bioRxiv Science⌕ Search

Biology subjects

Cheng, W.

Publications and source records attributed to Cheng, W..

At least 55 records · Page 3Linked to original sources

Multi-omics and Pharmacological Characterization of Patient-derived Glioma Cell Lines

Glioblastoma (GBM) is the most common brain tumor and is currently incurable. Primary GBM cultures are widely used tools for screening potentially therapeutic drugs; however, there is a lack of genomic and pharmacological characterization of these primary GBM cultures. Here, we collected 52 patient-derived glioma cell (PDGC) lines and characterized them through whole- genome sequencing (WGS), RNA-seq, and drug response screening. We identified three molecular subtypes among PDGCs: mesenchymal (MES), proneural (PN), and oxidative phosphorylation (OXPHOS). Upon profiling the responses of PDGCs to 214 drugs, we found that the PN subtype PDGCs were sensitive to tyrosine kinase inhibitors, whereas the OXPHOS subtype PDGCs were sensitive to histone deacetylase inhibitors, oxidative phosphorylation inhibitors, and HMG-CoA reductase inhibitors. PN and OXPHOS subtype PDGCs stably formed tumors in vivo upon intracranial transplantation into immunodeficient mice, while most MES subtype PDGCs were incapable of tumorigenesis in vivo. In addition, profiling and follow-up investigations showed that the serum-free culture system used for PDGCs enriched and propagated rare MYC/MYCN- amplified glioma cells. Our study provides a resource for understanding primary glioma cell cultures and aiding clinical translation. SignificanceOur study provides a resource for patient-derived glioma cell lines (PDGCs) on transcriptome, genome, drug response, and tumorigenic abilities. PDGCs are categorized into PN, MES, and OXPHOS subtypes, with MES-subtype PDGCs incapable of tumorigenesis in vivo. Notably, the serum-free culture system for PDGCs enriches glioma cells with MYC/MYCN amplification.

cancer biology↗

Molecular basis for potent B cell responses to antigen displayed on particles of viral size

Although it has long been appreciated that multivalent antigens - and particularly viral epitope display - produce extremely rapid, robust, and T-independent humoral immune responses, the biochemical basis for such potency has been incompletely understood. Here we take advantage of a set of neutral liposomes of viral size that are engineered to display affinity mutants of the model antigen (Ag) hen egg lysozyme at precisely varied density. We show that particulate Ag display by liposomes induces highly potent B cell responses that are dose-and density-dependent but affinity-independent. Titrating dose of particulate, but not soluble, Ag reveals bimodal Erk phosphorylation and cytosolic calcium increases. Particulate Ag induces signal amplification downstream of the B cell receptor (BCR) by selectively evading LYN-dependent inhibitory pathways, but in vitro potency is independent of CD19. Importantly, Ag display on viral-sized particles signals independently of MYD88 and IRAK1/4, but activates NF-{kappa}B robustly in a manner that mimics T cell help. Together, such biased signaling by particulate Ag promotes MYC expression and reduces the threshold required for B cell proliferation relative to soluble Ag. These findings uncover a molecular basis for highly sensitive B cell response to viral Ag display and remarkable potency of virus-like particle vaccines that is not merely accounted for by avidity and BCR cross-linking, and is independent of the contribution of B cell nucleic acid-sensing machinery.

immunology↗

Quantitative proteomics combined with network pharmacology analysis unveils the biological basis of Schisandrin B in treating diabetic nephropathy

BackgroundDiabetic nephropathy (DN) is a major complication of diabetes. Schisandrin B (Sch) is a natural pharmaceutical monomer that was shown to prevent kidney damage caused by diabetes and restore its function. However, there is still a lack of comprehensive and systematic understanding of the mechanism of Sch treatment in DN. ObjectiveWe aim to provide a systematic overview of the mechanisms of Sch in multiple pathways to treat DN in rats. MethodsStreptozocin was used to build a DN rat model, which was further treated with Sch. The possible mechanism of Sch protective effects against DN was predicted using network pharmacology and was verified by quantitative proteomics analysis. ResultsHigh dose Sch treatment significantly downregulated fasting blood glucose, creatinine, blood urea nitrogen, and urinary protein levels and reduced collagen deposition in the glomeruli and tubule-interstitium of DN rats. The activities of superoxide dismutase (SOD) and plasma glutathione peroxidase (GSH-Px) in the kidney of DN rats significantly increased with Sch treatment. In addition, the levels of IL-6, IL-1{beta}, and TNF- were significantly reduced in DN rats treated with Sch. 11 proteins that target both Sch and DN were enriched in pathways such as MAPK signaling, PI3K-Akt signaling, renal cell carcinoma, gap junction, endocrine resistance, and TNF signaling. Furthermore, quantitative proteomics showed that Xaf1 was downregulated in the model vs. control group and upregulated in the Sch-treated vs. model group. Five proteins, Crb3, Tspan4, Wdr45, Zfp512, and Tmigd1, were found to be upregulated in the model vs. control group and downregulated in the Sch vs. model group. Three intersected proteins between the network pharmacology prediction and proteomics results, Crb3, Xaf1, and Tspan4, were identified. ConclusionSch functions by relieving oxidative stress and the inflammatory response by regulating Crb3, Xaf1, and Tspan4 protein expression levels to treat DN disease.

pharmacology and toxicology↗

Improving protein secondary structure prediction by deep language models and transformer networks

Protein secondary structure prediction is useful for many applications. It can be considered a language translation problem, i.e., translating a sequence of 20 different amino acids into a sequence of secondary structure symbols (e.g., alpha helix, beta strand, and coil). Here, we develop a novel protein secondary structure predictor called TransPross based on the transformer network and attention mechanism widely used in natural language processing to directly extract the evolutionary information from the protein language (i.e., raw multiple sequence alignment (MSA) of a protein) to predict the secondary structure. The method is different from traditional methods that first generate a MSA and then calculate expert-curated statistical profiles from the MSA as input. The attention mechnism used by TransPross can effectively capture long-range residue-residue interactions in protein sequences to predict secondary structures. Benchmarked on several datasets, TransPross outperforms the state-of-art methods. Moreover, our experiment shows that the prediction accuracy of TransPross positively correlates with the depth of MSAs and it is able to achieve the average prediction accuracy (i.e., Q3 score) above 80% for hard targets with few homologous sequences in their MSAs. TransPross is freely available at https://github.com/BioinfoMachineLearning/TransPro

bioinformatics↗

Predicting Pacific cod spawning habitat in a changing climate

Warming temperatures elicit shifts in habitat use and geographic distributions of fishes, with uneven effects across life stages. Spawners and embryos are particularly sensitive to environmental conditions, with direct impacts of temperature on spawning habitat, as well as indirect connections between their population dynamics and fisheries effort, productivity, and management. Here, we ask how changing environmental conditions and thermal sensitivities of developing embryos confer spatiotemporal variability of thermally-suitable spawning habitat for Pacific cod in the eastern Bering Sea. Specifically, we use bottom temperature values from regionally downscaled global climate models coupled with an experimentally-derived relationship between hatch success and temperature to predict how the extent, mean latitude, and consistency of suitable spawning habitat has changed in the past and may change into the future. We then validate our predictions of suitable spawning habitat with distributions of adults and larvae and examine whether thermal habitat availability relates to recruitment success into the adult cod into the population. We find that the extent and mean latitude of suitable spawning habitat increase over time, particularly if no climate change mitigation occurs in the future. Hotspots of suitable spawning habitat are consistent across shorter time periods but do shift across the Bering Sea shelf by the end of the century. Finally, we find no correlation between the availability of suitable spawning habitat and annual estimates of recruitment. Collectively, our results suggest that as temperatures warm, the availability of suitable spawning habitat will increase and expand spatially and, thus, is not likely to limit recruitment. This work highlights the importance of coupling experimental data with climate models to identify the complex and mechanistic dynamics among temperature, life histories, and ecology, and offers a pathway for examining life stage-specific changes in habitat use and distribution with continued climate change.

ecology↗

Efficient meta-analysis of multivariate genome-wide association studies with Meta-MOSTest

MotivationGenome-wide association studies (GWAS) have been successful in identifying genetic variants associated with a particular phenotype. However, many complex phenotypes are influenced by multiple genetic variants with small effects. Detecting the genetic pleiotropy can provide insights into biological mechanisms influencing complex human phenotypes. The recently developed Multivariate Omnibus Statistical Test (MOSTest) has proven to be efficient and powerful, suited for complex large-scale data. The method substantially increased discovery of genetic variants associated with brain MRI phenotypes in the UK Biobank compared to conventionally use multivariate approach. Here we extend the MOSTest to meta-analysis (Meta-MOSTest), facilitating data analysis of multiple phenotypes across multiple cohorts. We evaluated our updated approach in the UK Biobank using brain MRI phenotypes, by comparing the discovery yield of the single-cohort MOSTest versus Meta-MOSTest through simulating sub-cohorts of different sample sizes from 265 to 26501 subjects. ResultsOur method works efficiently on large-scale cohorts with a large number of MRI phenotypes. We found that lower per-cohort sample sizes resulted in a reduced discovery yield indicating a loss of statistical power. However, with a minimum sample size of 250 subjects across cohorts, Meta-MOSTest was equivalent to MOSTest on discovery yield while maintaining a well-calibrated type I error and equivalent statistical power. We conclude that Meta-MOSTest is a useful tool for multivariate analysis across separate brain imaging genetics cohorts. Availability and implementationAll codes are freely available on GitHub: MOSTest and Meta-MOSTest.

bioinformatics↗

Hiplot: a comprehensive and easy-to-use web service boosting publication-ready biomedical data visualization

Modern web techniques provide an unprecedented opportunity for leveraging complex biomedical data generating in clinical, omics, and mechanism experiments. Currently, the functions for carrying out publication-ready biomedical data visualization represent primary technical hurdles in the state-of-art omics-based web services, whereas the demand for visualization-based interactive data mining is ever-growing. Here, we propose an easy-to-use web service, Hiplot (https://hiplot.com.cn), equipping with comprehensive and interactive biomedical data visualization functions (230+) including basic statistics, multi-omics, regression, clustering, dimensional reduction, meta-analysis, survival analysis, risk modeling, etc. We used the demo and real datasets to demonstrate the usage workflow and the core functions of Hiplot. It permits users to conveniently and interactively complete a few specialized visualization tasks that previously could only be done by senior bioinformatics or biostatistics researchers. A modern web client with efficient user interfaces and interaction methods has been implemented based on the custom components library and the extensible plugin system. The versatile output can also be produced in different environments via using the cross-platform portable command-line interface (CLI) program, Hctl. A switchable view between the editable data table and the file uploader/path selection could facilitate data importing, previewing, and exporting, while the plumber-based response strategy significantly reduced the time costs for generating basic scientific graphics. Diversified layouts, themes/styles, and color palettes in this website allow users to create high-quality and publication-ready graphics. Researchers devoted to both life and data science may benefit from the emerging web service.

bioinformatics↗

Multivariate genetic analysis of personality and cognitive traits reveals abundant pleiotropy and improves prediction

Personality and cognition are heritable mental traits, and their genetic determinants may be distributed across interconnected brain functions. However, previous studies have employed univariate approaches which reduce complex traits to summary measures. We applied the "pleiotropy-informed" multivariate omnibus statistical test (MOSTest) to genome-wide association studies (GWAS) of 35 item and task-level measures of neuroticism and cognition from the UK Biobank (n=336,993). We identified 431 significant genetic loci and found evidence of abundant pleiotropy across personality and cognitive domains. Functional characterisation implicated genes with significant tissue-specific expression in all tested brain tissues and enriched in brain-specific gene-sets. We conditioned independent GWAS of the Big 5 personality traits and cognition on our multivariate findings, which boosted genetic discovery in other personality traits and improved polygenic prediction. These findings advance our understanding of the polygenic architecture of complex mental traits, indicating a prominence of pleiotropic genetic effects across higher-order domains of mental function. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=200 SRC="FIGDIR/small/481967v2_ufig1.gif" ALT="Figure 1"> View larger version (89K): org.highwire.dtl.DTLVardef@1641108org.highwire.dtl.DTLVardef@9a24eeorg.highwire.dtl.DTLVardef@1eeae99org.highwire.dtl.DTLVardef@108db7f_HPS_FORMAT_FIGEXP M_FIG C_FIG

genetics↗

Uncertainty Quantification in Variable Selection for Genetic Fine-Mapping using Bayesian Neural Networks

In this paper, we propose a new approach for variable selection using a collection of Bayesian neural networks with a focus on quantifying uncertainty over which variables are selected. Motivated by fine-mapping applications in statistical genetics, we refer to our framework as an "ensemble of single-effect neural networks" (ESNN) which generalizes the "sum of single-effects" regression framework by both accounting for nonlinear structure in genotypic data (e.g., dominance effects) and having the capability to model discrete phenotypes (e.g., case-control studies). Through extensive simulations, we demonstrate our methods ability to produce calibrated posterior summaries such as credible sets and posterior inclusion probabilities, particularly for traits with genetic architectures that have significant proportions of non-additive variation driven by correlated variants. Lastly, we use real data to demonstrate that the ESNN framework improves upon the state-of-the-art for identifying true effect variables underlying various complex traits.

bioinformatics↗

CyanoHABs: inevitable evolutionary ecological consequence for low nutrient-requiring cyanobacteria in water eutrophication

Cyanobacterial harmful algal blooms (CyanoHABs) arise as cyanobacteria dominate phytoplankton communities when nutrient levels increase from oligotrophic state. From a wholistic perspective, this longstanding altered phytoplankton structure results from two conditions: one sufficient condition that cyanobacteria can grow maximally with elevated nutrients; one necessary condition that co-living algae cannot grow fast or dominate at the same levels. The sufficient condition, the special mechanism of CyanoHABs at the population level, has been established as the synergistic interaction between superior cyanobacterial ecophysiology and elevated nutrients. But it is unknown how these functions arise or whether they are under directed evolution to water eutrophication. The necessary condition, the general mechanism of CyanoHABs at the community level, is little understood: why co-living algae cannot form blooms as cyanobacteria? Literature and bioinformatics analyses show that the superior ecophysiology undergoes no directed positive evolution to worldwide eutrophication in general or any local eutrophic waters in particular; instead, these functions are under strong purifying selection and likely acquired through early adaptive radiation in nutrient-deficient conditions, as functions enabling extant cyanobacteria to occupy other niches. The general mechanism turns out to be quite straightforward: cyanobacteria are simple life forms and thus have lower per capita nutrient demand for growth, compared to co-existing eukaryotic algae in cell size and structure, genome size, size of genome-scale metabolic networks, cell content, nutrient requirement. Lower nutrient demand is proved by existing field nutrient supplementation. Both the special and general mechanisms of CyanoHABs are tentative frameworks awaiting further theoretic improvement and empirical assessment.

evolutionary biology↗

A novel analytical decoder of BOLD signals for dissociating latent neurobehavioral processes

How to retrieve latent neurobehavioral processes from complex neurobiological signals is an important yet unresolved challenge. Here, we develop a novel approach, orthogonal-Decoding multi-Cognitive Processes (DeCoP), to reveal underlying latent neurobehavioral processing and show that its performance is superior to traditional non-orthogonal decoding in terms of both false inference and robustness. Processing value and salience information are two fundamental but mutually confounded pathways of reward reinforcement essential for decision making. During reward/punishment anticipation, we applied DeCoP to decode brain-wide responses into spatially overlapping, yet functionally independent, evaluation and readiness processes, which are modulated differentially by meso-limbic vs nigro-striatal dopamine systems. Using DeCoP, we further demonstrated that most brain regions only encoded abstract information but not the exact input, except for dACC and insula. Furthermore, our novel analytical principle could be applied generally to decode multiple latent neurobehavioral processes and thus advance both the design and hypothesis testing for cognitive tasks.

neuroscience↗

A flexible summary-based colocalization method with application to the mucin Cystic Fibrosis lung disease modifier locus

Mucus obstruction is a central feature in the Cystic Fibrosis (CF) airways. A genome-wide association study (GWAS) of lung disease by the CF Gene Modifier Consortium (CFGMC) identified a significant locus containing two mucin genes, MUC20 and MUC4. Expression quantitative trait locus (eQTL) analysis using human nasal epithelial (HNE) from 94 CF Canadians in the CFGMC demonstrated MUC4 eQTLs that mirrored the lung association pattern in the region, suggesting that MUC4 expression may mediate CF lung disease. Complications arose, however, with colocalization testing using existing methods: the locus is complex and the associated SNPs span a 0.2Mb region with high linkage disequilibrium and evidence of eQTLs for multiple genes and tissues (heterogeneity). We previously developed the Simple Sum (SS), a powerful colocalization test in regions with heterogeneity, but SS assumed eQTLs to be present to achieve type I error control. Here we propose a two-stage SS (SS2) colocalization test that avoids a prior eQTL assumptions, accounts for multiple hypothesis testing and the composite null hypothesis and enables meta-analysis. We compare SS2 to published approaches through simulation and demonstrate type I error control for all settings with the greatest power in the presence of high LD and heterogeneity. Applying SS2 to the MUC20/MUC4 CF lung disease locus with eQTLs from CF HNE revealed significant colocalization with MUC4 (p = 1.71x10-5) rather than MUC20. The SS2 is a powerful method to inform the responsible gene(s) at a locus and guide future functional studies. SS2 has been implemented in the application LocusFocus (locusfocus.research.sickkids.ca).

genetics↗

Persistent DNA damage promotes microglial dysfunction in Ataxia-telangiectasia

The autosomal recessive genome instability disorder Ataxia-telangiectasia, caused by mutations in ATM kinase, is characterised by the progressive loss of cerebellar neurons. We find that DNA damage associated with ATM loss results in dysfunctional behaviour of human microglia, immune cells of the central nervous system. Microglial dysfunction is mediated by the pro-inflammatory RELB/p52 non-canonical NF-{kappa}B transcriptional pathway and leads to excessive phagocytic clearance of neurites. Pathological phagocytosis of neuronal processes by microglia has also been observed in multiple sclerosis, Alzheimers and progranulin deficiency, suggesting a common mechanism that promotes neuronal damage. Activation of the RELB/p52 pathway in ATM-deficient microglia is driven by persistent DNA damage and is dependent on the NIK kinase. These results provide insights into the underlying mechanisms of aberrant microglial behaviour in Ataxia-telangiectasia, potentially contributing to neurodegeneration.

cell biology↗

Multi-omics consensus ensemble refines the classification of muscle-invasive bladder cancer with stratified prognosis, tumour microenvironment and distinct sensitivity to frontline therapies

The molecular classification of muscle-invasive bladder cancer (MIBC) based on transcriptomic signatures has been extensively studied. The complementary nature of information provided by different molecular profiles motivated us to refine MIBC classification by aggregating multi-omics data. We generated a consensus ensemble through ten multi-omics integrative clustering approaches on 396 MIBCs from TCGA. A total of 701 MIBCs from different sequencing technologies were used for external validation. Associations between subtypes and prognosis, molecular profiles, the tumour microenvironment, and potential response to frontline therapies were further analyzed. Nearest template prediction and random forest classification were used to develop a predictive signature/classifier for MIBC refinement. We identified four integrative consensus subtypes of MIBC, which were further designated basal-inflamed, basal-noninflamed, luminal-excluded and luminal-desert by immune profiling. Of note, the refinement of basal-like MIBC classification adds to the literature by identifying a basal-noninflamed MIBC subtype presenting with a significantly poor outcome and a global immune-cold phenotype, which might be triggered by Chr4 deletion and high activation of the oncogenic NRF2 pathway. In contrast, basal-inflamed MIBC showed high immunocyte infiltration and high expression of potential targets for immunotherapy. Using an external metastatic MIBC cohort in which patients received anti-PD-L1 treatment, we suggested that basal-inflamed MIBC had a higher likelihood of responding to immunotherapy than other MIBCs. The R package "refineMIBC" was offered as a research tool to refine MIBC from a single-sample perspective (https://github.com/xlucpu/refineMIBC). This consensus ensemble refines the intrinsic MIBC subtypes, which provides a blueprint for the clinical development of rational targeted and immunotherapeutic strategies.

cancer biology↗

Redefining replication in multi-ancestry genome-wide association studies

Since 2005, genome-wide association (GWA) datasets have been largely biased toward sampling European ancestry individuals, and recent studies have shown that GWA results estimated from self-identified European individuals are not transferable to non-European individuals due to various confounding challenges. Here, we demonstrate that enrichment analyses which aggregate SNP-level association statistics at multiple genomic scales--from genes to genomic regions and pathways--have been underutilized in the GWA era and can generate biologically interpretable hypotheses regarding the genetic basis of complex trait architecture. We illustrate examples of the robust associations generated by enrichment analyses while studying 25 continuous traits assayed in 566,786 individuals from seven diverse self-identified human ancestries in the UK Biobank and the Biobank Japan, as well as 44,348 admixed individuals from the PAGE consortium including cohorts of African-American, Hispanic and Latin American, Native Hawaiian, and American Indian/Alaska Native individuals. We identify 1,000 gene-level associations that are genome-wide significant in at least two ancestry cohorts across these 25 traits, as well as highly conserved pathway associations with triglyceride levels in European, East Asian, and Native Hawaiian cohorts.

genomics↗

Multivariate genome-wide association study identifies 1735 unique genetic loci associated with cortical morphology.

Brain morphology has been shown to be highly heritable, yet only a small portion of the heritability is explained by the genetic variants discovered so far. Here we exploit the distributed nature of genetic effects across the brain and apply the Multivariate Omnibus Statistical Test (MOSTest) to genome-wide association studies (GWAS) of vertex-wise structural magnetic resonance imaging (MRI) cortical measures from N=35,657 participants in the UK Biobank. We identified 695 loci for cortical surface area and 539 for cortical thickness, in total 780 unique genetic loci associated with cortical morphology. This reflects an approximate 10-fold increase compared to the commonly applied univariate GWAS methods. Power analysis indicates that applying MOSTest to vertex-wise structural MRI data triples the effective sample size compared to conventional univariate GWAS approaches. Functional follow up including gene-based analyses implicate 10% of all protein-coding genes and point towards pathways involved in neurogenesis and cell differentiation.

genetics↗

Assessing Study Reproducibility through MMRI: A Novel Approach for Large-scale High-throughput Association Studies

Magnetic Resonance Imaging (MRI) technology has been increasingly used in large-scale association studies. Reproducibility of statistically significant findings generated by MRI-based association studies, especially structural MRI (sMRI) and functional MRI (fMRI), has been recently heavily debated. However, there is still a lack of overall reproducibility assessment for MRI-based association studies. It is also crucial to elucidate the relationship between overall reproducibility and sample size in an experimental design. In this study, we proposed an overall reproducibility index for large-scale high-throughput MRI-based association studies. We performed the overall reproducibility assessments for several recent large sMRI/fMRI databases and observed satisfactory overall reproducibility. Furthermore, we performed the sample size evaluation for the purpose of achieving a desirable overall reproducibility. Additionally, we evaluated the overall reproducibility of GMV changes for UKB vs. PPMI and UKB vs. HCP. We demonstrated that both sample size and some experimental factors play important roles in the overall reproducibility for different experiments. In summary, a systematic assessment of overall reproducibility is fundamental and crucial in the current large-scale high-throughput MRI-based research.

neuroscience↗

Multi-scale Genomic Inference using Biologically Annotated Neural Networks

In this article, we present Biologically Annotated Neural Networks (BANNs), a nonlinear probabilistic framework for association mapping in genome-wide association (GWA) studies. BANNs are feedforward models with partially connected architectures that are based on biological annotations. This setup yields a fully interpretable neural network where the input layer encodes SNP-level effects, and the hidden layer models the aggregated effects among SNP-sets. We treat the weights and connections of the network as random variables with prior distributions that reflect how genetic effects manifest at different genomic scales. The BANNs software uses variational inference to provide posterior summaries which allow researchers to simultaneously perform (i) mapping with SNPs and (ii) enrichment analyses with SNP-sets on complex traits. Through simulations, we show that our method improves upon state-of-the-art association mapping and enrichment approaches across a wide range of genetic architectures. We then further illustrate the benefits of BANNs by analyzing real GWA data assayed in approximately 2,000 heterogenous stock of mice from the Wellcome Trust Centre for Human Genetics and approximately 7,000 individuals from the Framingham Heart Study. Lastly, using a random subset of individuals of European ancestry from the UK Biobank, we show that BANNs is able to replicate known associations in high and low-density lipoprotein cholesterol content. Author SummaryA common goal in genome-wide association (GWA) studies is to characterize the relationship between genotypic and phenotypic variation. Linear models are widely used tools in GWA analyses, in part, because they provide significance measures which detail how individual single nucleotide polymorphisms (SNPs) are statistically associated with a trait or disease of interest. However, traditional linear regression largely ignores non-additive genetic variation, and the univariate SNP-level mapping approach has been shown to be underpowered and challenging to interpret for certain trait architectures. While nonlinear methods such as neural networks are well known to account for complex data structures, these same algorithms have also been criticized as "black box" since they do not naturally carry out statistical hypothesis testing like classic linear models. This limitation has prevented nonlinear regression approaches from being used for association mapping tasks in GWA applications. Here, we present Biologically Annotated Neural Networks (BANNs): a flexible class of feedforward models with partially connected architectures that are based on biological annotations. The BANN framework uses approximate Bayesian inference to provide interpretable probabilistic summaries which can be used for simultaneous (i) mapping with SNPs and (ii) enrichment analyses with SNP-sets (e.g., genes or signaling pathways). We illustrate the benefits of our method over state-of-the-art approaches using extensive simulations. We also demonstrate the ability of BANNs to recover novel and previously discovered genomic associations using quantitative traits from the Wellcome Trust Centre for Human Genetics, the Framingham Heart Study, and the UK Biobank.

genetics↗