understand the role of 5-hmC in the epigenomic landscape of pluripotent cells, here we profile the genome-wide 5-hmC distribution and correlate it with the genomic profiles of 11 diverse histone modifications and six transcription factors in human ES cells. By integrating genomic 5-hmC signals with maps of histone enrichment, we link particular pluripotency-associated chromatin contexts with 5-hmC. Intriguingly, through additional correlations with defined chromatin signatures at promoter and enhancer subtypes, we show distinct enrichment of 5-hmC at enhancers marked with H3K4me1 and H3K27ac. These results suggest potential role(s) for 5-hmC in the regulation of specific promoters and enhancers. In addition, our results provide a detailed epigenomic map of 5-hmC from which to pursue future functional studies on the diverse regulatory roles associated with 5-hmC.