Search bioRxiv⌕ Search

Biology subjects

Peng, J.

Publications and source records attributed to Peng, J..

At least 181 records · Page 10Linked to original sources
bioRxiv2021

Jumper Enables Discontinuous Transcript Assembly in Coronaviruses

Genes in SARS-CoV-2 and, more generally, in viruses in the order of Nidovirales are expressed by a process of discontinuous transcription mediated by the viral RNA-dependent RNA polymerase. This process is distinct from alternative splicing in eukaryotes, rendering current transcript assembly methods unsuitable to Nidovirales sequencing samples. Here, we introduce the DO_SCPLOWISCONTINUOUSC_SCPLOW TO_SCPLOWRANSCRIPTC_SCPLOW AO_SCPLOWSSEMBLYC_SCPLOW problem of finding transcripts [Formula] and their abundances c given an alignment [Formula] under a maximum likelihood model that accounts for varying transcript lengths. Underpinning our approach is the concept of a segment graph, a directed acyclic graph that, distinct from the splice graph used to characterize alternative splicing, has a unique Hamiltonian path. We provide a compact characterization of solutions as subsets of non-overlapping edges in this graph, enabling the formulation of an efficient mixed integer linear program. We show using simulations that our method, JO_SCPLOWUMPERC_SCPLOW, drastically outperforms existing methods for classical transcript assembly. On short-read data of SARS-CoV-1 and SARS-CoV-2 samples, we find that JO_SCPLOWUMPERC_SCPLOW not only identifies canonical transcripts that are part of the reference transcriptome, but also predicts expression of non-canonical transcripts that are well supported by direct evidence from long-read data, presence in multiple, independent samples or a conserved core sequence. JO_SCPLOWUMPERC_SCPLOW enables detailed analyses of Nidovirales transcriptomes. Code availabilitySoftware is available at https://github.com/elkebir-group/Jumper

bioinformatics↗
bioRxiv2021

Intermolecular interactions drive protein adaptive and co-adaptive evolution at both species and population levels

Proteins are the building blocks for almost all the functions in cells. Understanding the molecular evolution of proteins and the forces that shape protein evolution is essential in understanding the basis of function and evolution. Previous studies have shown that adaptation frequently occurs at the protein surface, such as in genes involved in host-pathogen interactions. However, it remains unclear whether adaptive sites are distributed randomly or at regions associated with particular structural or functional characteristics across the genome, since many proteins lack structural or functional annotations. Here, we seek to tackle this question by combining large-scale bioinformatic prediction, structural analysis, phylogenetic inference, and population genomic analysis of Drosophila protein-coding genes. We found that protein sequence adaptation is more relevant to function-related rather than structure-related properties. Interestingly, intermolecular interactions contribute significantly to protein adaptation. We further showed that intermolecular interactions, such as physical interactions may play a role in the co-adaptation of fast-adaptive proteins. We found that strongly differentiated amino acids across geographic regions in protein-coding genes are mostly adaptive, which may contribute to the long-term adaptive evolution. This strongly indicates that a number of adaptive sites tend to be repeatedly mutated and selected in evolution, in the past, present, and maybe future. Our results highlight the important roles of intermolecular interactions and co-adaptation in the adaptive evolution of proteins both at the species and population levels.

evolutionary biology↗
bioRxiv2021

Large field of view-spatially resolved transcriptomics at nanoscale resolution

Spatially resolved transcriptomic technologies are promising tools to study cell fate decisions in a physical microenvironment, which is fundamental for enhancing our knowledge of mammalian development. However, the imbalance between resolution, transcript capture and field of view of current methodologies precludes their systematic application to analyze relatively large and three-dimensional mid- and late-gestation mammalian embryos. Here, we combined DNA nanoball (DNB) patterned arrays and tissue RNA capture to create SpaTial Enhanced REsolution Omics-sequencing (Stereo-seq). This approach allows transcriptomic profiling of large histological sections with high resolution and sensitivity. We have applied Stereo-seq to study the kinetics and directionality of transcriptional variation in a time course of mouse organogenesis. We used this information to gain insight into the molecular basis of regional specification, neuronal migration and differentiation in the developing brain. Furthermore, we mapped the expression of a panel of developmental disease-related loci on our global transcriptomic maps to define the spatiotemporal windows of tissue vulnerability. Our panoramic atlas constitutes an essential resource to investigate longstanding questions concerning normal and abnormal mammalian development.

cell biology↗
bioRxiv2020

A Novel SARS-CoV-2 Multitope Protein/Peptide Vaccine Candidate is Highly Immunogenic and Prevents Lung Infection in an Adeno Associated Virus Human Angiotensin-Converting Enzyme 2 (AAV hACE2) Mouse Model

A novel multitope protein-peptide vaccine against Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) infection and disease is described in this report. The initial development and characterization experiments are presented along with proof-of-concept studies for the vaccine candidate UB-612. UB-612 consists of eight components rationally designed for induction of potently neutralizing antibodies and broad T cell responses against SARS-CoV-2: the S1-RBD-sFc fusion protein, six synthetic peptides (one universal peptide and five SARS-CoV-2-derived peptides), a proprietary CpG TLR-9 agonist at low concentration as an excipient, and aluminum phosphate adjuvant. Through immunogenicity studies in Guinea pigs and rats, we optimized the design of protein/peptide immunogens and selected an adjuvant system, yielding a vaccine that provides excellent S1-RBD binding and high neutralizing antibody responses, robust cellular responses, and a Th1-oriented response at low doses. In challenge studies, UB- 612 vaccination reduced viral load and prevented development of disease in mouse and non-human primate challenge models. With a Phase 1 trial completed, a Phase 2 trial ongoing in Taiwan, and additional trials planned to support global authorizations, UB-612 is a highly promising and differentiated vaccine candidate for prevention of SARS-CoV-2 infection and COVID-19 disease. Author SummarySARS-CoV-2 virus, the causative agent of Coronavirus Disease 2019 (COVID-19), has spread globally since its origin in 2019, causing an unprecedented public health crisis that has resulted in greater than 4.7 million deaths worldwide. Many vaccines are under development to limit disease spread and reduce the number of cases, but additional candidates that promote a robust immune response are needed. Here, we describe a multitope protein-peptide vaccine platform that is unique among COVID-19 vaccines. The advantages of our approach are induction of both high levels of neutralizing antibodies as well as a Th/CTL response in the vaccinated host, which mimics the immune response that occurs after natural infection with SARS-CoV-2. We demonstrate that our vaccine is immunogenic and effective in preventing disease in several animal models, including AAV- hACE-2 transduced mice, and both rhesus and cynomolgus macaques. Importantly, no immunopathology was observed in the lungs of immunized animals, therefore showing that antibody-dependent enhancement (ADE) does not occur. Our study provides an additional, novel vaccine candidate for advancement in clinical trials to treat and prevent SARS-CoV-2 infection and COVID-19 disease.

immunology↗
bioRxiv2020

Unified neural pathways that gate affective pain and multisensory innate threat signals to the amygdala

Perception of aversive sensory stimuli such as pain and innate threat cues is essential for animal survival. The amygdala is critical for aversive sensory perception, and it has been suggested that multiple parallel pathways independently relay aversive cues from each sensory modality to the amygdala. However, a convergent pathway that relays multisensory aversive cues to the amygdala has not been identified. Here, we report that neurons expressing calcitonin gene-related peptide (CGRP) in the parvocellular subparafasicular thalamic nucleus (SPFp) are necessary and sufficient for affective-motivational pain perception by forming a spino-thalamo-amygdaloid pain pathway. In addition, we find that this thalamic CGRP pain pathway, together with well-known parabrachio-amygdaloid CGRP pain pathway, is critical for the perception of multisensory innate threat cues. The discovery of unified pathways that collectively gate aversive sensory stimuli from all sensory modalities may provide critical circuit-based insights for developing therapeutic interventions for affective pain- and innate fear-related disorders.

neuroscience↗
bioRxiv2020

DAGBagM: Learning directed acyclic graphs of mixed variables with an application to identify prognostic protein biomarkers in ovarian cancer

MotivationDirected gene/protein regulatory networks inferred by applying directed acyclic graph (DAG) models to proteogenomic data has been shown effective for detecting causal biomarkers of clinical outcomes. However, there remain unsolved challenges in DAG learning to jointly model clinical outcome variables, which often take binary values, and biomarker measurements, which usually are continuous variables. Therefore, in this paper, we propose a new tool, DAGBagM, to learn DAGs with both continuous and binary nodes. By using appropriate models for continuous and binary variables, DAGBagM allows for either type of nodes to be parents or children nodes in the learned graph. DAGBagM also employs a bootstrap aggregating strategy to reduce false positives and achieve better estimation accuracy. Moreover, the aggregation procedure provides a flexible framework to robustly incorporate prior information on edges for DAG reconstruction. ResultsAs shown by simulation studies, DAGBagM performs better in identifying edges between continuous and binary nodes, as compared to commonly used strategies of either treating binary variables as continuous or discretizing continuous variables. Moreover, DAGBagM outperforms several popular DAG structure learning algorithms including the score-based hill climbing (HC) algorithm, constraint-based PC-algorithm (PC-alg), and the hybrid method max-min hill climbing (MMHC) even for constructing DAG with only continuous nodes. The HC implementation in the R package DAGBagM is much faster than that in a widely used DAG learning R package bnlearn. When applying DAGBagM to proteomics datasets from ovarian cancer studies, we identify potential prognostic protein biomarkers in ovarian cancer. Availability and implementationDAGBagM is made available as a github repository https://github.com/jie108/dagbagM.

systems biology↗
bioRxiv2020

Behavioral and genomic sensory adaptations underlying the pest activity of Drosophila suzukii

Studying how novel phenotypes originate and evolve is fundamental to the field of evolutionary biology as it allows us to understand how organismal diversity is generated and maintained. However, determining the basis of novel phenotypes is challenging as it involves orchestrated changes at multiple biological levels. Here, we aim to overcome this challenge by using a comparative species framework combining behavioral, gene expression, and genomic analyses to understand the evolutionary novel egg-laying substrate-choice behavior of the invasive pest species Drosophila suzukii. First, we used egg-laying behavioral assays to understand the evolution of ripe fruit oviposition preference in D. suzukii as compared to closely related species D. subpulchrella and D. biarmipes, as well as D. melanogaster. We show that D. subpulchrella and D. biarmipes lay eggs on both ripe and rotten fruits, suggesting that the transition to ripe fruit preference was gradual. Secondly, using two-choice oviposition assays, we studied how D. suzukii, D. subpulchrella, D. biarmipes and D. melanogaster differentially process key sensory cues distinguishing ripe from rotten fruit during egg-laying. We found that D. suzukiis preference for ripe fruit is in part mediated through a species-specific preference for stiff substrates. Lastly, we sequenced and annotated a high-quality genome for D. subpulchrella. Using comparative genomic approaches, we identified candidate genes involved in D. suzukiis ability to seek out and target ripe fruits. Our results provide detail to the stepwise evolution of pest activity in D. suzukii, indicating important cues used by this species when finding a host, and the molecular mechanisms potentially underlying their adaptation to a new ecological niche.

evolutionary biology↗
bioRxiv2020

Manage flowering time in Miscanthus and sugarcane to facilitate intra- and intergeneric crosses

Miscanthus is a close relative of Saccharum and a potentially valuable genetic resource for improving sugarcane. Differences in flowering time within and between Miscanthus and Saccharum hinders intra- and interspecific hybridizations. A series of greenhouse experiments were conducted over three years to determine how to synchronize flowering time of Saccharum and Miscanthus genotypes. We found that day length was an important factor influencing when Miscanthus and Saccharum flowered. Sugarcane could be induced to flower in a central Illinois greenhouse using supplemental lighting to reduce the rate at which days shortened during the autumn and winter to 1 min d-1, which allowed us to synchronize the flowering of some sugarcane genotypes with Miscanthus genotypes primarily from low latitudes. In a complementary growth chamber experiment, we evaluated 33 Miscanthus genotypes, including 28 M. sinensis, 2 M. floridulus, and 3 M. xgiganteus collected from 20.9{degrees} S to 44.9{degrees} N for response to three day lengths (10 h, 12.5 h, and 15 h). High latitude-adapted M. sinensis flowered mainly under 15 h days, but unexpectedly, short days resulted in short, stocky plants that did not flower; in some cases, flag leaves developed under short days but heading did not occur. In contrast, for M. sinensis and M. floridulus from low latitudes, shorter day lengths typically resulted in earlier flowering, and for some low latitude genotypes, 15 h days resulted in no flowering. However, the highest ratio of reproductive shoots to total number of culms was typically observed for 12.5 h or 15 h days. Latitude of origin was significantly associated with culm length, and the shorter the days, the stronger the relationship. Nearly all entries achieved maximal culm length under the 15 h treatment, but the nearer to the equator an accession originated, the less of a difference in culm length between the short-day treatments and the 15 h day treatment. Under short days, short culms for high-latitude accessions was achieved by different physiological mechanisms for M. sinensis genetic groups from the mainland in comparison to those from Japan; for mainland accessions, the mechanism was reduced internode length, whereas for Japanese accessions the phyllochron under short days was greater than under long days. Thus, for M. sinensis, short days typically hastened floral induction, consistent with the expectations for a facultative short-day plant. However, for high latitude accessions of M. sinensis, days less than 12.5 h also signaled that plants should prepare for winter by producing many short culms with limited elongation and development; moreover, this response was also epistatic to flowering. Thus, to flower M. sinensis that originates from high latitudes synchronously with sugarcane, the former needs day lengths >12.5 h (perhaps as high as 15 h), whereas that the latter needs day lengths <12.5 h.

plant biology↗
bioRxiv2020

VIP-HL: Semi-automated ACMG/AMP variant interpretation platform for genetic hearing loss

PurposeThe American College of Medical Genetics and Genomics, and the Association for Molecular Pathology (ACMG/AMP) have proposed a set of evidence-based guidelines to support sequence variant interpretation. The ClinGen hearing loss expert panel (HL-EP) introduced further specifications into the ACMG/AMP framework for genetic hearing loss. This study aimed to semi-automate the HL ACMG/AMP rules. MethodsVIP-HL aggregates information from external databases to automate 13 out of 24 ACMG/AMP rules specified by HL-EP, namely PVS1, PS1, PM1, PM2, PM4, PM5, PP3, BA1, BS1, BS2, BP3, BP4, and BP7. ResultsWe benchmarked VIP-HL using 50 variants where 83 rules were activated by the HL expert panel. VIP-HL concordantly activated 96% (80/83) rules, significantly higher than that of by InterVar (47%; 39/83). Of 4948 ClinVar star 2+ variants from 142 deafness-related genes, VIP-HL achieved an overall variant interpretation concordance in 88.0% (4353/4948). VIP-HL is available with a user-friendly web interface at http://hearing.genetics.bgi.com/. ConclusionVIP-HL is an integrated online tool for reliable automated variant classification in hearing loss genes. It assists curators in variant interpretation and provides a platform for users to share classifications with each other.

bioinformatics↗
bioRxiv2020

Concerted evolution reveals co-adapted amino acid substitutions in frogs that prey on toxic toads

Gene duplication is an important source of evolutionary innovation, but the adaptive division-of-labor between duplicates can be opposed by ongoing gene conversion between them. Here we document a tandem duplication of Na+,K+-ATPase subunit 1 (ATP1A1) shared by frogs in the genus Leptodactylus, a group of species that feeds on toxic toads. One ATP1A1 paralog evolved resistance to toad toxins while the other paralog retained ancestral susceptibility. We show that the two Leptodactylus paralogs are distinguished by 12 amino acid substitutions that were maintained by strong selection that counteracted the homogenizing effect of gene conversion. Protein-engineering experiments show that two major-effect substitutions confer toxin resistance, whereas the 10 additional substitutions mitigate deleterious pleiotropic effects on enzyme function. Our results highlight how trans-specific, neofunctionalized gene duplicates can provide unique insights into interactions between adaptive substitutions and the genetic backgrounds on which they arise. One Sentence SummarySelection counteracts gene conversion to maintain an adaptive division-of-labor between tandemly duplicated genes.

evolutionary biology↗
bioRxiv2020

Genome-Wide Sequencing as a First-Tier Screening Test for Short Tandem Repeat Expansions

Short tandem repeat (STR) expansions cause several neurological and neuromuscular disorders. Screening for STR expansions in genome-wide (exome and genome) sequencing data can enable diagnosis, optimal clinical management/treatment, and accurate genetic counselling of patients with repeat expansion disorders. We assessed the performance of lobSTR, HipSTR, RepeatSeq, ExpansionHunter, TREDPARSE, GangSTR, STRetch, and exSTRa - bioinformatics tools that have been developed to detect and/or genotype STR expansions - on experimental and simulated genome sequence data with known STR expansions aligned using two different aligners, Isaac and BWA. We then adjusted the parameter settings to optimize the sensitivity and specificity of the STR tools and fed the optimized results into a machine-learning decision tree classifier to determine the best combination of tools to detect full mutation expansions with high diagnostic sensitivity and specificity. The decision tree model supported using ExpansionHunters full mutation calls with those of either STRetch or exSTRa for detection of full mutations with precision, recall, and F1-score of 90%, 100%, and 95%, respectively. We used this pipeline to screen the BWA-aligned exome or genome sequence data of 306 families of children with suspected genetic disorders for pathogenic expansions of known disease STR loci. We identified 27 samples, 17 with an apparent full-mutation expansion of the AR, ATXN1, ATXN2, ATXN8, DMPK, FXN, HTT, or TBP locus, nine with an intermediate or premutation allele in the FMR1 locus, and one with a borderline allele in the ATXN2 locus. We report the concordance between our bioinformatics findings and the clinical PCR results in a subset of these samples. Implementation of our bioinformatics workflow can improve the detection of disease STR expansions in exome and genome sequence diagnostics and enhance clinical outcomes for patients with repeat expansion disorders.

genetics↗
bioRxiv2020

Characterization of SARS-CoV-2 viral diversity within and across hosts

In light of the current COVID-19 pandemic, there is an urgent need to accurately infer the evolutionary and transmission history of the virus to inform real-time outbreak management, public health policies and mitigation strategies. Current phylogenetic and phylodynamic approaches typically use consensus sequences, essentially assuming the presence of a single viral strain per host. Here, we analyze 621 bulk RNA sequencing samples and 7,540 consensus sequences from COVID-19 patients, and identify multiple strains of the virus, SARS-CoV-2, in four major clades that are prevalent within and across hosts. In particular, we find evidence for (i) within-host diversity across phylogenetic clades, (ii) putative cases of recombination, multi-strain and/or superinfections as well as (iii) distinct strain profiles across geographical locations and time. Our findings and algorithms will facilitate more detailed evolutionary analyses and contact tracing that specifically account for within-host viral diversity in the ongoing COVID-19 pandemic as well as future pandemics.

bioinformatics↗
bioRxiv2020

SimiC: A Single Cell Gene Regulatory Network Inference method with Similarity Constraints

Single-cell RNA-Sequencing has made it possible to infer high-resolution gene regulatory networks (GRNs), providing deep biological insights by revealing regulatory interactions at single-cell resolution. However, current single-cell GRN analysis methods produce only a single GRN per input dataset, potentially missing relationships between cells from different phenotypes. To address this issue, we present SimiC, a single-cell GRN inference method that produces a GRN per phenotype while imposing a similarity constraint that forces a smooth transition between GRNs, allowing for a direct comparison between different states, treatments, or conditions. We show that jointly inferring GRNs can uncover variation in regulatory relationships across phenotypes that would have otherwise been missed. Moreover, SimiC can recapitulate complex regulatory dynamics across a range of systems, both model and non-model alike. Taken together, we establish a new approach to quantitating regulatory architectures between the GRNs of distinct cellular phenotypes, with far-reaching implications for systems biology.

bioinformatics↗
bioRxiv2020

Identification of a common deletion in the spike protein of SARS-CoV-2

Two notable features have been identified in the SARS-CoV-2 genome: (1) the receptor binding domain of SARS-CoV-2; (2) a unique insertion of twelve nucleotide or four amino acids (PRRA) at the S1 and S2 boundary. For the first feature, the similar RBD identified in SARs-like virus from pangolin suggests the RBD in SARS-CoV-2 may already exist in animal host(s) before it transmitted into human. The left puzzle is the history and function of the insertion at S1/S2 boundary, which is uniquely identified in SARS-CoV-2. In this study, we identified two variants from the first Guangdong SARS-CoV-2 cell strain, with deletion mutations on polybasic cleavage site (PRRAR) and its flank sites. More extensive screening indicates the deletion at the flank sites of PRRAR could be detected in 3 of 68 clinical samples and half of 22 in vitro isolated viral strains. These data indicate (1) the deletion of QTQTN, at the flank of polybasic cleavage site, is likely benefit the SARS-CoV-2 replication or infection in vitro but under strong purification selection in vivo since it is rarely identified in clinical samples; (2) there could be a very efficient mechanism for deleting this region from viral genome as the variants losing 23585-23599 is commonly detected after two rounds of cell passage. The mechanistic explanation for this in vitro adaptation and in vivo purification processes (or reverse) that led to such genomic changes in SARS-CoV-2 requires further work. Nonetheless, this study has provided valuable clues to aid further investigation of spike protein function and virus evolution. The deletion mutation identified in vitro isolation should be also noted for current vaccine development.

genetics↗
bioRxiv2020

Characterizing sensitivity and coverage of clinical WGS as a diagnostic test for genetic disorders

BackgroundWith the reduce of cost and incomparable advantages, WGS is likely to change the way of clinical diagnosis of rare and undiagnosed diseases. However, the sensitivity and breadth of coverage of clinical WGS as a diagnostic test for genetic disorders has not been fully evaluated, especially for CNV detection. MethodsAll the samples underwent WGS using MGISEQ-2000. The performance of NA12878, YH cell line, and the Chinese trios were measured for sensitivity, PPV, depth and breadth of coverage. We also compared the performance of WES and WGS using NA12878. The sensitivity and PPV were tested using family-based trio design in the Chinese trios. We also developed a systematic WGS pipeline for the analysis of 8 clinical cases with known disease-causing variants. ResultsIn general, the sensitivity and PPV for SNV/indel detection increased with mean depth, and reached a plateau at a ~40X mean depth using down-sampling samples of NA12878. With a mean depth of 40X, the sensitivity of homozyous and heterozygous SNPs of NA12878 was >99.25% and >99.50% respectively, and PPVs were 99.97% and 98.96%. Homozygous and heterozygous indels showed lower sensitivity and PPV. The sensitivity and PPV is still not 100% even with a mean depth of ~150X. We also observed a substantial variation in the sensitivity of CNV detection across different tools, especially in CNVs with a size of less than 1kb. In general, the breadth of coverage for disease-associated genes and CNVs increased with mean depth. The sensitivity and coverage of WGS (~40X) is better than WES (~120X). Among the Chinese trios with ~40X mean depth, the sensitivity in the offspring was >99.48% and >96.36% for SNP and indel detection, and PPVs were 99.86% and 97.93%. All the 12 previously validated variants in the 8 clinical cases were successfully detected by our WGS pipeline. ConclusionsThe current standard of a mean depth of 40X may be sufficient for SNV/indel detection and identification of most CNVs. Clinical scientists should know the range of sensitivity and PPV for different classes of variants for a particular WGS pipeline, which would be useful when interpreting and delivering clinical reports.

genomics↗
bioRxiv2020

Structure-Based Design, Synthesis and Biological Evaluation of Peptidomimetic Aldehydes as a Novel Series of Antiviral Drug Candidates Targeting the SARS-CoV-2 Main Protease

SARS-CoV-2 is the etiological agent responsible for the COVID-19 outbreak in Wuhan. Specific antiviral drug are urgently needed to treat COVID-19 infections. The main protease (Mpro) of SARS-CoV-2 is a key CoV enzyme that plays a pivotal role in mediating viral replication and transcription, which makes it an attractive drug target. In an effort to rapidly discover lead compounds targeting Mpro, two compounds (11a and 11b) were designed and synthesized, both of which exhibited excellent inhibitory activity with an IC50 value of 0.05 M and 0.04 M respectively. Significantly, both compounds exhibited potent anti-SARS-CoV-2 infection activity in a cell-based assay with an EC50 value of 0.42 M and 0.33 M, respectively. The X-ray crystal structures of SARS-CoV-2 Mpro in complex with 11a and 11b were determined at 1.5 [A] resolution, respectively. The crystal structures showed that 11a and 11b are covalent inhibitors, the aldehyde groups of which are bound covalently to Cys145 of Mpro. Both compounds showed good PK properties in vivo, and 11a also exhibited low toxicity which is promising drug leads with clinical potential that merits further studies.

biochemistry↗
bioRxiv2020

Synthesis of diagnostic quality cancer pathology images

Deep learning-based computer vision methods have recently made remarkable breakthroughs in the analysis and classification of cancer pathology images. However, there has been relatively little investigation of the utility of deep neural networks to synthesize medical images. In this study, we evaluated the efficacy of generative adversarial networks (GANs) to synthesize high resolution pathology images of ten histological types of cancer, including five cancer types from The Cancer Genome Atlas (TCGA) and the five major histological subtypes of ovarian carcinoma. The quality of these images was assessed using a comprehensive survey of board-certified pathologists (n = 9) and pathology trainees (n = 6). Our results show that the real and synthetic images are classified by histotype with comparable accuracies, and the synthetic images are visually indistinguishable from real images. Furthermore, we trained deep convolutional neural networks (CNNs) to diagnose the different cancer types and determined that the synthetic images perform as well as additional real images when used to supplement a small training set. These findings have important applications in proficiency testing of medical practitioners and quality assurance in clinical laboratories. Furthermore, training of computer-aided diagnostic systems can benefit from synthetic images where labeled datasets are limited (e.g., rare cancers). We have created a publicly available website where clinicians and researchers can attempt questions from the image survey at http://gan.aimlab.ca/.

bioinformatics↗
bioRxiv2020

Batch Equalization with a Generative Adversarial Network

Advances in automation and imaging have made it possible to capture large image datasets for experiments that span multiple weeks with multiple experimental batches of data. However, accurate biological comparisons across the batches is challenged by the batch-to-batch variation due to uncontrollable experimental noise (e.g., different stain intensity or illumination conditions). To mediate the batch variation (i.e. the batch effect), we developed a batch equalization method that can transfer images from one batch to another while preserving the biological phenotype. The equalization method is trained as a generative adversarial network (GAN), using the StarGAN architecture that has shown considerable ability in doing style transfer for consumer images. After incorporating an additional objective that disentangles batch effect from biological features using an existing GAN framework, we show that the equalized images have less batch information as determined by a batch-prediction task and perform better in a biologically relevant task (e.g., Mechanism of Action prediction).

cell biology↗