Metabolic Dynamics and Prediction of Gestational Age and Time to Delivery in Pregnant Women.

Metabolism during pregnancy is a dynamic and precisely programmed process, the failure of which can bring devastating consequences to the mother and fetus. To define a high-resolution temporal profile of metabolites during healthy pregnancy, we analyzed the untargeted metabolome of 784 weekly blood samples from 30 pregnant women. Broad changes and a highly choreographed profile were revealed: 4,995 metabolic features (of 9,651 total), 460 annotated compounds (of 687 total), and 34 human metabolic pathways (of 48 total) were significantly changed during pregnancy. Using linear models, we built a metabolic clock with five metabolites that time gestational age in high accordance with ultrasound (R = 0.92). Furthermore, two to three metabolites can identify when labor occurs (time to delivery within two, four, and eight weeks, AUROC ≥ 0.85). Our study represents a weekly characterization of the human pregnancy metabolome, providing a high-resolution landscape for understanding pregnancy with potential clinical utilities.

A whole-slide foundation model for digital pathology from real-world data.

Digital pathology poses unique computational challenges, as a standard gigapixel slide may comprise tens of thousands of image tiles1-3. Prior models have often resorted to subsampling a small portion of tiles for each slide, thus missing the important slide-level context4. Here we present Prov-GigaPath, a whole-slide pathology foundation model pretrained on 1.3 billion 256 × 256 pathology image tiles in 171,189 whole slides from Providence, a large US health network comprising 28 cancer centres. The slides originated from more than 30,000 patients covering 31 major tissue types. To pretrain Prov-GigaPath, we propose GigaPath, a novel vision transformer architecture for pretraining gigapixel pathology slides. To scale GigaPath for slide-level learning with tens of thousands of image tiles, GigaPath adapts the newly developed LongNet5 method to digital pathology. To evaluate Prov-GigaPath, we construct a digital pathology benchmark comprising 9 cancer subtyping tasks and 17 pathomics tasks, using both Providence and TCGA data6. With large-scale pretraining and ultra-large-context modelling, Prov-GigaPath attains state-of-the-art performance on 25 out of 26 tasks, with significant improvement over the second-best method on 18 tasks. We further demonstrate the potential of Prov-GigaPath on vision-language pretraining for pathology7,8 by incorporating the pathology reports. In sum, Prov-GigaPath is an open-weight foundation model that achieves state-of-the-art performance on various digital pathology tasks, demonstrating the importance of real-world data and whole-slide modelling.

Longitudinal multi-omics of host-microbe dynamics in prediabetes.

Type 2 diabetes mellitus (T2D) is a growing health problem, but little is known about its early disease stages, its effects on biological processes or the transition to clinical T2D. To understand the earliest stages of T2D better, we obtained samples from 106 healthy individuals and individuals with prediabetes over approximately four years and performed deep profiling of transcriptomes, metabolomes, cytokines, and proteomes, as well as changes in the microbiome. This rich longitudinal data set revealed many insights: first, healthy profiles are distinct among individuals while displaying diverse patterns of intra- and/or inter-personal variability. Second, extensive host and microbial changes occur during respiratory viral infections and immunization, and immunization triggers potentially protective responses that are distinct from responses to respiratory viral infections. Moreover, during respiratory viral infections, insulin-resistant participants respond differently than insulin-sensitive participants. Third, global co-association analyses among the thousands of profiled molecules reveal specific host-microbe interactions that differ between insulin-resistant and insulin-sensitive individuals. Last, we identified early personal molecular signatures in one individual that preceded the onset of T2D, including the inflammation markers interleukin-1 receptor agonist (IL-1RA) and high-sensitivity C-reactive protein (CRP) paired with xenobiotic-induced immune signalling. Our study reveals insights into pathways and responses that differ between glucose-dysregulated and healthy individuals during health and disease and provides an open-access data resource to enable further research into healthy, prediabetic and T2D states.

Multi-omic profiling of simultaneous ductal carcinoma in situ and invasive breast cancer.

PURPOSE: The progression of ductal carcinoma in situ (DCIS) to invasive breast carcinoma (IBC) in humans is highly variable. To better understand the relationship between them, we performed a multi-omic characterization of co-occurring DCIS and IBC lesions in a cohort of individuals. METHODS: Formalin-fixed paraffin-embedded tissue samples from 50 patients with co-occurring DCIS and IBC lesions were subjected to DNA-seq and whole transcriptome RNA-seq. Paired DCIS and IBC multi-omics profiles were then interrogated for DNA mutations, gene expression profiles and pathway analysis. RESULTS: Most small variants and copy number variations were shared between co-occurring DCIS and IBC lesions, with IBC exhibiting on average a higher degree of additional mutations. However, 36% of co-occurring lesions shared no common mutations and 49% shared no common copy number variations. The most frequent genomic variants in both DCIS and IBC were PIK3CA, TP53, KMT2C, MAP3K1, GATA3 and SF3B1, with KMT2C being more frequent in DCIS and TP53 and MAP3K1 more frequent in IBC, though the numbers are too small for definitive conclusions. The most frequent copy number variations were seen in MCL1, CKSB1 and ERBB2. ERBB2 changes were not seen in IBC unless present in the corresponding DCIS. Transcriptional profiles were highly distinct between DCIS and IBC, with DCIS exhibiting upregulation of immune-related signatures, while IBC showed significant overexpression in genes and pathways associated with cell division and proliferation. Interestingly, DCIS and IBC exhibited significant differential expression of different components of extracellular matrix (ECM) formation and regulation, with DCIS showing overexpression of ECM-membrane interaction components while IBC showed upregulation of genes associated with fibronectin and invadopodia. CONCLUSION: While most co-occurring DCIS and IBC were mutationally similar and suggestive of a common clonal progenitor, transcriptionally the lesions are highly distinct, with IBC expressing key pathways that facilitate invasion and proliferation. These results are suggestive of additional levels of regulation, epigenetic or other, that facilitate the acquisition of invasive properties during tumor evolution.

Unsupervised mRNA-seq classification of heart transplant endomyocardial biopsies.

BACKGROUND: Endomyocardial biopsy (EMB) is currently considered the gold standard for diagnosing cardiac allograft rejection. However, significant limitations related to histological interpretation variability are well-recognized. We sought to develop a methodology to evaluate EMB solely based on gene expression, without relying on histology interpretation. METHODS: Sixty-four EMBs were obtained from 47 post-heart transplant recipients, who were evaluated for allograft rejection. EMBs were subjected to mRNA sequencing, in which an unsupervised classification algorithm was used to identify the molecular signatures that best classified the EMBs. Cytokine and natriuretic peptide peripheral blood profiling was also performed. Subsequently, we performed gene network analysis to identify the gene modules and gene ontology to understand their biological relevance. We correlated our findings with the unsupervised and histological classifications. RESULTS: Our algorithm classifies EMBs into three categories based solely on clusters of gene expression: unsupervised classes 1, 2, and 3. Unsupervised and histological classifications were closely related, with stronger gene module-phenotype correlations for the unsupervised classes. Gene ontology enrichment analysis revealed processes impacting on the regulation of cardiac and mitochondrial function, immune response, and tissue injury response. Significant levels of cytokines and natriuretic peptides were detected following the unsupervised classification. CONCLUSION: We have developed an unsupervised algorithm that classifies EMBs into three distinct categories, without relying on histology interpretation. These categories were highly correlated with mitochondrial, immune, and tissue injury response. Significant cytokine and natriuretic peptide levels were detected within the unsupervised classification. If further validated, the unsupervised classification could offer a more objective EMB evaluation.

Gut microbiota analyses of inflammatory bowel diseases from a representative Saudi population.

BACKGROUND: Crohn's diseases and ulcerative colitis, both of which are chronic immune-mediated disorders of the gastrointestinal tract are major contributors to the overarching Inflammatory bowel diseases. It has become increasingly evident that the pathological processes of IBDs results from interactions between genetic and environmental factors, which can skew immune responses against normal intestinal flora. METHODS: The aim of this study is to assess and analyze the taxa diversity and relative abundances in CD and UC in the Saudi population. We utilized a sequencing strategy that targets all variable regions in the 16 S rRNA gene using the Swift Amplicon 16 S rRNA Panel on Illumina NovaSeq 6000. RESULTS: The composition of stool 16 S rRNA was analyzed from 219 patients with inflammatory bowel disease and from 124 healthy controls. We quantified the abundance of microbial communities to examine any significant differences between subpopulations of samples. At the genus level, two genera in particular, Veillonella and Lachnoclostridium showed significant association with CD versus controls. There were significant differences between subjects with CD versus UC, with the top differential genera spanning Akkermansia, Harryflintia, Maegamonas and Phascolarctobacterium. Furthermore, statistically significant taxa diversity in microbiome composition was observed within the UC and CD groups. CONCLUSIONS: In conclusion we have shown that there are significant differences in gut microbiota between UC, CD and controls in a Saudi Arabian inflammatory bowel disease cohort. This reinforces the need for further studies in large populations that are ethnically and geographically diverse. In addition, our results show the potential to develop classifiers that may have add additional richness of context to clinical diagnosis of UC and CD with larger inflammatory bowel disease cohorts.

iNetModels 2.0: an interactive visualization and database of multi-omics data.

It is essential to reveal the associations between various omics data for a comprehensive understanding of the altered biological process in human wellness and disease. To date, very few studies have focused on collecting and exhibiting multi-omics associations in a single database. Here, we present iNetModels, an interactive database and visualization platform of Multi-Omics Biological Networks (MOBNs). This platform describes the associations between the clinical chemistry, anthropometric parameters, plasma proteomics, plasma metabolomics, as well as metagenomics for oral and gut microbiome obtained from the same individuals. Moreover, iNetModels includes tissue- and cancer-specific Gene Co-expression Networks (GCNs) for exploring the connections between the specific genes. This platform allows the user to interactively explore a single feature's association with other omics data and customize its particular context (e.g. male/female specific). The users can also register their data for sharing and visualization of the MOBNs and GCNs. Moreover, iNetModels allows users who do not have a bioinformatics background to facilitate human wellness and disease research. iNetModels can be accessed freely at https://inetmodels.com without any limitation.

Oral microbiota analyses of paediatric Saudi population reveals signatures of dental caries.

BACKGROUND: Oral microbiome sequencing has revealed key links between microbiome dysfunction and dental caries. However, these efforts have largely focused on Western populations, with few studies on the Middle Eastern communities. The current study aimed to identify the composition and abundance of the oral microbiota in saliva samples of children with different caries levels using machine learning approaches. METHODS: Oral microbiota composition and abundance were identified in 250 Saudi participants with high dental caries and 150 with low dental caries using 16 S rRNA sequencing on a NextSeq 2000 SP flow cell (Illumina, CA) using 250 bp paired-end reads, and attempted to build a classifier using random forest models to assist in the early detection of caries. RESULTS: The ADONIS test results indicate that there was no significant association between sex and Bray-Curtis dissimilarity (p ~ 0.93), but there was a significant association with dental caries status (p ~ 0.001). Using an alpha level of 0.05, five differentially abundant operational taxonomic units (OTUs) were identified between males and females as the main effect along with four differentially abundant OTUs between high and low dental caries. The mean metrics for the optimal hyperparameter combination using the model with only differentially abundant OTUs were: Accuracy (0.701); Matthew's correlation coefficient (0.0509); AUC (0.517) and F1 score (0.821) while the mean metrics for random forest model using all OTUs were:0.675; 0.054; 0.611 and 0.796 respectively. CONCLUSION: The assessment of oral microbiota samples in a representative Saudi Arabian population for high and low metrics of dental caries yields signatures of abundances and diversity.

Gut microbiota analyses of Saudi populations for type 2 diabetes-related phenotypes reveals significant association.

BACKGROUND: Large-scale gut microbiome sequencing has revealed key links between microbiome dysfunction and metabolic diseases such as type 2 diabetes (T2D). To date, these efforts have largely focused on Western populations, with few studies assessing T2D microbiota associations in Middle Eastern communities where T2D prevalence is now over 20%. We analyzed the composition of stool 16S rRNA from 461 T2D and 119 non-T2D participants from the Eastern Province of Saudi Arabia. We quantified the abundance of microbial communities to examine any significant differences between subpopulations of samples based on diabetes status and glucose level. RESULTS: In this study we performed the largest microbiome study ever conducted in Saudi Arabia, as well as the first-ever characterization of gut microbiota T2D versus non-T2D in this population. We observed overall positive enrichment within diabetics compared to healthy individuals and amongst diabetic participants; those with high glucose levels exhibited slightly more positive enrichment compared to those at lower risk of fasting hyperglycemia. In particular, the genus Firmicutes was upregulated in diabetic individuals compared to non-diabetic individuals, and T2D was associated with an elevated Firmicutes/Bacteroidetes ratio, consistent with previous findings. CONCLUSION: Based on diabetes status and glucose levels of Saudi participants, relatively stable differences in stool composition were perceived by differential abundance and alpha diversity measures. However, community level differences are evident in the Saudi population between T2D and non-T2D individuals, and diversity patterns appear to vary from well-characterized microbiota from Western cohorts. Comparing overlapping and varying patterns in gut microbiota with other studies is critical to assessing novel treatment options in light of a rapidly growing T2D health epidemic in the region. As a rapidly emerging chronic condition in Saudi Arabia and the Middle East, T2D burdens have grown more quickly and affect larger proportions of the population than any other global region, making a regional reference T2D-microbiome dataset critical to understanding the nuances of disease development on a global scale.

A rapid and universal liquid chromatograph-mass spectrometry-based platform, refmAb-Q nSMOL, for monitoring monoclonal antibody therapeutics.

Accurate quantitation of antibodies is critical for development of monoclonal antibody therapeutics (mAbs). Therapeutic drug monitoring has been applied to measure levels of mAbs in clinics for dose adjustment for autoimmune disease. Trough levels of mAbs can be a biomarker for cancer immunotherapy. Thus, the deployment of a rapid and universal platform for mAb monitoring may benefit processes ranging from drug development to clinical practice for a wide spectrum of diseases. However, mAb monitoring often requires development and conduct of an individual ligand binding assay such as ELISA, which is impractical to scale. We streamlined quantitation of antibody therapeutics by a nano-surface and molecular-orientation limited (nSMOL) proteolysis assay using LC-MS with a universal reference antibody (refmAb-Q), for accurate multiplexed quantitation of unique signature peptides derived from mAbs. This innovative refmAb-Q nSMOL platform may provide a practical solution for quantitating an ever-increasing number of mAbs from developmental to clinical use settings.

Network analyses identify liver-specific targets for treating liver diseases.

We performed integrative network analyses to identify targets that can be used for effectively treating liver diseases with minimal side effects. We first generated co-expression networks (CNs) for 46 human tissues and liver cancer to explore the functional relationships between genes and examined the overlap between functional and physical interactions. Since increased de novo lipogenesis is a characteristic of nonalcoholic fatty liver disease (NAFLD) and hepatocellular carcinoma (HCC), we investigated the liver-specific genes co-expressed with fatty acid synthase (FASN). CN analyses predicted that inhibition of these liver-specific genes decreases FASN expression. Experiments in human cancer cell lines, mouse liver samples, and primary human hepatocytes validated our predictions by demonstrating functional relationships between these liver genes, and showing that their inhibition decreases cell growth and liver fat content. In conclusion, we identified liver-specific genes linked to NAFLD pathogenesis, such as pyruvate kinase liver and red blood cell (PKLR), or to HCC pathogenesis, such as PKLR, patatin-like phospholipase domain containing 3 (PNPLA3), and proprotein convertase subtilisin/kexin type 9 (PCSK9), all of which are potential targets for drug development.

Applying genomics in heart transplantation.

While advances in patient care and immunosuppressive pharmacotherapies have increased the lifespan of heart allograft recipients, there are still significant comorbidities post-transplantation and 5-year survival rates are still significant, at approximately 70%. The last decade has seen massive strides in genomics and other omics fields, including transcriptomics, with many of these advances now starting to impact heart transplant clinical care. This review summarizes a number of the key advances in genomics which are relevant for heart transplant outcomes, and we highlight the translational potential that such knowledge may bring to patient care within the next decade.

Fluorescence tracking demonstrates T cell recirculation is transiently impaired by radiation therapy to the tumor.

T cells recirculate through tissues and lymphatic organs to scan for their cognate antigen. Radiation therapy provides site-specific cytotoxicity to kill cancer cells but also has the potential to eliminate the tumor-specific T cells in field. To dynamically study the effect of radiation on CD8 T cell recirculation, we used the Kaede mouse model to photoconvert tumor-infiltrating cells and monitor their movement out of the field of radiation. We demonstrate that radiation results in loss of CD8 T cell recirculation from the tumor to the lymph node and to distant sites. Using scRNASeq, we see decreased proliferating CD8 T cells in the tumor following radiation therapy resulting in a proportional enrichment in exhausted phenotypes. By contrast, 5 days following radiation increased recirculation of T cells from the tumor to the tumor draining lymph node corresponds with increased immunosurveillance of the treated tumor. These data demonstrate that tumor radiation therapy transiently impairs systemic T cell recirculation from the treatment site to the draining lymph node and distant untreated tumors. This may inform timing therapies to improve systemic T cell-mediated tumor immunity.

FcγRIIB Is an Immune Checkpoint Limiting the Activity of Treg-Targeting Antibodies in the Tumor Microenvironment.

Preclinical murine data indicate that fragment crystallizable (Fc)-dependent depletion of intratumoral regulatory T cells (Treg) is a major mechanism of action of anti-CTLA-4. However, the two main antibodies administered to patients (ipilimumab and tremelimumab) do not recapitulate these effects. Here, we investigate the underlying mechanisms responsible for the limited Treg depletion observed with these therapies. Using an immunocompetent murine model humanized for CTLA-4 and Fcγ receptors (FcγR), we show that ipilimumab and tremelimumab exhibit limited Treg depletion in tumors. Immune profiling of the tumor microenvironment (TME) in both humanized mice and humans revealed high expression of the inhibitory Fc receptor, FcγRIIB, which limits antibody-dependent cellular cytotoxicity/phagocytosis. Blocking FcγRIIB in humanized mice rescued the Treg-depleting capacity and antitumor activity of ipilimumab. Furthermore, Fc engineering of antibodies targeting Treg-associated targets (CTLA-4 or CCR8) to minimize FcγRIIB binding significantly enhanced Treg depletion, resulting in increased antitumor activity across various tumor models. Our results define the inhibitory FcγRIIB as an immune checkpoint limiting antibody-mediated Treg depletion in the TME, and demonstrate Fc engineering as an effective strategy to overcome this limitation and improve the efficacy of Treg-targeting antibodies.

Integrative multi-omics profiling in human decedents receiving pig heart xenografts.

In a previous study, heart xenografts from 10-gene-edited pigs transplanted into two human decedents did not show evidence of acute-onset cellular- or antibody-mediated rejection. Here, to better understand the detailed molecular landscape following xenotransplantation, we carried out bulk and single-cell transcriptomics, lipidomics, proteomics and metabolomics on blood samples obtained from the transplanted decedents every 6 h, as well as histological and transcriptomic tissue profiling. We observed substantial early immune responses in peripheral blood mononuclear cells and xenograft tissue obtained from decedent 1 (male), associated with downstream T cell and natural killer cell activity. Longitudinal analyses indicated the presence of ischemia reperfusion injury, exacerbated by inadequate immunosuppression of T cells, consistent with previous findings of perioperative cardiac xenograft dysfunction in pig-to-nonhuman primate studies. Moreover, at 42 h after transplantation, substantial alterations in cellular metabolism and liver-damage pathways occurred, correlating with profound organ-wide physiological dysfunction. By contrast, relatively minor changes in RNA, protein, lipid and metabolism profiles were observed in decedent 2 (female) as compared to decedent 1. Overall, these multi-omics analyses delineate distinct responses to cardiac xenotransplantation in the two human decedents and reveal new insights into early molecular and immune responses after xenotransplantation. These findings may aid in the development of targeted therapeutic approaches to limit ischemia reperfusion injury-related phenotypes and improve outcomes.

Myeloid MyD88 restricts CD8+ T cell response to radiation therapy in pancreatic cancer.

Radiation therapy induces immunogenic cell death in cancer cells, whereby released endogenous adjuvants are sensed by immune cells to direct adaptive immune responses. TLRs expressed on several immune subtypes recognize innate adjuvants to direct downstream inflammatory responses in part via the adapter protein MyD88. We generated Myd88 conditional knockout mice to interrogate its contribution to the immune response to radiation therapy in distinct immune populations in pancreatic cancer. Surprisingly, Myd88 deletion in Itgax (CD11c)-expressing dendritic cells had little discernable effects on response to RT in pancreatic cancer and elicited normal T cell responses using a prime/boost vaccination strategy. Myd88 deletion in Lck-expressing T cells resulted in similar or worsened responses to radiation therapy compared to wild-type mice and lacked antigen-specific CD8+ T cell responses from vaccination, similar to observations in Myd88-/- mice. Lyz2-specific loss of Myd88 in myeloid populations rendered tumors more susceptible to radiation therapy and elicited normal CD8+ T cell responses to vaccination. scRNAseq in Lyz2-Cre/Myd88fl/fl mice revealed gene signatures in macrophages and monocytes indicative of enhanced type I and II interferon responses, and improved responses to RT were dependent on CD8+ T cells and IFNAR1. Together, these data implicate MyD88 signaling in myeloid cells as a critical source of immunosuppression that hinders adaptive immune tumor control following radiation therapy.

FcyRIIB is a novel immune checkpoint in the tumor microenvironment limiting activity of Treg-targeting antibodies.

Despite pre-clinical murine data supporting T regulatory (Treg) cell depletion as a major mechanism by which anti-CTLA-4 antibodies function in vivo, the two main antibodies tested in patients (ipilimumab and tremelimumab) have failed to demonstrate similar effects. We report analogous findings in an immunocompetent murine model humanized for CTLA-4 and Fcy receptors (hCTLA-4/hFcyR mice), where both ipilimumab and tremelimumab fail to show appreciable Treg depletion. Immune profiling of the tumor microenvironment (TME) in both mice and human samples revealed upregulation of the inhibitory Fcy receptor, FcyRIIB, which limits the ability of the antibody Fc fragment of human anti-CTLA-4 antibodies to induce effective antibody dependent cellular cytotoxicty/phagocytosis (ADCC/ADCP). Blocking FcyRIIB in humanized mice rescues Treg depleting capacity and anti-tumor activity of ipilimumab. For another target, CC motif chemokine receptor 8 (CCR8), which is selectively expressed on tumor infiltrating Tregs, we show that Fc engineering to enhance binding to activating Fc receptors, while limiting binding to the inhibitory Fc receptor, leads to consistent Treg depletion and single-agent activity across multiple tumor models, including B16, MC38 and MB49. These data reveal the importance of reducing engagement to the inhibitory Fc receptor to optimize Treg depletion by TME targeting antibodies. Our results define the inhibitory FcyRIIB receptor as a novel immune checkpoint limiting antibody-mediated Treg depletion in tumors, and demonstrate Fc variant engineering as a means to overcome this limitation and augment efficacy for a repertoire of antibodies currently in use or under clinical evaluation in oncology.

Toward structuring real-world data: Deep learning for extracting oncology information from clinical text with patient-level supervision.

Most detailed patient information in real-world data (RWD) is only consistently available in free-text clinical documents. Manual curation is expensive and time consuming. Developing natural language processing (NLP) methods for structuring RWD is thus essential for scaling real-world evidence generation. We propose leveraging patient-level supervision from medical registries, which are often readily available and capture key patient information, for general RWD applications. We conduct an extensive study on 135,107 patients from the cancer registry of a large integrated delivery network (IDN) comprising healthcare systems in five western US states. Our deep-learning methods attain test area under the receiver operating characteristic curve (AUROC) values of 94%-99% for key tumor attributes and comparable performance on held-out data from separate health systems and states. Ablation results demonstrate the superiority of these advanced deep-learning methods. Error analysis shows that our NLP system sometimes even corrects errors in registrar labels.

Uterine Inflammatory Myofibroblastic Tumors: Proposed Risk Stratification Model Using Integrated Clinicopathologic and Molecular Analysis.

Inflammatory myofibroblastic tumor (IMT) of the uterus is a rare mesenchymal tumor with largely benign behavior; however, a small subset demonstrate aggressive behavior. While clinicopathologic features have been previously associated with aggressive behavior, these reports are based on small series, and these features are imperfect predictors of clinical behavior. IMTs are most commonly driven by ALK fusions, with additional pathogenic molecular alterations being reported only in rare examples of extrauterine IMTs. In this study, a series of 11 uterine IMTs, 5 of which demonstrated aggressive behavior, were evaluated for clinicopathologic variables and additionally subjected to capture-based next-generation sequencing with or without whole-transcriptome RNA sequencing. In the 6 IMTs without aggressive behavior, ALK fusions were the sole pathogenic alteration. In contrast, all 5 aggressive IMTs harbored pathogenic molecular alterations and numerous copy number changes in addition to ALK fusions, with the majority of the additional alterations present in the primary tumors. We combined our series with cases previously reported in the literature and performed statistical analyses to propose a novel clinicopathologic risk stratification score assigning 1 point each for: age above 45 years, size≥5 cm,≥4 mitotic figures per 10 high-power field, and infiltrative borders. No tumors with 0 points had an aggressive outcome, while 21% of tumors with 1 to 2 points and all tumors with ≥3 points had aggressive outcomes. We propose a 2-step classification model that first uses the clinicopathologic risk stratification score to identify low-risk and high-risk tumors, and recommend molecular testing to further classify intermediate-risk tumors.

A Random Forest Genomic Classifier for Tumor Agnostic Prediction of Response to Anti-PD1 Immunotherapy.

Tumor mutational burden (TMB), a surrogate for tumor neoepitope burden, is used as a pan-tumor biomarker to identify patients who may benefit from anti-program cell death 1 (PD1) immunotherapy, but it is an imperfect biomarker. Multiple additional genomic characteristics are associated with anti-PD1 responses, but the combined predictive value of these features and the added informativeness of each respective feature remains unknown. We evaluated whether machine learning (ML) approaches using proposed determinants of anti-PD1 response derived from whole exome sequencing (WES) could improve prediction of anti-PD1 responders over TMB alone. Random forest classifiers were trained on publicly available anti-PD1 data (n = 104), and subsequently tested on an independent anti-PD1 cohort (n = 69). Both the training and test datasets included a range of cancer types such as non-small cell lung cancer (NSCLC), head and neck squamous cell carcinoma (HNSCC), melanoma, and smaller numbers of patients from other tumor types. Features used include summaries such as TMB and number of frameshift mutations, as well as more gene-level features such as counts of mutations associated with immune checkpoint response and resistance. Both ML algorithms demonstrated area under the receiver-operator curves (AUC) that exceeded TMB alone (AUC 0.63 "human-guided," 0.64 "cluster," and 0.58 TMB alone). Mutations within oncogenes disproportionately modulate anti-PD1 responses relative to their overall contribution to tumor neoepitope burden. The use of a ML algorithm evaluating multiple proposed genomic determinants of anti-PD1 responses modestly improves performance over TMB alone, highlighting the need to integrate other biomarkers to further improve model performance.