Machine learning in rare disease.

High-throughput profiling methods (such as genomics or imaging) have accelerated basic research and made deep molecular characterization of patient samples routine. These approaches provide a rich portrait of genes, molecular pathways and cell types involved in disease phenotypes. Machine learning (ML) can be a useful tool for extracting disease-relevant patterns from high-dimensional datasets. However, depending upon the complexity of the biological question, machine learning often requires many samples to identify recurrent and biologically meaningful patterns. Rare diseases are inherently limited in clinical cases, leading to few samples to study. In this Perspective, we outline the challenges and emerging solutions for using ML for small sample sets, specifically in rare diseases. Advances in ML methods for rare diseases are likely to be informative for applications beyond rare diseases for which few samples exist with high-dimensional data. We propose that the method community prioritize the development of ML techniques for rare disease research.

Cross-platform normalization enables machine learning model training on microarray and RNA-seq data simultaneously.

Large compendia of gene expression data have proven valuable for the discovery of novel biological relationships. Historically, most available RNA assays were run on microarray, while RNA-seq is now the platform of choice for many new experiments. The data structure and distributions between the platforms differ, making it challenging to combine them directly. Here we perform supervised and unsupervised machine learning evaluations to assess which existing normalization methods are best suited for combining microarray and RNA-seq data. We find that quantile and Training Distribution Matching normalization allow for supervised and unsupervised model training on microarray and RNA-seq data simultaneously. Nonparanormal normalization and z-scores are also appropriate for some applications, including pathway analysis with Pathway-Level Information Extractor (PLIER). We demonstrate that it is possible to perform effective cross-platform normalization using existing methods to combine microarray and RNA-seq data for machine learning applications.

GenomicSuperSignature facilitates interpretation of RNA-seq experiments through robust, efficient comparison to public databases.

Millions of transcriptomic profiles have been deposited in public archives, yet remain underused for the interpretation of new experiments. We present a method for interpreting new transcriptomic datasets through instant comparison to public datasets without high-performance computing requirements. We apply Principal Component Analysis on 536 studies comprising 44,890 human RNA sequencing profiles and aggregate sufficiently similar loading vectors to form Replicable Axes of Variation (RAV). RAVs are annotated with metadata of originating studies and by gene set enrichment analysis. Functionality to associate new datasets with RAVs, extract interpretable annotations, and provide intuitive visualization are implemented as the GenomicSuperSignature R/Bioconductor package. We demonstrate the efficient and coherent database search, robustness to batch effects and heterogeneous training data, and transfer learning capacity of our method using TCGA and rare diseases datasets. GenomicSuperSignature aids in analyzing new gene expression data in the context of existing databases using minimal computing resources.

Macrophages in SHH subgroup medulloblastoma display dynamic heterogeneity that varies with treatment modality.

Tumor-associated macrophages (TAMs) play an important role in tumor immunity and comprise of subsets that have distinct phenotype, function, and ontology. Transcriptomic analyses of human medulloblastoma, the most common malignant pediatric brain cancer, showed that medulloblastomas (MBs) with activated sonic hedgehog signaling (SHH-MB) have significantly more TAMs than other MB subtypes. Therefore, we examined MB-associated TAMs by single-cell RNA sequencing of autochthonous murine SHH-MB at steady state and under two distinct treatment modalities: molecular-targeted inhibitor and radiation. Our analyses reveal significant TAM heterogeneity, identify markers of ontologically distinct TAM subsets, and show the impact of brain microenvironment on the differentiation of tumor-infiltrating monocytes. TAM composition undergoes dramatic changes with treatment and differs significantly between molecular-targeted and radiation therapy. We identify an immunosuppressive monocyte-derived TAM subset that emerges with radiation therapy and demonstrate its role in regulating T cell and neutrophil infiltration in MB.

annoFuse: an R Package to annotate, prioritize, and interactively explore putative oncogenic RNA fusions.

BACKGROUND: Gene fusion events are significant sources of somatic variation across adult and pediatric cancers and are some of the most clinically-effective therapeutic targets, yet low consensus of RNA-Seq fusion prediction algorithms makes therapeutic prioritization difficult. In addition, events such as polymerase read-throughs, mis-mapping due to gene homology, and fusions occurring in healthy normal tissue require informed filtering, making it difficult for researchers and clinicians to rapidly discern gene fusions that might be true underlying oncogenic drivers of a tumor and in some cases, appropriate targets for therapy. RESULTS: We developed annoFuse, an R package, and shinyFuse, a companion web application, to annotate, prioritize, and explore biologically-relevant expressed gene fusions, downstream of fusion calling. We validated annoFuse using a random cohort of TCGA RNA-Seq samples (N = 160) and achieved a 96% sensitivity for retention of high-confidence fusions (N = 603). annoFuse uses FusionAnnotator annotations to filter non-oncogenic and/or artifactual fusions. Then, fusions are prioritized if previously reported in TCGA and/or fusions containing gene partners that are known oncogenes, tumor suppressor genes, COSMIC genes, and/or transcription factors. We applied annoFuse to fusion calls from pediatric brain tumor RNA-Seq samples (N = 1028) provided as part of the Open Pediatric Brain Tumor Atlas (OpenPBTA) Project to determine recurrent fusions and recurrently-fused genes within different brain tumor histologies. annoFuse annotates protein domains using the PFAM database, assesses reciprocality, and annotates gene partners for kinase domain retention. As a standard function, reportFuse enables generation of a reproducible R Markdown report to summarize filtered fusions, visualize breakpoints and protein domains by transcript, and plot recurrent fusions within cohorts. Finally, we created shinyFuse for algorithm-agnostic interactive exploration and plotting of gene fusions. CONCLUSIONS: annoFuse provides standardized filtering and annotation for gene fusion calls from STAR-Fusion and Arriba by merging, filtering, and prioritizing putative oncogenic fusions across large cancer datasets, as demonstrated here with data from the OpenPBTA project. We are expanding the package to be widely-applicable to other fusion algorithms and expect annoFuse to provide researchers a method for rapidly evaluating, prioritizing, and translating fusion findings in patient tumors.

Integrative Analysis Identifies Candidate Tumor Microenvironment and Intracellular Signaling Pathways that Define Tumor Heterogeneity in NF1.

Neurofibromatosis type 1 (NF1) is a monogenic syndrome that gives rise to numerous symptoms including cognitive impairment, skeletal abnormalities, and growth of benign nerve sheath tumors. Nearly all NF1 patients develop cutaneous neurofibromas (cNFs), which occur on the skin surface, whereas 40-60% of patients develop plexiform neurofibromas (pNFs), which are deeply embedded in the peripheral nerves. Patients with pNFs have a ~10% lifetime chance of these tumors becoming malignant peripheral nerve sheath tumors (MPNSTs). These tumors have a severe prognosis and few treatment options other than surgery. Given the lack of therapeutic options available to patients with these tumors, identification of druggable pathways or other key molecular features could aid ongoing therapeutic discovery studies. In this work, we used statistical and machine learning methods to analyze 77 NF1 tumors with genomic data to characterize key signaling pathways that distinguish these tumors and identify candidates for drug development. We identified subsets of latent gene expression variables that may be important in the identification and etiology of cNFs, pNFs, other neurofibromas, and MPNSTs. Furthermore, we characterized the association between these latent variables and genetic variants, immune deconvolution predictions, and protein activity predictions.

Making Workshops Work: Insights from EDAMAME.

Microbiology, like many areas of life science research, is increasingly data-intensive. As such, bioinformatics and data science skills have become essential to leverage microbiome sequencing data for discovery. Short intensive courses have sprung up as formal computational training opportunities at individual institutions fail to meet demands. In this issue, Shade et al. (A. Shade, T. K. Dunivin, J. Choi, T. K. Teal, et al., mSystems 4:e00297-19, 2019, https://doi.org/10.1128/mSystems.00297-19) share their experience and approach in executing the annual, weeklong Explorations in Data Analysis for Metagenomic Advances in Microbial Ecology (EDAMAME) workshop from 2014 to 2018. EDAMAME introduced learners to general scientific computing concepts and domain-specific data analysis approaches. Workshop learners self-reported appreciable gains in understanding and ability. This report on the EDAMAME workshop strategy and lessons learned will help others in the life sciences to plan, execute, and assess short hands-on computing-intensive courses that support research in a particular domain.

MultiPLIER: A Transfer Learning Framework for Transcriptomics Reveals Systemic Features of Rare Disease.

Most gene expression datasets generated by individual researchers are too small to fully benefit from unsupervised machine-learning methods. In the case of rare diseases, there may be too few cases available, even when multiple studies are combined. To address this challenge, we utilize transfer learning to extract coordinated expression patterns and use learned patterns to analyze small rare disease datasets. We trained a pathway-level information extractor (PLIER) model on a large public data compendium comprising multiple experiments, tissues, and biological conditions and then transferred the model to small datasets in an approach we call MultiPLIER. Models constructed from the public data compendium included features that aligned well to known biological factors and were more comprehensive than those constructed from individual datasets or conditions. When transferred to rare disease datasets, the models describe biological processes related to disease severity more effectively than models trained only on a given dataset.

Metabolic pathways and immunometabolism in rare kidney diseases.

OBJECTIVES: To characterise renal tissue metabolic pathway gene expression in different forms of glomerulonephritis. METHODS: Patients with nephrotic syndrome (NS), antineutrophil cytoplasmic antibody-associated vasculitis (AAV), systemic lupus erythematosus (SLE) and healthy living donors (LD) were studied. Clinically indicated renal biopsies were obtained at time of diagnosis and microdissected into glomerular and tubulointerstitial compartments. Microarray-derived differential gene expression of 88 genes representing critical enzymes of metabolic pathways and 25 genes related to immune cell markers was compared between disease groups. Correlation analyses measured relationships between metabolic pathways, kidney function and cytokine production. RESULTS: Reduced steady state levels of mRNA species were enriched in pathways of oxidative phosphorylation and increased in the pentose phosphate pathway (PPP) with maximal perturbation in AAV and SLE followed by NS, and least in LD. Transcript regulation was isozymes specific with robust regulation in hexokinases, enolases and glucose transporters. Intercorrelation networks were observed between enzymes of the PPP (eg, transketolase) and macrophage markers (eg, CD68) (r=0.49, p<0.01). Increased PPP transcript levels were associated with reduced glomerular filtration rate in the glomerular (r=-0.49, p<0.01) and tubulointerstitial (r=-0.41, p<0.01) compartments. PPP expression and tumour necrosis factor activation were tightly co-expressed (r=0.70, p<0.01). CONCLUSION: This study demonstrated concordant alterations of the renal transcriptome consistent with metabolic reprogramming across different forms of glomerulonephritis. Activation of the PPP was tightly linked with intrarenal macrophage marker expression, reduced kidney function and increased production of cytokines. Modulation of glucose metabolism may offer novel immune-modulatory therapeutic approaches in rare kidney diseases.

Mycophenolate Mofetil Treatment of Systemic Sclerosis Reduces Myeloid Cell Numbers and Attenuates the Inflammatory Gene Signature in Skin.

Fewer than half of patients with systemic sclerosis demonstrate modified Rodnan skin score improvement during mycophenolate mofetil (MMF) treatment. To understand the molecular basis for this observation, we extended our prior studies and characterized molecular and cellular changes in skin biopsies from subjects with systemic sclerosis treated with MMF. Eleven subjects completed ≥24 months of MMF therapy. Two distinct skin gene expression trajectories were observed across six of these subjects. Three of the six subjects showed attenuation of the inflammatory signature by 24 months, paralleling reductions in CCL2 mRNA expression in skin and reduced numbers of macrophages and myeloid dendritic cells in skin biopsies. MMF cessation at 24 months resulted in an increased inflammatory score, increased CCL2 mRNA and protein levels, modified Rodnan skin score rebound, and increased numbers of skin myeloid cells in these subjects. In contrast, three other subjects remained on MMF >24 months and showed a persistent decrease in inflammatory score, decreasing or stable modified Rodnan skin score, CCL2 mRNA reductions, sera CCL2 protein levels trending downward, reduction in monocyte migration, and no increase in skin myeloid cell numbers. These data summarize molecular changes during MMF therapy that suggest reduction of innate immune cell numbers, possibly by attenuating expression of chemokines, including CCL2.

A Multimodal Strategy Used by a Large c-di-GMP Network.

The Pseudomonas fluorescens genome encodes more than 50 proteins predicted to be involved in c-di-GMP signaling. Here, we demonstrated that, tested across 188 nutrients, these enzymes and effectors appeared capable of impacting biofilm formation. Transcriptional analysis of network members across ∼50 nutrient conditions indicates that altered gene expression can explain a subset of but not all biofilm formation responses to the nutrients. Additional organization of the network is likely achieved through physical interaction, as determined via probing ∼2,000 interactions by bacterial two-hybrid assays. Our analysis revealed a multimodal regulatory strategy using combinations of ligand-mediated signals, protein-protein interaction, and/or transcriptional regulation to fine-tune c-di-GMP-mediated responses. These results create a profile of a large c-di-GMP network that is used to make important cellular decisions, opening the door to future model building and the ability to engineer this complex circuitry in other bacteria.IMPORTANCE Cyclic diguanylate (c-di-GMP) is a key signaling molecule regulating bacterial biofilm formation, and many microbes have up to dozens of proteins that make, break, or bind this dinucleotide. A major open issue in the field is how signaling specificity is conferred in the unpartitioned space of a bacterial cell. Here, we took a systems approach, using mutational analysis, transcriptional studies, and bacterial two-hybrid analysis to interrogate this network. We found that a majority of enzymes are capable of impacting biofilm formation in a context-dependent manner, and we revealed examples of two or more modes of regulation (i.e., transcriptional control with protein-protein interaction) being utilized to generate an observable impact on biofilm formation.

Integrated machine learning pipeline for aberrant biomarker enrichment (i-mAB): characterizing clusters of differentiation within a compendium of systemic lupus erythematosus patients.

Clusters of differentiation (CD) are cell surface biomarkers that denote key biological differences between cell types and disease state. CD-targeting therapeutic monoclonal antibodies (mABs) afford rich trans-disease repositioning opportunities. Within a compendium of systemic lupus erythematous (SLE) patients, we applied the Integrated machine learning pipeline for aberrant biomarker enrichment (i-mAB) to profile de novo gene expression features affecting CD20, CD22 and CD30 gene aberrance. First, a novel Relief-based algorithm identified interdependent features(p=681) predicting treatment-naïve SLE patients (balanced accuracy=0.822). We then compiled CD-associated expression profiles using regularized logistic regression and pathway enrichment analyses. On an independent general cell line model system data, we replicated associations (in silico) of BCL7A (padj=1.69e-9) and STRBP(padj=4.63e-8) with CD22; NCOA2(padj=7.00e-4), ATN1 (padj=1.71e-2), and HOXC4(padj=3.34e-2) with CD30; and PHOSPHO1, a phosphatase linked to bone mineralization, with both CD22(padj=4.37e-2) and CD30(padj=7.40e-3). Utilizing carefully aggregated secondary data and leveraging a priori hypotheses, i-mAB fostered robust biomarker profiling among interdependent biological features.

A novel multi-network approach reveals tissue-specific cellular modulators of fibrosis in systemic sclerosis.

BACKGROUND: Systemic sclerosis (SSc) is a multi-organ autoimmune disease characterized by skin fibrosis. Internal organ involvement is heterogeneous. It is unknown whether disease mechanisms are common across all involved affected tissues or if each manifestation has a distinct underlying pathology. METHODS: We used consensus clustering to compare gene expression profiles of biopsies from four SSc-affected tissues (skin, lung, esophagus, and peripheral blood) from patients with SSc, and the related conditions pulmonary fibrosis (PF) and pulmonary arterial hypertension, and derived a consensus disease-associate signature across all tissues. We used this signature to query tissue-specific functional genomic networks. We performed novel network analyses to contrast the skin and lung microenvironments and to assess the functional role of the inflammatory and fibrotic genes in each organ. Lastly, we tested the expression of macrophage activation state-associated gene sets for enrichment in skin and lung using a Wilcoxon rank sum test. RESULTS: We identified a common pathogenic gene expression signature-an immune-fibrotic axis-indicative of pro-fibrotic macrophages (MØs) in multiple tissues (skin, lung, esophagus, and peripheral blood mononuclear cells) affected by SSc. While the co-expression of these genes is common to all tissues, the functional consequences of this upregulation differ by organ. We used this disease-associated signature to query tissue-specific functional genomic networks to identify common and tissue-specific pathologies of SSc and related conditions. In contrast to skin, in the lung-specific functional network we identify a distinct lung-resident MØ signature associated with lipid stimulation and alternative activation. In keeping with our network results, we find distinct MØ alternative activation transcriptional programs in SSc-associated PF lung and in the skin of patients with an "inflammatory" SSc gene expression signature. CONCLUSIONS: Our results suggest that the innate immune system is central to SSc disease processes but that subtle distinctions exist between tissues. Our approach provides a framework for examining molecular signatures of disease in fibrosis and autoimmune diseases and for leveraging publicly available data to understand common and tissue-specific disease processes in complex human diseases.

A Functional Genomic Meta-Analysis of Clinical Trials in Systemic Sclerosis: Toward Precision Medicine and Combination Therapy.

Systemic sclerosis is an orphan, systemic autoimmune disease with no FDA-approved treatments. Its heterogeneity and rarity often result in underpowered clinical trials making the analysis and interpretation of associated molecular data challenging. We performed a meta-analysis of gene expression data from skin biopsies of patients with systemic sclerosis treated with five therapies: mycophenolate mofetil, rituximab, abatacept, nilotinib, and fresolimumab. A common clinical improvement criterion of -20% or -5 modified Rodnan skin score was applied to each study. We applied a machine learning approach that captured features beyond differential expression and was better at identifying targets of therapies than the differential expression alone. Regardless of treatment mechanism, abrogation of inflammatory pathways accompanied clinical improvement in multiple studies suggesting that high expression of immune-related genes indicates active and targetable disease. Our framework allowed us to compare different trials and ask if patients who failed one therapy would likely improve on a different therapy, based on changes in gene expression. Genes with high expression at baseline in fresolimumab nonimprovers were downregulated in mycophenolate mofetil improvers, suggesting that immunomodulatory or combination therapy may have benefitted these patients. This approach can be broadly applied to increase tissue specificity and sensitivity of differential expression results.

Stress granules and RNA processing bodies are novel autoantibody targets in systemic sclerosis.

BACKGROUND: Autoantibody profiles represent important patient stratification markers in systemic sclerosis (SSc). Here, we performed serum-immunoprecipitations with patient antibodies followed by mass spectrometry (LC-MS/MS) to obtain an unbiased view of all possible autoantibody targets and their associated molecular complexes recognized by SSc. METHODS: HeLa whole cell lysates were immunoprecipitated (IP) using sera of patients with SSc clinically positive for autoantibodies against RNA polymerase III (RNAP3), topoisomerase 1 (TOP1), and centromere proteins (CENP). IP eluates were then analyzed by LC-MS/MS to identify novel proteins and complexes targeted in SSc. Target proteins were examined using a functional interaction network to identify major macromolecular complexes, with direct targets validated by IP-Western blots and immunofluorescence. RESULTS: A wide range of peptides were detected across patients in each clinical autoantibody group. Each group contained peptides representing a broad spectrum of proteins in large macromolecular complexes, with significant overlap between groups. Network analyses revealed significant enrichment for proteins in RNA processing bodies (PB) and cytosolic stress granules (SG) across all SSc subtypes, which were confirmed by both Western blot and immunofluorescence. CONCLUSIONS: While strong reactivity was observed against major SSc autoantigens, such as RNAP3 and TOP1, there was overlap between groups with widespread reactivity seen against multiple proteins. Identification of PB and SG as major targets of the humoral immune response represents a novel SSc autoantigen and suggests a model in which a combination of chronic and acute cellular stresses result in aberrant cell death, leading to autoantibody generation directed against macromolecular nucleic acid-protein complexes.

Molecular characterization of systemic sclerosis esophageal pathology identifies inflammatory and proliferative signatures.

INTRODUCTION: Esophageal involvement in patients with systemic sclerosis (SSc) is common, but tissue-specific pathological mechanisms are poorly understood. There are no animal scleroderma esophagus models and esophageal smooth muscle cells dedifferentiate in culture prohibiting in vitro studies. Esophageal fibrosis is thought to disrupt smooth muscle function and lead to esophageal dilatation, but autopsy studies demonstrate esophageal smooth muscle atrophy and the absence of fibrosis in the majority of SSc cases. Herein, we perform a detailed characterization of SSc esophageal histopathology and molecular signatures at the level of gene expression. METHODS: Esophageal biopsies were prospectively obtained during esophagogastroduodenoscopy in 16 consecutive SSc patients and 7 subjects without SSc. Upper and lower esophageal biopsies were evaluated for histopathology and gene expression. RESULTS: Individual patient's upper and lower esophageal biopsies showed nearly identical patterns of gene expression. Similar to skin, inflammatory and proliferative gene expression signatures were identified suggesting that molecular subsets are a universal feature of SSc end-target organ pathology. The inflammatory signature was present in biopsies without high numbers of infiltrating lymphocytes. Molecular classification of esophageal biopsies was independent of SSc skin subtype, serum autoantibodies and esophagitis. CONCLUSIONS: Proliferative and inflammatory molecular gene expression subsets in tissues from patients with SSc may be a conserved, reproducible component of SSc pathogenesis. The inflammatory signature is observed in biopsies that lack large inflammatory infiltrates suggesting that immune activation is a major driver of SSc esophageal pathogenesis.

Roles of mouse sperm-associated alpha-L-fucosidases in fertilization.

Sperm-associated α-L-fucosidases have been implicated in fertilization in many species. Previously, we documented the existence of α-L-fucosidase in mouse cauda epididymal contents, and showed that sperm-associated α-L-fucosidase is cryptically stored within the acrosome and reappears within the sperm equatorial segment after the acrosome reaction. The enrichment of sperm membrane-associated α-L-fucosidase within the equatorial segment of acrosome-reacted cells implicates its roles during fertilization. Here, we document the absence of α-L-fucosidase in mouse oocytes and early embryos, and define roles of sperm associated α-L-fucosidase in fertilization using specific inhibitors and competitors. Mouse sperm were pretreated with deoxyfuconojirimycin (DFJ, an inhibitor of α-L-fucosidase) or with anti-fucosidase antibody; alternatively, mouse oocytes were pretreated with purified human liver α-L-fucosidase. Five-millimolar DFJ did not inhibit sperm-zona pellucida (ZP) binding, membrane binding, or fusion and penetration, but anti-fucosidase antibody and purified human liver α-L-fucosidase significantly decreased the frequency of these events. To evaluate sperm-associated α-L-fucosidase enzyme activity in post-fusion events, DFJ-pretreated sperm were microinjected into oocytes, and 2-pronuclear (2-PN) embryos were treated with 5 mM DFJ with no significant effects, suggesting that α-L-fucosidase enzyme activity does not play a role in post-fusion events and/or early embryo development in mice. The recognition and binding of mouse sperm to the ZP and oolemma involves the glycoprotein structure of α-L-fucosidase, but not its catalytic action. These observations suggest that deficits in fucosidase protein and/or the presence of anti-fucosidase antibody may be responsible for some types of infertility.