BioAutoMATED: An end-to-end automated machine learning tool for explanation and design of biological sequences.

The design choices underlying machine-learning (ML) models present important barriers to entry for many biologists who aim to incorporate ML in their research. Automated machine-learning (AutoML) algorithms can address many challenges that come with applying ML to the life sciences. However, these algorithms are rarely used in systems and synthetic biology studies because they typically do not explicitly handle biological sequences (e.g., nucleotide, amino acid, or glycan sequences) and cannot be easily compared with other AutoML algorithms. Here, we present BioAutoMATED, an AutoML platform for biological sequence analysis that integrates multiple AutoML methods into a unified framework. Users are automatically provided with relevant techniques for analyzing, interpreting, and designing biological sequences. BioAutoMATED predicts gene regulation, peptide-drug interactions, and glycan annotation, and designs optimized synthetic biology components, revealing salient sequence characteristics. By automating sequence modeling, BioAutoMATED allows life scientists to incorporate ML more readily into their work.

Target-agnostic drug prediction integrated with medical record analysis uncovers differential associations of statins with increased survival in COVID-19 patients.

Drug repurposing requires distinguishing established drug class targets from novel molecule-specific mechanisms and rapidly derisking their therapeutic potential in a time-critical manner, particularly in a pandemic scenario. In response to the challenge to rapidly identify treatment options for COVID-19, several studies reported that statins, as a drug class, reduce mortality in these patients. However, it is unknown if different statins exhibit consistent function or may have varying therapeutic benefit. A Bayesian network tool was used to predict drugs that shift the host transcriptomic response to SARS-CoV-2 infection towards a healthy state. Drugs were predicted using 14 RNA-sequencing datasets from 72 autopsy tissues and 465 COVID-19 patient samples or from cultured human cells and organoids infected with SARS-CoV-2. Top drug predictions included statins, which were then assessed using electronic medical records containing over 4,000 COVID-19 patients on statins to determine mortality risk in patients prescribed specific statins versus untreated matched controls. The same drugs were tested in Vero E6 cells infected with SARS-CoV-2 and human endothelial cells infected with a related OC43 coronavirus. Simvastatin was among the most highly predicted compounds (14/14 datasets) and five other statins, including atorvastatin, were predicted to be active in > 50% of analyses. Analysis of the clinical database revealed that reduced mortality risk was only observed in COVID-19 patients prescribed a subset of statins, including simvastatin and atorvastatin. In vitro testing of SARS-CoV-2 infected cells revealed simvastatin to be a potent direct inhibitor whereas most other statins were less effective. Simvastatin also inhibited OC43 infection and reduced cytokine production in endothelial cells. Statins may differ in their ability to sustain the lives of COVID-19 patients despite having a shared drug target and lipid-modifying mechanism of action. These findings highlight the value of target-agnostic drug prediction coupled with patient databases to identify and clinically evaluate non-obvious mechanisms and derisk and accelerate drug repurposing opportunities.

SOPHIE: Generative Neural Networks Separate Common and Specific Transcriptional Responses.

Genome-wide transcriptome profiling identifies genes that are prone to differential expression (DE) across contexts, as well as genes with changes specific to the experimental manipulation. Distinguishing genes that are specifically changed in a context of interest from common differentially expressed genes (DEGs) allows more efficient prediction of which genes are specific to a given biological process under scrutiny. Currently, common DEGs or pathways can only be identified through the laborious manual curation of experiments, an inordinately time-consuming endeavor. Here we pioneer an approach, Specific cOntext Pattern Highlighting In Expression data (SOPHIE), for distinguishing between common and specific transcriptional patterns using a generative neural network to create a background set of experiments from which a null distribution of gene and pathway changes can be generated. We apply SOPHIE to diverse datasets including those from human, human cancer, and bacterial pathogen Pseudomonas aeruginosa. SOPHIE identifies common DEGs in concordance with previously described, manually and systematically determined common DEGs. Further molecular validation indicates that SOPHIE detects highly specific but low-magnitude biologically relevant transcriptional changes. SOPHIE's measure of specificity can complement log2 fold change values generated from traditional DE analyses. For example, by filtering the set of DEGs, one can identify genes that are specifically relevant to the experimental condition of interest. Consequently, these results can inform future research directions. All scripts used in these analyses are available at https://github.com/greenelab/generic-expression-patterns. Users can access https://github.com/greenelab/sophie to run SOPHIE on their own data.

Nutritional deficiency in an intestine-on-a-chip recapitulates injury hallmarks associated with environmental enteric dysfunction.

Environmental enteric dysfunction (EED)-a chronic inflammatory condition of the intestine-is characterized by villus blunting, compromised intestinal barrier function and reduced nutrient absorption. Here we show that essential genotypic and phenotypic features of EED-associated intestinal injury can be reconstituted in a human intestine-on-a-chip lined by organoid-derived intestinal epithelial cells from patients with EED and cultured in nutrient-deficient medium lacking niacinamide and tryptophan. Exposure of the organ chip to such nutritional deficiencies resulted in congruent changes in six of the top ten upregulated genes that were comparable to changes seen in samples from patients with EED. Chips lined with healthy epithelium or with EED epithelium exposed to nutritional deficiencies resulted in severe villus blunting and barrier dysfunction, and in the impairment of fatty acid uptake and amino acid transport; and the chips with EED epithelium exhibited heightened secretion of inflammatory cytokines. The organ-chip model of EED-associated intestinal injury may facilitate the analysis of the molecular, genetic and nutritional bases of the disease and the testing of candidate therapeutics for it.

Target-agnostic drug prediction integrated with medical record analysis uncovers differential associations of statins with increased survival in COVID-19 patients.

Importance: Drug repurposing requires distinguishing established drug class targets from novel molecule-specific mechanisms and rapidly derisking their therapeutic potential in a time-critical manner, particularly in a pandemic scenario. In response to the challenge to rapidly identify treatment options for COVID-19, several studies reported that statins, as a drug class, reduce mortality in these patients. However, it is unknown if different statins exhibit consistent function or may have varying therapeutic benefit. Objectives: To test if different statins differ in their ability to exert protective effects based on molecular computational predictions and electronic medical record analysis. Main Outcomes and Measures: A Bayesian network tool was used to predict drugs that shift the host transcriptomic response to SARS-CoV-2 infection towards a healthy state. Drugs were predicted using 14 RNA-sequencing datasets from 72 autopsy tissues and 465 COVID-19 patient samples or from cultured human cells and organoids infected with SARS-CoV-2, with a total of 2,436 drugs investigated. Top drug predictions included statins, which were then assessed using electronic medical records containing over 4,000 COVID-19 patients on statins to determine mortality risk in patients prescribed specific statins versus untreated matched controls. The same drugs were tested in Vero E6 cells infected with SARS-CoV-2 and human endothelial cells infected with a related OC43 coronavirus. Results: Simvastatin was among the most highly predicted compounds (14/14 datasets) and five other statins, including atorvastatin, were predicted to be active in > 50% of analyses. Analysis of the clinical database revealed that reduced mortality risk was only observed in COVID-19 patients prescribed a subset of statins, including simvastatin and atorvastatin. In vitro testing of SARS-CoV-2 infected cells revealed simvastatin to be a potent direct inhibitor whereas most other statins were less effective. Simvastatin also inhibited OC43 infection and reduced cytokine production in endothelial cells. Conclusions and Relevance: Different statins may differ in their ability to sustain the lives of COVID-19 patients despite having a shared drug target and lipid-modifying mechanism of action. These findings highlight the value of target-agnostic drug prediction coupled with patient databases to identify and clinically evaluate non-obvious mechanisms and derisk and accelerate drug repurposing opportunities.

Mechanical control of innate immune responses against viral infection revealed in a human lung alveolus chip.

Mechanical breathing motions have a fundamental function in lung development and disease, but little is known about how they contribute to host innate immunity. Here we use a human lung alveolus chip that experiences cyclic breathing-like deformations to investigate whether physical forces influence innate immune responses to viral infection. Influenza H3N2 infection of mechanically active chips induces a cascade of host responses including increased lung permeability, apoptosis, cell regeneration, cytokines production, and recruitment of circulating immune cells. Comparison with static chips reveals that breathing motions suppress viral replication by activating protective innate immune responses in epithelial and endothelial cells, which are mediated in part through activation of the mechanosensitive ion channel TRPV4 and signaling via receptor for advanced glycation end products (RAGE). RAGE inhibitors suppress cytokines induction, while TRPV4 inhibition attenuates both inflammation and viral burden, in infected chips with breathing motions. Therefore, TRPV4 and RAGE may serve as new targets for therapeutic intervention in patients infected with influenza and other potential pandemic viruses that cause life-threatening lung inflammation.

Self-assembling short immunostimulatory duplex RNAs with broad spectrum antiviral activity.

The current COVID-19 pandemic highlights the need for broad-spectrum antiviral therapeutics. Here we describe a new class of self-assembling immunostimulatory short duplex RNAs that potently induce production of type I and type III interferon (IFN-I and IFN-III), in a wide range of human cell types. These RNAs require a minimum of 20 base pairs, lack any sequence or structural characteristics of known immunostimulatory RNAs, and instead require a unique conserved sequence motif (sense strand: 5'-C, antisense strand: 3'-GGG) that mediates end-to-end dimer self-assembly of these RNAs by Hoogsteen G-G base-pairing. The presence of terminal hydroxyl or monophosphate groups, blunt or overhanging ends, or terminal RNA or DNA bases did not affect their ability to induce IFN. Unlike previously described immunostimulatory siRNAs, their activity is independent of TLR7/8, but requires the RIG-I/IRF3 pathway that induces a more restricted antiviral response with a lower proinflammatory signature compared with poly(I:C). Immune stimulation mediated by these duplex RNAs results in broad spectrum inhibition of infections by many respiratory viruses with pandemic potential, including SARS-CoV-2, SARS-CoV, MERS-CoV, and influenza A, as well as the common cold virus HCoV-NL63 in both cell lines and human Lung Chips that mimic organ-level lung pathophysiology. These short dsRNAs can be manufactured easily, and thus potentially could be harnessed to produce broad-spectrum antiviral therapeutics at low cost.

A human-airway-on-a-chip for the rapid identification of candidate antiviral therapeutics and prophylactics.

The rapid repurposing of antivirals is particularly pressing during pandemics. However, rapid assays for assessing candidate drugs typically involve in vitro screens and cell lines that do not recapitulate human physiology at the tissue and organ levels. Here we show that a microfluidic bronchial-airway-on-a-chip lined by highly differentiated human bronchial-airway epithelium and pulmonary endothelium can model viral infection, strain-dependent virulence, cytokine production and the recruitment of circulating immune cells. In airway chips infected with influenza A, the co-administration of nafamostat with oseltamivir doubled the treatment-time window for oseltamivir. In chips infected with pseudotyped severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), clinically relevant doses of the antimalarial drug amodiaquine inhibited infection but clinical doses of hydroxychloroquine and other antiviral drugs that inhibit the entry of pseudotyped SARS-CoV-2 in cell lines under static conditions did not. We also show that amodiaquine showed substantial prophylactic and therapeutic activities in hamsters challenged with native SARS-CoV-2. The human airway-on-a-chip may accelerate the identification of therapeutics and prophylactics with repurposing potential.

Deep-Learning Resources for Studying Glycan-Mediated Host-Microbe Interactions.

Glycans, the most diverse biopolymer, are shaped by evolutionary pressures stemming from host-microbe interactions. Here, we present machine learning and bioinformatics methods to leverage the evolutionary information present in glycans to gain insights into how pathogens and commensals interact with hosts. By using techniques from natural language processing, we develop deep-learning models for glycans that are trained on a curated dataset of 19,299 unique glycans and can be used to study and predict glycan functions. We show that these models can be utilized to predict glycan immunogenicity and the pathogenicity of bacterial strains, as well as investigate glycan-mediated immune evasion via molecular mimicry. We also develop glycan-alignment methods and use these to analyze virulence-determining glycan motifs in the capsular polysaccharides of bacterial pathogens. These resources enable one to identify and study glycan motifs involved in immunogenicity, pathogenicity, molecular mimicry, and immune evasion, expanding our understanding of host-microbe interactions.

Trisomy 21 activates the kynurenine pathway via increased dosage of interferon receptors.

Trisomy 21 (T21) causes Down syndrome (DS), affecting immune and neurological function by ill-defined mechanisms. Here we report a large metabolomics study of plasma and cerebrospinal fluid, showing in independent cohorts that people with DS produce elevated levels of kynurenine and quinolinic acid, two tryptophan catabolites with potent immunosuppressive and neurotoxic properties, respectively. Immune cells of people with DS overexpress IDO1, the rate-limiting enzyme in the kynurenine pathway (KP) and a known interferon (IFN)-stimulated gene. Furthermore, the levels of IFN-inducible cytokines positively correlate with KP dysregulation. Using metabolic tracing assays, we show that overexpression of IFN receptors encoded on chromosome 21 contribute to enhanced IFN stimulation, thereby causing IDO1 overexpression and kynurenine overproduction in cells with T21. Finally, a mouse model of DS carrying triplication of IFN receptors exhibits KP dysregulation. Together, our results reveal a mechanism by which T21 could drive immunosuppression and neurotoxicity in DS.

Sequential specification of oligodendrocyte lineage cells by distinct levels of Hedgehog and Notch signaling.

During development of the central nervous system oligodendrocyte precursor cells (OPCs) give rise to both myelinating oligodendrocytes and NG2 glia, which are the most proliferative cells in the adult mammalian brain. NG2 glia retain characteristics of OPCs, and some NG2 glia produce oligodendrocytes, but many others persist throughout adulthood. Why some OPCs differentiate as oligodendrocytes during development whereas others persist as OPCs and acquire characteristics of NG2 glia is not known. Using zebrafish spinal cord as a model, we found that OPCs that differentiate rapidly as oligodendrocytes and others that remain as OPCs arise in sequential waves from distinct neural progenitors. Additionally, oligodendrocyte and persistent OPC fates are specified during a defined critical period by small differences in Shh signaling and Notch activity, which modulates Shh signaling response. Thus, our data indicate that OPCs fated to produce oligodendrocytes or remain as OPCs during development are specified as distinct cell types, raising the possibility that the myelinating potential of OPCs is set by graded Shh signaling activity.

Metastatic cells are preferentially vulnerable to lysosomal inhibition.

Molecular alterations that confer phenotypic advantages to tumors can also expose specific therapeutic vulnerabilities. To search for potential treatments that would selectively affect metastatic cells, we examined the sensitivity of lineage-related human bladder cancer cell lines with different lung colonization abilities to chloroquine (CQ) or bafilomycin A1, which are inhibitors of lysosome function and autophagy. Both CQ and bafilomycin A1 were more cytotoxic in vitro to highly metastatic cells compared with their less metastatic counterparts. Genetic inactivation of macroautophagy regulators and lysosomal proteins indicated that this was due to greater reliance on the lysosome but not upon macroautophagy. To identify the mechanism underlying these effects, we generated cells resistant to CQ in vitro. Surprisingly, selection for in vitro CQ resistance was sufficient to alter gene expression patterns such that unsupervised cluster analysis of whole-transcriptome data indicated that selection for CQ resistance alone created tumor cells that were more similar to the poorly metastatic parental cells from which the metastatic cells were derived; importantly, these tumor cells also had diminished metastatic ability in vivo. These effects were mediated in part by differential expression of the transcriptional regulator ID4 (inhibitor of DNA binding 4); depletion of ID4 both promoted in vitro CQ sensitivity and restored lung colonization and metastasis of CQ-resistant cells. These data demonstrate that selection for metastasis ability confers selective vulnerability to lysosomal inhibitors and identify ID4 as a potential biomarker for the use of lysosomal inhibitors to reduce metastasis in patients.

GSEA-InContext: identifying novel and common patterns in expression experiments.

Motivation: Gene Set Enrichment Analysis (GSEA) is routinely used to analyze and interpret coordinate pathway-level changes in transcriptomics experiments. For an experiment where less than seven samples per condition are compared, GSEA employs a competitive null hypothesis to test significance. A gene set enrichment score is tested against a null distribution of enrichment scores generated from permuted gene sets, where genes are randomly selected from the input experiment. Looking across a variety of biological conditions, however, genes are not randomly distributed with many showing consistent patterns of up- or down-regulation. As a result, common patterns of positively and negatively enriched gene sets are observed across experiments. Placing a single experiment into the context of a relevant set of background experiments allows us to identify both the common and experiment-specific patterns of gene set enrichment. Results: We compiled a compendium of 442 small molecule transcriptomic experiments and used GSEA to characterize common patterns of positively and negatively enriched gene sets. To identify experiment-specific gene set enrichment, we developed the GSEA-InContext method that accounts for gene expression patterns within a background set of experiments to identify statistically significantly enriched gene sets. We evaluated GSEA-InContext on experiments using small molecules with known targets to show that it successfully prioritizes gene sets that are specific to each experiment, thus providing valuable insights that complement standard GSEA analysis. Availability and implementation: GSEA-InContext implemented in Python, Supplementary results and the background expression compendium are available at: https://github.com/CostelloLab/GSEA-InContext.

Next-Generation Machine Learning for Biological Networks.

Machine learning, a collection of data-analytical techniques aimed at building predictive models from multi-dimensional datasets, is becoming integral to modern biological research. By enabling one to generate models that learn from large datasets and make predictions on likely outcomes, machine learning can be used to study complex cellular systems such as biological networks. Here, we provide a primer on machine learning for life scientists, including an introduction to deep learning. We discuss opportunities and challenges at the intersection of machine learning and network biology, which could impact disease biology, drug discovery, microbiome research, and synthetic biology.

Eya3 promotes breast tumor-associated immune suppression via threonine phosphatase-mediated PD-L1 upregulation.

Eya proteins are critical developmental regulators that are highly expressed in embryogenesis but downregulated after development. Amplification and/or re-expression of Eyas occurs in many tumor types. In breast cancer, Eyas regulate tumor progression by acting as transcriptional cofactors and tyrosine phosphatases. Intriguingly, Eyas harbor a separate threonine (Thr) phosphatase activity, which was previously implicated in innate immunity. Here we describe what we believe to be a novel role for Eya3 in mediating triple-negative breast cancer-associated immune suppression. Eya3 loss decreases tumor growth in immune-competent mice and is associated with increased numbers of infiltrated CD8+ T cells, which, when depleted, reverse the effects of Eya3 knockdown. Mechanistically, Eya3 utilizes its Thr phosphatase activity to dephosphorylate Myc at pT58, resulting in a stabilized form. We show that Myc is required for Eya3-mediated increases in PD-L1, and that rescue of PD-L1 in Eya3-knockdown cells restores tumor progression. Finally, we demonstrate that Eya3 significantly correlates with PD-L1 in human breast tumors, and that tumors expressing high levels of Eya3 have a decreased CD8+ T cell signature. Our data uncover a role for Eya3 in mediating tumor-associated immune suppression, and suggest that its inhibition may enhance checkpoint therapies.