Biology subjects

Cai, L.

Publications and source records attributed to Cai, L..

At least 91 records · Page 5Linked to original sources

bioRxiv2022

A cryptic transactivation domain of EZH2 binds AR and AR's splice variant promoting oncogene activation and tumorous transformation

Enhancer of Zeste Homolog 2 (EZH2) and androgen receptor (AR) are crucial chromatin regulators involved in the development and/or progression of prostate tumor, including advanced castration-resistant prostate cancer (CRPC). To sustain prostate tumorigenicity, EZH2 establishes noncanonical biochemical interaction with AR for mediating oncogene activation, in addition to its canonical role as a transcriptional repressor and enzymatic subunit of Polycomb Repressive Complex 2 (PRC2). However, the molecular basis underlying non-canonical activities of EZH2 in prostate cancer remains elusive and therapeutic strategies for targeting EZH2:AR-mediated oncogene activation activities are also lacking. Here, we report that a cryptic transactivation domain of EZH2 (EZH2TAD) binds both AR and AR spliced variant 7 (AR-V7, an AR variant enriched in CRPC), mediating assembly and/or recruitment of transactivation-related machineries at genomic sites that lack PRC2 binding. Such noncanonical targets of EZH2:AR/AR-V7:(co)activators are enriched for the clinically-relevant oncogenes. We also show that EZH2TAD is required for the chromatin recruitment of EZH2, for EZH2-mediated oncogene activation, and for CRPC growth in vitro and in vivo. To completely block EZH2s multifaceted oncogenic activities in prostate cancer, we employed MS177, a recently developed proteolysis targeting chimera (PROTAC) of EZH2. Strikingly, MS177 achieved on-target depletion of both EZH2s canonical (EZH2:PRC2) and noncanonical (EZH2TAD:AR/AR-V7:coactivators) complexes in prostate tumor, eliciting much more potent antitumor effects than the catalytic inhibitors of EZH2. Overall, this study reports previously unappreciated requirements of EZH2TAD for mediating EZH2s noncanonical (co)activator recruitment and gene-activation functions in prostate tumor and suggests EZH2-targeting PROTACs as potentially attractive therapeutics for the treatment of aggressive prostate tumors that rely on the circuits wired by EZH2 and AR.

cancer biology↗

bioRxiv2022

The small cell lung cancer neuroendocrine transdifferentiation explorer

SCLC is a high-grade neuroendocrine (NE) cancer that exhibits cellular plasticity. Transdifferention into non-NE cells creates considerable intra-tumoral heterogeneity, enhanced metastasis, greater tumor burden, and treatment resistance. Similar NE transdifferentiation has been observed in neuroblastoma (NBL). Targeting NE plasticity and cooperation between NE and non-NE cells in the tumor microenvironment may provide an avenue to enhance and restore sensitivity to available treatments. Although substantial transcriptomic changes take place upon NE transdifferentiation, conservation of these changes has not been investigated. In this study, we extensively curated genes associated with NE transdifferentiation in SCLC. We collected 35 datasets and compared the NE score-associated transcriptome across studies, for SCLC vs. NBL human tumors, human NBL tumors vs. cell lines, SCLC human tumors vs. tumors from genetically engineered mouse models (GEMMs), and SCLC GEMM uncultured cancer cells vs. cultured cancer cells. We have also created a user-friendly web application for researchers to explore these results. This work establishes a useful resource for researchers to understand the NE transdifferentiation landscape and explore context-dependent NE associations in SCLC and NBL.

cancer biology↗

bioRxiv2022

Comprehensive isotopomer analysis of glutamate and aspartate in small tissue samples

Stable isotopes are powerful tools to assess metabolism. 13C labeling is detected using nuclear magnetic resonance spectroscopy (NMRS) or mass spectrometry (MS). MS has excellent sensitivity but generally cannot discriminate among different 13C positions (isotopomers), whereas NMRS is less sensitive but reports some isotopomers. Here, we develop an MS method that reports all 16 aspartate and 32 glutamate isotopomers while requiring 1% of the sample used for NMRS. This method discriminates between pathways that result in the same number of 13C labels in aspartate and glutamate, providing enhanced specificity over conventional MS. We demonstrate regional metabolic heterogeneity within human tumors, document the impact of fumarate hydratase deficiency in human renal cancers, and investigate the contributions of TCA cycle turnover and CO2 recycling to isotope labeling in vivo. This method can accompany NMRS or standard MS to provide outstanding sensitivity in isotope labeling experiments, particularly in vivo.

biochemistry↗

bioRxiv2022

Longitudinal structural change of gray matter before and after hemispherotomy

BackgroundHemispherotomy is an effective surgery developed to treat refractory epilepsy caused by diffuse unihemispheric pathologies. Post-surgery neuroplastic mechanisms supporting behavioral recovery after left and right hemispherotomy as well as their difference remain largely unclear. MethodsIn a large cohort of 57 pediatric patients who underwent hemispherotomy, voxel-wise GMV in unaffected regions (contralesional cerebrum and bilateral cerebellum) and behavioral abilities were assessed before and after surgery. Using linear mixed models, we evaluated changes in GMV and behavioral scores, and how GMV changes correlated with improvements in behavioral scores. In addition, three machine learning regression algorithms were applied to assess whether preoperative GMV can predict postoperative behavior. ResultsIn the two patient groups (29 left hemispherotomy patients, age of surgery mean {+/-} standard deviation = 3.5 {+/-} 2.5; 28 right hemispherotomy patients, age of surgery 4.6 {+/-} 2.5), widespread increases in the postoperative GMV in the contralateral cerebrum and ipsilateral cerebellum but decreases in the postoperative GMV in the contralateral cerebellum were consistently observed. Importantly, the decrease in GMV in the contralateral cerebellum was significantly correlated with improvement in behavioral scores in right but not left hemispherotomy patients. Moreover, the preoperative GMV around the most longitudinally changed locations significantly predicted postoperative behavioral scores in left but not right hemispherotomy patients. ConclusionsOur findings indicate an important role for the contralateral cerebellum in the recovery after right hemispherotomy, and postoperative behavioral scores can be predicted with preoperative GMV features. What is already known on this topicThe contralesional cerebrum plays a key role in the recovery after hemispherotomy. What this study addsDespite suffering from gray matter atrophy, GMV decrease in contralesional cerebellum is associated with improvement in behavioral score in patients after hemispherotomy. How this study might affect research, practice or policyThese results provide novel insight into the prominence of the contralateral cerebellum in the recovery after hemispherotomy and highlight the clinical role of presurgery GMV in the prediction of postsurgery behavior.

neuroscience↗

bioRxiv2022

Glutamine metabolism modulates chondrocyte inflammatory response

Osteoarthritis is the most common joint disease in the world with significant societal consequences, but lacks effective disease modifying interventions. The pathophysiology consists of a prominent inflammatory component that can be targeted to prevent cartilage degradation and structural defects. Intracellular metabolism has emerged as a culprit of the inflammatory response in chondrocytes, with both processes co-regulating each other. The role of glutamine metabolism in chondrocytes, especially in the context of inflammation, lacks a thorough understanding and is the focus of this work. We display that chondrocytes utilize glutamine for energy production and anabolic processes. Furthermore, we show that glutamine deprivation itself causes metabolic reprogramming and decreases the inflammatory response of chondrocytes through inhibition of NF-{kappa}B activity. Finally, we display that glutamine deprivation promotes autophagy and that ammonia is an inhibitor of autophagy. Overall, we identify a relationship between glutamine metabolism and inflammatory signaling and display the need for increased study of chondrocyte metabolic systems.

biochemistry↗

bioRxiv2022

Development and application of an uncapped mRNA platform

A novel uncapped mRNA platform was developed. Five lipid nanoparticle (LNP)-encapsulated mRNA constructs were made to evaluate several aspects of our platform, including transfection efficiency and durability in vitro and in vivo and the activation of humoral and cellular immunity in several animal models. The constructs were eGFP-mRNA-LNP (for enhanced green fluorescence mRNA), Fluc-mRNA-LNP (for firefly luciferase mRNA), S{delta}T-mRNA-LNP (for Delta strain SARS-CoV-2 spike protein trimer mRNA), gDED-mRNA-LNP (for truncated glycoprotein D mRNA coding ectodomain from herpes simplex virus type 2 (HSV2)) and gDFR-mRNA-LNP (for truncated HSV2 glycoprotein D mRNA coding amino acids 1[~]400). Quantifiable target protein expression was achieved in vitro and in vivo with eGFP-and Fluc-mRNA-LNP. S{delta}T-mRNA-LNP, gDED-mRNA-LNP and gDFR-mRNA-LNP induced both humoral and cellular immune responses comparable to those obtained by previously reported capped mRNA-LNP constructs. Notably, S{delta}T-mRNA-LNP elicited neutralizing antibodies in hamsters against the Omicron and Delta strains. Additionally, gDED-mRNA-LNP and gDFR-mRNA-LNP induced potent neutralizing antibodies in rabbits and mice. The mRNA constructs with uridine triphosphate (UTP) outperformed those with N1-methylpseudouridine triphosphate (N1m{psi}TP) in the induction of antibodies via S{delta}T-mRNA-LNP. Our uncapped, process-simplified, and economical mRNA platform may have broad utility in vaccines and protein replacement drugs.

bioengineering↗

bioRxiv2022

Respiratory mucosal vaccination of peptide-poloxamine-DNA nanoparticles provides complete protection against lethal SARS-CoV-2 challenge

The ongoing SARS-CoV-2 pandemic represents a brutal reminder of the continual threat of mucosal infectious diseases. Mucosal immunity may provide robust protection at the predominant sites of SARS-CoV-2 infection. However, it remains unclear whether respiratory mucosal administration of DNA vaccines could confer protective immune responses against SARS-CoV-2 challenge due to the insurmountable barriers posed by the airway. Here, we applied self-assembled peptide-poloxamine nanoparticles with mucus-penetrating properties for pulmonary inoculation of a COVID-19 DNA vaccine (pSpike/PP-sNp). Not only displays the pSpike/PP-sNp superior gene-transfection and favorable biocompatibility in the mouse airway, but pSpike/PP-sNp promotes a tripartite immunity consisting of systemic, cellular and mucosal immune responses that are characterized by mucosal IgA secretion, high levels of neutralizing antibodies, and resident memory phenotype T-cell responses in the lungs of mice. Most importantly, pSpike/PP-sNp completely eliminates SARS-CoV-2 infection in both upper and lower respiratory tracts and enables 100% survival rate of mice following lethal SARS-CoV-2 challenge. Our findings indicate PP-sNp might be a promising platform in mediating DNA vaccines to elicit all-around mucosal immunity against SARS-CoV-2.

immunology↗

bioRxiv2022

Evolution and diversification dynamics of butterflies

Butterflies are a diverse and charismatic insect group that are thought to have diversified via coevolution with plants and in response to dispersals following key geological events. These hypotheses have been poorly tested at the macroevolutionary scale because a comprehensive phylogenetic framework and datasets on global distributions and larval hosts of butterflies are lacking. We sequenced 391 genes from nearly 2,000 butterfly species to construct a new, phylogenomic tree of butterflies representing 92% of all genera and aggregated global distribution records and larval host datasets. We found that butterflies likely originated in what is now the Americas, [~]100 Ma, shortly before the Cretaceous Thermal Maximum, then crossed Beringia and diversified in the Paleotropics. The ancestor of modern butterflies likely fed on Fabaceae, and most extant families were present before the K/Pg extinction. The majority of butterfly dispersals occurred from the tropics (especially the Neotropics) to temperate zones, largely supporting a "cradle" pattern of diversification. Surprisingly, host breadth changes and shifts to novel host plants had only modest impacts.

evolutionary biology↗

bioRxiv2022

The largest haplotype-phased and chromosome-scale genome assembly of the fungal pathogen, Puccinia polysora, by HiFi reads and Hi-C data

Rust fungi are characterized by large genomes with high repeat content and have two haploid nuclei in most life stages, which makes achieving high-quality genome assemblies challenging. Here, we described a pipeline using HiFi reads and Hi-C data to assemble a gigabase-sized fungal pathogen, Puccinia polysora f.sp. zeae, to haplotype-phased and chromosome-scale. The final assembled genome is 1.71 Gbp, with ~850 Mbp and 18 chromosomes in each haplotype, being currently one of the two giga-scale fungi assembled to chromosome level. Transcript-based annotation identified 47,512 genes for dikaryotic genome with a similar number for each haplotype. A high level of interhaplotype variation was found with 10% haplotype-specific BUSCO genes, 5.8 SNPs/kbp and structural variation accounting for 3% of the genome size. The P. polysora genome displayed over 85% repeat contents, with genome-size expansion and copy number increasing of species-specific orthogroups. Interestingly, these features did not affect overall synteny with other Puccinia species having smaller genomes. Fine-time-point transcriptomics revealed seven clusters of co-expressed secreted proteins that are conserved between two haplotypes. The fact that candidate effectors interspersed with all genes indicated the absence of a "two-speed genome" evolution in P. polysora. Genome resequencing of 79 additional isolates revealed a clonal population structure of P. polysora in China with low geographic differentiation. Nevertheless, a minor population differentiated from the major population by having mutations on secreted proteins including AvrRppC, indicating the ongoing virulence to evade recognition by RppC, a major resistance gene in Chinese corn cultivars. The high-quality assembly provides valuable genomic resources for future studies on disease management and the evolution of P. polysora.

genomics↗

bioRxiv2022

Jak2V617F Reversible Activation Shows an Essential Requirement for Jak2V617F in Myeloproliferative Neoplasms

Janus kinases (JAKs) mediate cytokine signaling, cell growth and hematopoietic differentiation.1 Gain-of-function mutations activating JAK2 signaling are seen in the majority of myeloproliferative neoplasm (MPN) patients, most commonly due to the JAK2V617F driver allele.2 While clinically-approved JAK inhibitors improve symptoms and outcomes in MPNs, remissions are rare, and mutant allele burden does not substantively change with chronic JAK inhibitor therapy in most patients.3, 4 This has been postulated to be due to incomplete dependence on constitutive JAK/STAT signaling, alternative signaling pathways, and/or the presence of cooperating disease alleles;5 however we hypothesize this is due to the inability of current JAK inhibitors to potently and specifically abrogate mutant JAK2 signaling. We therefore developed a conditionally inducible mouse model allowing for sequential activation, and then inactivation, of Jak2V617F from its endogenous locus using a Dre-rox/Cre-lox dual orthogonal recombinase system. Deletion of oncogenic Jak2V617Fabrogates the MPN disease phenotype, induces mutant-specific cell loss including in hematopoietic stem/progenitor cells, and extends overall survival to an extent not observed with pharmacologic JAK inhibition. Furthermore, reversal of Jak2V617F in MPN cells with antecedent loss of Tet26, 7 abrogates the MPN phenotype and inhibits mutant stem cell persistence suggesting cooperating epigenetic-modifying alleles do not alter dependence on mutant JAK/STAT signaling. Our results suggest that mutant-specific inhibition of JAK2V617F represents the best therapeutic approach for JAK2V617F-mutant MPN and demonstrate the therapeutic relevance of a dual-recombinase system to assess mutant-specific oncogenic dependencies in vivo.

cancer biology↗

bioRxiv2022

Modeling clonal evolution and oncogenic dependency in vivo in the context of hematopoietic transformation

Cancer evolution is a multifaceted process involving the acquisition of somatic mutations and progressive epigenetic dysregulation of cellular fate. Both cell-intrinsic mechanisms and environmental interactions provide selective pressures capable of promoting clonal evolution and expansion, with single-cell and bulk DNA sequencing offering increased resolution into this process1-4. Advances in genome editing, single-cell biology and expressed lentiviral barcoding have enabled new insights into how transcriptional/epigenetic states change with clonal evolution5,6. Despite the extensive catalog of genomic alterations revealed by resequencing studies7,8, there remain limited means to functionally model and perturb this evolutionary process in experimental systems9. Here we integrated multi-recombinase (Cre, Flp, and Dre) tools for modeling reversible, sequential mutagenesis from premalignant clonal hematopoiesis to acute myeloid leukemia. We demonstrate that somatic acquisition of Flt3 activating mutations elicits distinct phases of acute and chronic activation resulting in differential cooperativity with Npm1 and Dnmt3a disease alleles. We next developed a generalizable allelic framework allowing for the reversible expression of oncogenic mutations at their endogenous loci. We found that reversal of mutant Flt3 resulted in rapid leukemic regression with distinct alterations in cellular compartments depending upon co-occurring mutations. These studies provide a path to model sequential mutagenesis and deterministically investigate mechanisms of transformation and oncogenic dependency in the context of clonal evolution.

cancer biology↗

bioRxiv2022

Manipulating the 3D organization of the largest synthetic yeast chromosome

Whether synthetic genomes can power life has attracted broad interest in the synthetic biology field, especially when the synthetic genomes are extensively modified with thousands of designer features. Here we report de novo synthesis of the largest eukaryotic chromosome thus far, synIV, a 1,454,621-bp Saccharomyces cerevisiae chromosome resulting from extensive genome streamlining and modification. During the construction of synIV, we developed megachunk assembly combined with a hierarchical integration strategy, which significantly increased the accuracy and flexibility of synthetic chromosome construction and facilitated chromosome debugging. In addition to the drastic sequence changes made to synIV by rewriting it, we further manipulated the three-dimensional structure of synIV in the yeast nucleus to explore spatial gene regulation within the nuclear space. Surprisingly, we found few gene expression changes, suggesting that positioning inside the yeast nucleoplasm plays a minor role in gene regulation. Lastly, we tethered synIV to the inner nuclear membrane via its hundreds of loxPsym sites and observed transcriptional repression of the entire chromosome, demonstrating chromosome-wide transcription manipulation without changing the DNA sequences. Our manipulation of the spatial structure of the largest synthetic yeast chromosome shed light on higher-order architectural design of the synthetic genomes. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=199 SRC="FIGDIR/small/487066v2_ufig1.gif" ALT="Figure 1"> View larger version (64K): org.highwire.dtl.DTLVardef@18bf586org.highwire.dtl.DTLVardef@1a471d5org.highwire.dtl.DTLVardef@12d7690org.highwire.dtl.DTLVardef@1c32510_HPS_FORMAT_FIGEXP M_FIG C_FIG HighlightsO_LIDe novo synthesis of the largest eukaryotic chromosome, synIV C_LIO_LISynIV shows similar 3D structure to wild-type IV, despite thousands of changes made to it C_LIO_LI"Inside-out" repositioning of synIV in nucleus shows minor transcriptional changes C_LIO_LIMultipoint tethering synIV to inner nuclear membrane represses transcription of whole chromosome C_LI

synthetic biology↗

bioRxiv2022

Machine learning improves global models of plant diversity

Despite the paramount role of plant diversity for ecosystem functioning, biogeochemical cycles, and human welfare, knowledge of its global distribution is incomplete, hampering basic research and biodiversity conservation. Here, we used machine learning (random forests, extreme gradient boosting, neural networks) and conventional statistical methods (generalised linear models, generalised additive models) to model species richness and phylogenetic richness of vascular plants worldwide based on 830 regional plant inventories including c. 300,000 species and predictors of past and present environmental conditions. Machine learning showed an outstanding performance, explaining up to 80.9% of species richness and 83.3% of phylogenetic richness. Current climate and environmental heterogeneity emerged as the primary drivers, while past environmental conditions left only small but detectable imprints on plant diversity. Finally, we combined predictions from multiple modelling techniques (ensemble predictions) to reveal global patterns and centres of plant diversity at multiple resolutions down to 7,774 km2. Our predictive maps provide the most accurate estimates of global plant diversity available to date at grain sizes relevant for conservation and macroecology.

ecology↗

bioRxiv2022

An intergeneric recombinant geminivirus causes soybean stay-green disease

Soybean is one of the most valuable legume crops in the world with high nutritional value. Recently, the outbreak of soybean stay-green syndrome has swept the soybean production in the Huang-Huai-Hai region of China, resulting in huge yield losses, which has become an epidemic and prominent problem in soybean production. However, the cause of the stay-green syndrome remains obscure. Here, we report a novel intergeneric recombinant geminivirus which causes soybean stay-green symptoms. Viral small RNA-based screening identified a new recombinant geminvirus from field soybean stay-green samples. The complete genome sequence of the virus contains 2762 nucleotide (nt) and appears to be an intergeneric recombinant virus in which protein coding for coat protein (V1) is similar to member of genus Mastrevirus, whereas proteins coding for V2, C2, C3 are most similar to those of viruses in the Maldovirus genus, and C1 and C4 are most similar to virus in genus Begomovirus. Inoculation of the infectious clone of the recombinant geminivirus through Agrobacterium rhizogenes causes typical soybean stay-green syndrome which resembles field symptoms including delayed leaf senescence, flat pods and abnormal seeds. The recombinant geminivirus can be detected in seed coat but not in cotyledon and embryo, thus failing to be transmitted by seeds. Moreover, the genome variation and epidemiological dynamic analysis were also carried out to help the continuous epidemiological surveillance of this emerging geminivirus. Collectively, this new geminivirus is tentatively named soybean stay-green associated virus (SoSGV). Our determination of the causal agent of soybean stay-green disease will bolster efforts to develop effective management strategies to control this prevalent disease in the field.

microbiology↗

bioRxiv2021

Single-cell RNA-seq analysis reveals lung epithelial cell-specific contributions of Tet1 to allergic inflammation

Tet1 protects against house dust mite (HDM)-induced lung inflammation in mice and alters the lung methylome and transcriptome. In order to explore the role of Tet1 in individual lung epithelial cell types in HDM-induced inflammation, we established a model of HDM-induced lung inflammation in Tet1 knockout and littermate wildtype mice and studied EpCAM+ lung epithelial cells using single-cell RNA-seq analysis. We identified eight EpCAM+ lung epithelial cell types, among which AT2 cells were the most abundant. HDM challenge increased the percentage of alveolar progenitor cells (AP), broncho alveolar stem cells (BAS), and goblet cells, and decreased the percentage of AT2 and ciliated cells. Bulk and cell-type-specific analysis identified genes subject to Tet1 regulation and linked to augmented lung inflammation, including alarms, detoxification enzymes, oxidative stress response genes, and genes in tissue repair. The transcriptomic regulation was accompanied by alterations in TF activities. Trajectory analysis supports that HDM may enhance the differentiation of AP and BAS cells into AT2 cells, independent of Tet1. Collectively, our data showed that lung epithelial cells had common and unique transcriptomic signatures of allergic lung inflammation. Tet1 deletion altered transcriptomic networks in various lung epithelial cells, with an overall effect of promoting allergen-induced lung inflammation.

immunology↗

bioRxiv2021

The blackgrass genome reveals patterns of divergent evolution of non-target site resistance to herbicides.

Globally, weedy plants result in more crop yield loss than plant pathogens and insect pests combined. Much of the success of weeds rests with their ability to rapidly adapt in the face of human-mediated environmental management and change. The evolution of resistance to herbicides is an emblematic example of this rapid adaptation. Here, we focus on Alopecurus myosuroides (blackgrass), the most impactful agricultural weed in Europe. To gain insights into the evolutionary history and genomic mechanisms underlying adaptation in blackgrass, we assembled and annotated its large, complex genome. We show that non-target site herbicide resistance is oligogenic and likely evolves from standing genetic variation. We present evidence for divergent selection of resistance at the level of the genome in wild, evolved populations, though at the transcriptional level, resistance mechanisms are underpinned by similar patterns of up-regulation of stress- and defence-responsive gene families. These gene families are expanded in the blackgrass genome, suggesting that the large, duplicated, and dynamic genome plays a role in enabling rapid adaptation in blackgrass. These observations have wide significance for understanding rapid plant adaptation in novel stressful environments.

genomics↗

bioRxiv2021

PhyloHerb: A phylogenomic pipeline for processing genome skimming data for plants

O_LIPremise of the study: The application of high throughput sequencing, especially to herbarium specimens, is greatly accelerating biodiversity research. Among various techniques, low coverage Illumina sequencing of total genomic DNA (genome skimming) can simultaneously recover the plastid, mitochondrial, and nuclear ribosomal regions across hundreds of species. Here, we introduce PhyloHerb -- a bioinformatic pipeline to efficiently and effectively assemble phylogenomic datasets derived from genome skimming. C_LIO_LIMethods and Results: PhyloHerb uses either a built-in database or user-specified references to extract orthologous sequences using BLAST search. It outputs FASTA files and offers a suite of utility functions to assist with alignment, data partitioning, concatenation, and phylogeny inference. The program is freely available at https://github.com/lmcai/PhyloHerb/. C_LIO_LIConclusions: Using published data from Clusiaceae, we demonstrated that PhyloHerb can accurately identify genes using highly fragmented assemblies derived from sequencing older herbarium specimens. Our approach is effective at all taxonomic depths and is scalable to thousands of species. C_LI

evolutionary biology↗

bioRxiv2021

iEnhancer-CLA: Self-attention-based interpretable model for enhancers and their strength prediction

Enhancer is a class of non-coding DNA cis-acting elements that plays a crucial role in the development of eukaryotes for their transcription. Computational methods for predicting enhancers have been developed and achieve satisfactory performance. However, existing computational methods suffer from experience-based feature engineering and lack of interpretability, which not only limit the representation ability of the models to some extent, but also make it difficult to provide interpretable analysis of the model prediction findings.In this paper, we propose a novel deep-learning-based model, iEnhancer-CLA, for identifying enhancers and their strengths. Specifically, iEnhancer-CLA automatically learns sequence 1D features through multiscale convolutional neural networks (CNN), and employs a self-attention mechanism to represent global features formed by multiple elements (multibody effects). In particular, the model can provide an interpretable analysis of the enhancer motifs and key base signals by decoupling CNN modules and generating self-attention weights. To avoid the bias of setting hyperparameters manually, we construct Bayesian optimization methods to obtain model global optimization hyperparameters. The results demonstrate that our method outperforms existing predictors in terms of accuracy for identifying enhancers and their strengths. Importantly, our analyses found that the distribution of bases in enhancers is uneven and the base G contents are more enriched, while the distribution of bases in non-enhancers is relatively even. This result contributes to the improvement of prediction performance and thus facilitates revealing an in-depth understanding of the potential functional mechanisms of enhancers. Author summaryThe enhancers contain many subspecies and the accuracy of existing models is difficult to improve due to the small data set. Motivated by the need for accurate and efficient methods to predict enhancer types, we developed a self-attention deep learning model iEnhancer-CLA, the aim is to be able to distinguish effectively and quickly between subspecies of enhancers and whether they are enhancers or not. The model is able to learn sequence features effectively through the combination of multi-scale CNN blocks, BLSTM layers, and self-attention mechanisms, thus improving the accuracy of the model. Encouragingly, by decoupling the CNN layer it was found that the layer was effective in learning the motif of the sequences, which in combination with the self-attention weights could provide interpretability to the model. We further performed sequence analysis in conjunction with the model-generated weights and discovered differences in enhancer and non-enhancer sequence characteristics. This phenomenon can be a guide for the construction of subsequent models for identifying enhancer sequences.

bioinformatics↗