jpad journal

AND option

OR option

PREDICTION OF ALZHEIMER’S DISEASE BASED ON 3D GENOME SELECTED CIRCRNA

 

R. Chi1,2,*, K. Li1,2,*, K. Su1,2, L. Liu1,2, M. Feng1,2, X. Zhang3, J. Wang3, X. Li4, G. He1,2, Y. Shi1,2,5

 

1. Bio-X Institutes, Key Laboratory for the Genetics of Developmental and Neuropsychiatric Disorders, Shanghai Jiao Tong University, 1954 Huashan Road, Shanghai, 200030, China; 2. Shanghai Key Laboratory of Psychotic Disorders, and Brain Science and Technology Research Center, Shanghai Jiao Tong University, 1954 Huashan Road, Shanghai, 200030, China; 3. Shanghai Institute of Thoracic Oncology, Shanghai Chest Hospital, Shanghai Jiaotong University, Shanghai, 200030, China;
4. Shanghai Mental Health Center, Shanghai Jiao Tong University School of Medicine, Shanghai, 200030, China; 5. Research Institute for Doping Control, Shanghai University of Sport, Shanghai, 200438, China. * These authors contributed equally as co-first authors.

Corresponding Author: Guang He, Yi Shi, Bio-X Institutes, Key Laboratory for the Genetics of Developmental and Neuropsychiatric Disorders, Shanghai Jiao Tong University, 1954 Huashan Road, Shanghai, 200030, China; heguang@sjtu.edu.cn, yishi@sjtu.edu.cn

J Prev Alz Dis 2024;
Published online March 6, 2024, http://dx.doi.org/10.14283/jpad.2024.52

 


Abstract

Alzheimer’s disease (AD) is a neurodegenerative disease and there is by far no effective treatment for it, especially in its late stage. Circular RNAs (circRNAs), known as a class of non-coding RNAs are widely observed in eukaryotic transcriptomes, and are reported to play an important role in neurodegenerative diseases including AD. circRNAs usually act as microRNA (miRNA) inhibitors or «sponges» to regulate the function of miRNAs, leading to subsequent changes in protein activities and functions. Accumulating evidence indicates that circRNAs can serve as potential biomarker in AD early prediction. The functional roles of circRNAs are very versatile including miRNAs binding – thus affecting downstream gene expression, generating abnormally translated protein peptides, and affecting epigenetic modifications which subsequently affect AD related gene expressions. Therefore, identifying AD-related circRNAs can contribute to AD early diagnosis and intervention. In this work, we collected and curated an AD-related circRNA dataset; by exploring the circRNAs’ corresponding DNA loci distribution in chromatin 3D conformation (3D genome) and utilize the such 3D genome information, we were able to selected a concise yet predictively effective circRNA panel, based on which, significantly better AD prediction machine learning models were achieved.

Key words: Circular RNA, epigenetics, 3D genome, Alzheimer’s disease, machine learning.


 

Introduction

Alzheimer’s disease (AD) is a neurodegenerative disease and is one of the leading causes of dementia (1). AD mainly manifested as cognitive and functional decline during agin, and can eventually lead to individual mortality (2). Characterized by the accumulation of amyloid-beta peptides (Aβ) as well as neurofibrillary tangles (NFTs) in brain (3), the underlying AD initiation and progression mechanism can be traced to single or multiple causes, including the transport and accumulation of the transmembrane protein APP to the presynaptic terminal which may lead to the enrichment of Aβ at the synapse (4, 5). Among the various forms of Aβ, Aβ(1-42) is considered to be the most neurotoxic form, which aggregates and leads to apoptosis of neuronal cells by increasing intracellular calcium levels or peroxidizing lipids in cell membranes (6). Mainly present in the brain neurons axons, tau protein is a structural protein related to the neuron cytoskeleton, and the deposition of Aβ also contributes to the hyperphosphorylation of tau protein, resulting in its misfolding and the formation of NFT aggregates, these processes thus impair communication between neurons and eventually leading to neuronal death (7). There is by far no effective clinical trial for treating Alzheimer’s disease due to a lack of conclusive opinion about the cause of AD (8).
Circular RNAs (circRNAs) were first discovered in pathogens as a class of covalently closed RNA loop molecules generated via exon or intron back splicing (9). CircRNAs are widespread in eukaryotes, and studies have shown that circRNAs can be highly expressed in the nervous system, especially enriched in synapses (10, 11). Specifically, circRNAs accumulate in the central nervous system (CNS) of Drosophila and Mouse as age increases (12, 13). At present, the most studied role of circRNAs is that it plays as miRNA «sponge», which affects the expression of miRNA and regulates the downstream gene expression and protein level (14). In addition, studies have also suggested that circRNAs can be translated into protein peptides under certain conditions (15). Moreover, several groups have demonstrated that some circRNAs are specifically expressed in Alzheimer’s disease. For instance, cirRS-7 functions as the sponge of miRNA-7 and down-regulation of its activity may increase the level of endogenous miRNA-7, leading to the down-regulation of AD-related target bio-entities such as ubiquitin protein ligase, UBE2A and autophagic phagocytic protein, which are essential to eliminate amyloid protein in the brain of AD patients (16-18).
The 3D structure of the nucleus is determined by the connection of the cytoskeleton to the nucleus, the integration and composition of the nuclear layer, and the chromatin folding in the nucleus (19, 20). The nucleus of human cells contains 46 tightly arranged chromosomes which are hierarchically packaged in the nucleus and constitute the 3D structure of the nucleus (21). The folding and positioning of chromatins in the nucleus play important role in the gene regulation, epigenetic modification, DNA replication and many other cellular processes (22). Various techniques have been developed to determine the 3D structure of chromatins, such as chromosome conformation capture (3C) (23) and its whole-genome extension Hi-C (24). High-through chromosome conformation capture (Hi-C) is a high-throughput chromatin conformation capture technique that maps global chromatin interactions in eukaryotic genomes (25). The 3D conformational model of the human genome using Hi-C data (26) can help us to intuitively discover the distribution of chromatins, so as to explore the 3D positions and interactions between genes and regulatory elements (27, 28).
Machine learning methods are increasingly recognized and adopted in the medical field for risk prediction of various diseases, either based on clinical data (29, 30) or multi-omics data (28, 31). In this work, we collected an AD related circRNA dataset from Mount Sinai/JJ Peters VA Medical Center Brain Bank (MSBB) (32). After curation, each sample in the dataset has a corresponding 0-5 clinical dementia rating (CDR) score, indicating the clinical severity of AD (33). We then studied the distribution pattern of circRNAs in genome 3D structure and incorporated the 3D genome information to select a small yet effective circRNA panel that exhibited more discriminatory power to distinguish different CDR scores comparing to using all circRNA expression profile or the top correlation coefficient derived circRNA panel, which we believe could serve as a better AD prediction biomarker panel.

 

Material and Methods

AD related circRNA Dataset

The original RNA-seq data, clinical information, and QC sheets were downloaded from the Synapse portal (https://www.synapse.org/#!Synapse:syn3159438), via application in May 2022. To remove low quality samples, the samples marked as «Remap» and «Exclude» in the provided QC list of the cohort were removed. Samples with RNA sequencing RIN scores below 4, or rRNA rates above 5% were also removed. Furthermore, only samples belonging to the European group were kept in the dataset to be genetically coherent. There are four brain regions in the MSBB data set: STL, PHG, IFG and aPFC. The samples in the aPFC region are removed due to the uneven distribution of the disease process. In the end, 164 samples from the STL region, 138 samples from the PHG region and 157 samples from the IFG region were retained for further analyses.

CircRNA quantification

The FASTQ and BAM files were downloaded and processed using Picard tools (http://broadinstitute.github.io/picard/). The data processing followed the CIRIexplorer2 (34) pipeline, in which TopHat2 (35) was chosen to map sequences to the human reference genome GRCh37/hg19 (36). After annotation, transcripts with zero reads in more than 25% across the retained samples were removed, and transcripts with average circRNA expression higher than two reads in any CDR group were retained (37). The expression of the annotated circRNAs was further processed by log2(n+1) where n>0 is the original expression value, to cancel potential outlier effect and serve as input in the subsequent prediction process and used for Spearman correlation coefficient calculations.

Acquisition of 3D genome coordinates of circRNAs

We collected and processed Hi-C data of eight cell-lines, i.e., K562, IMR90_Rao, HUVEC, NHEK, GM12878, hESC, IMR90 and KBM7 (38, 39), and computationally built chromatin 3D structure models for each of the eight cell-lines (26). The 3D genome models were built based on bins with a bin size = 500kb. Each circRNA’s starting position was mapped to a corresponding bin and the bin’s spatial coordinates <x, y, z> were assigned to each circRNA as its 3D genome locus.

Machine learning model building and circRNA selection

We employed four widely adopted machine learning regression models, i.e., K-Nearest Neighbor regression (KNN), Linear Regression Model (LR), Support Vector Regression (SVR) and eXtreme Gradient Boosting(XGBoost) for the predictions in each scenario, and the models were implemented based on the R package CARET and Keras (40). The prediction power of each model was evaluated based on 10-fold cross validation. We employed classification and regression algorithms to predict the association between circRNA and the progression of AD. As illustrated in Figure 1, for the classification algorithm, we conducted differential analysis and density-based clustering on all circRNAs within each brain region. Subsequently, we selected the most differentially expressed circRNAs from each cluster to form a new dataset. For the regression algorithm, for circRNA in each brain region, we first used the Spearman correlation coefficient (41) to filter circRNAs that are insignificantly correlated with the AD score. We then use k-means cluster k=100 based on circRNA expression and calculated the model importance to selected the circRNA with the top importance score in each cluster to form a 100 circRNA medium panel. Further, we used density-based clustering DBSCAN (42) to group the 100 circRNAs based on their 3D genome spatial coordinates and selected from each cluster a representative circRNA to form a 50 circRNA panel. For each panel, the four machine learning prediction models were tested under the 50 times 10-fold cross validation. All source code and data are available on GitHub (https://github.com/Rachelrqc/3D-AD-circRNA-prediction) and can be run on a personal computer via RStudio (https://www.rstudio.com/).

Figure 1. The workflow of circRNA panel selections and their AD score prediction effectiveness evaluated by different ML prediction modes under the 10-fold cross validation framework

 

Results

3D positional relationships of co-expressed genes in the nucleus

Based on the 3D genome driven cell compartmentalization theory that we discovered before (26, 43, 44), the intuition of incorporating 3D genome information in circRNA selection relies on the conjecture that circRNAs’ corresponding DNA loci obey certain distribution pattern that may contribute to better circRNA panel selection and eventually benefits AD prediction. To examine such conjecture, we sought to find out the correlation of circRNA expressions with their spatial loci. As described hereinabove, circRNAs with a Spearman correlation coefficient (SCC) p-value greater than 0.05 were removed; circRNAs from the same gene were considered as one kind and the circRNA with the highest SCC was selected as the representative of each kind. K-means clustering (K = 100) was performed based on the expression profiles, and 100 clusters were obtained with an average cluster size 134 circRNAs. We discovered that some co-expressed circRNAs belonging to the same expression cluster tend to be co-localized in 3D genome space, as demonstrated in Figure 2.

Figure 2. The spatial distribution of different circRNA clusters computed based on K-means clustering using circRNA expression profiles, under the eight Hi-C cell-line derived 3D models

 

Biomarker selection and machine learning models

We tested the predictive performance of circRNA on the disease using two methods. For the classification model, we defined individuals with CDR scores of 0 and 0.5 as non-affected patients and those with CDR scores of 4-5 as Alzheimer’s disease patients. We employed the XGBoost classification model to predict disease status. Initially, we performed density clustering on all circRNAs within each dataset, dividing them into as many classes as possible to represent each region on the 3D genome. The final clustering distance was set to 0.12, and circRNAs not included in any cluster were considered outliers. Subsequently, we conducted differential analysis on these circRNAs and selected the most differentially expressed circRNAs in each class (p<0.05). Consequently, we obtained 312, 311, and 471 representative circRNAs in the three brain regions, respectively. For prediction, we used all circRNAs, circRNAs including outliers, and the density-clustered representative circRNAs separately. The classification model’s AUC results showed improvement after density clustering in all three brain regions.
For regression models, the predicted feature is the patients’ CDR scores. We tested three regression methods under the leave-one-out cross validation: K Nearest Neighbors (KNN), Linear Regression Model (LM), Support Vector Regression (SVR). To evaluate the advantage of selecting fewer circRNAs as AD biomarkers, all 3748 circRNAs in the dataset were first used as biomarkers, and the correlations between the predicted CDR scores from four methods with the true CDR scores are all relatively low, with the highest SVR method only achieving a R = 0.554. We then computed the SCC between each circRNA’s expression profile over samples with the true CDR score and selected the circRNAs with p<0.05 and obtained a 1337 panel. Furthermore, we use k-means cluster k=100 based on circRNA expression and calculated the model importance to selected the circRNA with the top importance score in each cluster to form a 100 circRNA medium panel. The correlations between circRNAs and disease stages in the three models were all improved to a certain extent as shown in Table 1 and Figure 3(D). In the end, in order to incorporate the 3D genome information, we collected the 3D genome spatial coordinates of these 100 circRNAs and conducted a density-based clustering (DBSCAN) procedure (radius cutoff set to 1.3). DBSCAN grouped circRNAs into eight clusters and 50 circRNAs were left along as outliers (Figure 3(C)). From each DBSCAN cluster, a representative circRNA (highest model contribution score) was selected to form the final biomarker panel and feed to the regression models. The prediction results show that, by adopting 3D genome information in such way, the CDR prediction efficacies of each regression method can be further improved. Among the three regression models, the K-Nearest Neighbor method outperformed the other three methods, with the correlation coefficient reaching 0.708 (Table 1 and Figure 3). After applying the same method for circRNA selection, the models’ predictive correlations for the other two brain regions also showed improvement. These results suggest incorporating 3D genomic information contributes to significantly better AD predictors.

Table 1. The correlation coefficient between the predicted CDR scores and the true CDR scores using different circRNA panels under the three regression methods

 

Figure 3. A). XGBoost model performance under different circRNA panel. (B). Expression heatmap of circRNAs with p of SCC less than 0.05. (C). CircRNA distribution of three brain regions after density-based cluster. The central black dot represents the nucleus position (0,0,0). (D). The distribution of the predicted CDR scores from different regression models under the three different AD stages using only the 3D genome selected circRNA panel

 

The final screened circRNA signature

The occurrence of AD may be affected by epigenetic phenomena such as DNA methylation or certain noncoding RNA interferences (45), and miRNAs play an important role in the latter (46). We aimed to explore the potential functions of circRNAs selected after integrating the 3D genome data. Here, we summarized the circRNAs identified by the regression models that were present in at least two regions(Figure 4(A)) and predicted their associated miRNAs using CircInteractome (47). As illustrated in Figure 4(C), among all the inferenced results, hsa_circ_0103896, hsa_circ_0117770, hsa_circ_0109315, hsa_circ_0102742, hsa_circ_0007558 and hsa_circ_0103284 have binding sites for hsa-miR-136, which targets AEG-1 and Bcl-2 express anti-apoptotic proteins. The expression of hsa-miR-136 is down-regulated in glioma, which leads to the apoptosis of glial cells (48). Additionally, the results of the GO enrichment analysis indicate that these genes are associated with synaptic membrane and neuron spine in terms of cellular compartment, involved in synaptic signal transmission in biological processes, and associated with prominent structural composition in molecular function. This may also suggest that circRNAs generated by reverse splicing of these genes are indeed involved in the information transmission processes of the nervous system (Figure 4(B)).

Figure 4. (A) Distribution of circRNAs screened by density-based cluster in three brain regions. (B) Distribution of parental genes of circRNA screened by density-based cluster in three regions. (C) Interaction between circRNAs present in at least two brain regions and their potential target miRNAs. (D) The results of GO analysis on cellular compartment, biological processes and molecular function of parental genes of circRNA screened by density-based cluster in three brain regions

 

Discussion

It was previously demonstrated that circRNAs play an important regulatory role in AD. In addition to ciRS-7, which is the most well-studied, other circRNAs such as circ_0000950 serve as the sponge of miR-103, increasing the mRNA expression of the pro-inflammatory gene PTGS2, which ultimately leads to neuronal cell apoptosis and the inhibition of synaptic growth (49). In this study, in each brain region, circRNAs were identified by combining 3D genomic information and circRNA expression profiles. Using these circRNAs as predictors, we performed classification and regression analyses on AD disease status through machine learning methods including SVM, LM, KNN and XGBoost. The results demonstrated that the correlation between circRNAs and AD disease status increased after selecting fewer but more informative circRNAs. By conducting K-means clustering based on circRNA expression profiles and projected the results into 3D genome space, we discovered that there is a link between co-expression and co-localization of the circRNAs. Then, by incorporating the 3D genome information, we were able to obtain a small yet effective circRNA panel that can distinguish AD severeness the best. In the end, we predicted the miRNA binding sites of the circRNAs, and discovered that the target miRNAs predicted to bind to these circRNAs were previously reported to be linked to the nervous system malfunction. In particular, hsa_circ_0109315, arising from the reverse splicing of ZNF91, among the ultimately selected circRNAs across the three brain regions. Notably, it has the potential to target hsa-miR-136, implying its plausible involvement in modulating the activities of the nervous system.
We observed in our results that circRNAs of similar expression pattern tended to have a similar distribution in the 3D genome conformation, which contribute to the foundation of our conjecture in the past eight years, that from DNA (50) to DNA methylation (51), to RNA transcription (52), to protein-protein interaction (53), and to degenerated epitope levels (26), we believe there exist “3D genome driven cell compartmentalization” phenomena. We hypothesize that it is the nature of cellular compartmentalization that chained all the co-localization together, to improve cellular process more efficient. Our results illustrated that for circRNAs within the same cellular compartment/region, one could choose a representative circRNA to better overcome the circRNA redundancy issues. Such results indicate that by combining 3D genome information, the density-based clustering selected circRNA is the most effective discriminatory panel.
In each stage of Alzheimer’s disease, relative genes are under regulation of multiple sources, including DNA methylation, histone modification and noncoding RNAs (ncRNAs), etc. As far as epigenetics is concerned, its abnormality can directly or indirectly affect gene expression via miRNAs binding, and studies have also shown that circRNAs can act as miRNA upstream regulators that affect DNA methylation (54). In addition, studies have shown that circRNA can bind to the CpG island of the gene promoter region and directly affect DNA methylation (54). There are also evidences suggest that circRNA expression is dynamically regulated according to each stage of disease progression to exert epigenetic regulatory effects on its downstream targets relevant to disease pathomechanisms (55). Moreover, IRES-driven or m6A modification can mediate circRNA translation (15), and protein fragments abnormally translated from circRNAs may affect RNA processing and play a critical role in neural development. Due to the stability of circRNAs and their enrichment in exosomes (56), they have great potential to serve as AD biomarkers. The AD period and disease development trend of the patient can be inferred by detecting the expression level of circRNA. In addition, identifying the target miRNAs and related pathways of related circRNA and their pathogenesis will help to develop targeted drugs for this specific circRNA, which will contribute to the treatment of AD.
As the potential impact of this study on daily medical practice, we envision a transformative influence on the diagnostic behavior, particularly in cases characterized by diagnostic uncertainty. In the presence of a doubtful case, our proposed circRNA panel serves as a valuable tool for clinicians, offering a more nuanced and reliable approach to AD diagnosis. The intricate regulatory roles of circRNAs in neurodegenerative processes, as highlighted in our study, could provide crucial insights into the underlying molecular mechanisms of AD, aiding in a more accurate and timely diagnosis. We anticipate that our circRNA-based approach can be seamlessly integrated into the routine diagnostic workflow of clinical or hospital laboratories. The methodology outlined in our study involves the collection and curation of an AD-related circRNA dataset, coupled with an exploration of circRNAs’ corresponding DNA loci distribution in chromatin 3D conformation. Leveraging this 3D genome information, we have identified a concise yet predictively effective circRNA panel. The practicality of implementing this approach lies in its compatibility with standard laboratory procedures, making it feasible for adoption in clinical settings. In cases where diagnostic uncertainty persists, our circRNA panel could serve as a valuable adjunct to existing diagnostic tools, providing clinicians with an additional layer of molecular information to enhance diagnostic accuracy. We envision a scenario where our findings contribute to a more personalized and precise approach to AD diagnosis, ultimately improving patient outcomes. While further validation and clinical testing are essential steps, our study lays a foundation for the potential integration of circRNA-based diagnostics into the broader landscape of routine medical practice.

Key Points

• Differentially expressed circRNAs related to AD obey certain distribution pattern in 3D genome (high order chromatin conformation) space.
• Selecting circRNAs from various 3D genome regions constitutes a more predictive biomarker panel for AD.
• Machine learning models can be built based on 3D genome selected circRNAs to achieve significantly better AD prediction outcome than ordinary biomarker selection strategy.

 

Ethics approval and consent to participate: Not applicable.

Availability of data and materials: Data and materials are available up on reasonable request via corresponding authors.

Competing interests: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Funding: This project was supported by the Key Research and Development Plan of the Ministry of Science and Technology (2022YFE0125300), the National Key Research and Development Program (2016YFC0906400), the Innovation Funding in Shanghai (20JC1418600 and 18JC1413100), the National Natural Science Foundation of China (82071262 and 81671326), the Natural Science Foundation of Shanghai (20ZR1427200 and 20511101900), the Shanghai Municipal Science and Technology Major Project (2017SHZDZX01), the Shanghai Key Laboratory of Psychotic Disorders (13DZ2260500), the Shanghai Leading Academic Discipline Project (B205), the Shanghai Jiao Tong University STAR Grant (YG2023ZD26, YG2022ZD024, and YG2022QN111), and Shanghai Gaofeng & Gaoyuan Project for University Academic Program Development (Shanghai University of Sport-2023).

Authors’ contributions: RC, KL, KS and LL participated in the omics and computational experiments. RC and KL conducted the computational experiments. RC provided the figures, tables, and drafted the manuscript. RC, KL and LL designed chromatin modeling. JW, XL and GH provided clinical suggestions and biological insights about AD. YS and GH initiated this project and supervised the whole workflow. YS and GH edited and finalized the manuscript. All the authors reviewed and proof read the manuscript and the experimental results.

 

References

1. Sonkusare SK, Kaul CL, Ramarao P. Dementia of Alzheimer’s disease and other neurodegenerative disorders–memantine, a new hope. Pharmacol Res. 2005;51(1):1-17. https://doi.org/10.1016/j.phrs.2004.05.005
2. Abeysinghe A, Deshapriya R, Udawatte C. Alzheimer’s disease; a review of the pathophysiological basis and therapeutic interventions. Life Sci. 2020;256:117996. https://doi.org/10.1016/j.lfs.2020.117996
3. Breijyeh Z, Karaman R. Comprehensive Review on Alzheimer’s Disease: Causes and Treatment. Molecules. 2020;25(24). https://doi.org/10.3390/molecules25245789
4. Mattson MP. Pathways towards and away from Alzheimer’s disease. Nature. 2004;430(7000):631-9. https://doi.org/10.1038/nature02621
5. Wang J, Gu BJ, Masters CL, Wang YJ. A systemic view of Alzheimer disease – insights from amyloid-beta metabolism beyond the brain. Nat Rev Neurol. 2017;13(11):703. https://doi.org/10.1038/nrneurol.2017.111
6. Pereira C, Agostinho P, Moreira PI, Cardoso SM, Oliveira CR. Alzheimer’s disease-associated neurotoxic mechanisms and neuroprotective strategies. Curr Drug Targets CNS Neurol Disord. 2005;4(4):383-403. https://doi.org/10.2174/1568007054546117
7. Naseri NN, Wang H, Guo J, Sharma M, Luo W. The complexity of tau in Alzheimer’s disease. Neurosci Lett. 2019;705:183-94. https://doi.org/10.1016/j.neulet.2019.04.022
8. Cummings JL, Tong G, Ballard C. Treatment Combinations for Alzheimer’s Disease: Current and Future Pharmacotherapy Options. J Alzheimers Dis. 2019;67(3):779-94. https://doi.org/10.3233/JAD-180766
9. Sanger HL, Klotz G, Riesner D, Gross HJ, Kleinschmidt AK. Viroids Are Single-Stranded Covalently Closed Circular Rna Molecules Existing as Highly Base-Paired Rod-Like Structures. P Natl Acad Sci USA. 1976;73(11):3852-6. https://doi.org/10.1073/pnas.73.11.3852
10. You XT, Vlatkovic I, Babic A, Will T, Epstein I, Tushev G, et al. Neural circular RNAs are derived from synaptic genes and regulated by development and plasticity. Nat Neurosci. 2015;18(4):603-+.https://doi.org/10.1038/nn.3975
11. Rybak-Wolf A, Stottmeister C, Glazar P, Jens M, Pino N, Giusti S, et al. Circular RNAs in the Mammalian Brain Are Highly Abundant, Conserved, and Dynamically Expressed. Mol Cell. 2015;58(5):870-85. https://doi.org/10.1016/j.molcel.2015.03.027
12. Westholm JO, Miura P, Olson S, Shenker S, Joseph B, Sanfilippo P, et al. Genome-wide Analysis of Drosophila Circular RNAs Reveals Their Structural and Sequence Properties and Age-Dependent Neural Accumulation. Cell Rep. 2014;9(5):1966-80. https://doi.org/10.1016/j.celrep.2014.10.062
13. Gruner H, Cortes-Lopez M, Cooper DA, Bauer M, Miura P. CircRNA accumulation in the aging mouse brain. Sci Rep-Uk. 2016;6. https://doi.org/10.1038/srep38907
14. Hansen TB, Jensen TI, Clausen BH, Bramsen JB, Finsen B, Damgaard CK, et al. Natural RNA circles function as efficient microRNA sponges. Nature. 2013;495(7441):384-8. https://doi.org/10.1038/nature11993
15. Shi Y, Jia X, Xu J. The new function of circRNA: translation. Clin Transl Oncol. 2020;22(12):2162-9. https://doi.org/10.1007/s12094-020-02371-1
16. Akhter R. Circular RNA and Alzheimer’s Disease. Adv Exp Med Biol. 2018;1087:239-43. https://doi.org/10.1007/978-981-13-1426-1_19
17. Cogswell JP, Ward J, Taylor IA, Waters M, Shi YL, Cannon B, et al. Identification of miRNA changes in Alzheimer’s disease brain and CSF yields putative biomarkers and insights into disease pathways. J Alzheimers Dis. 2008;14(1):27-41. https://doi.org/10.3233/jad-2008-14103
18. Lukiw W, Zhao YH, Rogaev E, Bhattacharjee S. A Circular RNA (circRNA) ciRS-7 in Alzheimer’s disease (AD) targets miRNA-7 trafficking and promotes deficits in the expression of the ubiquitin conjugase (UBE2A) and the epidermal growth factor receptor (EGFR). Faseb J. 2016;30. https://doi.org/10.1096/fasebj.30.1_supplement.587.1
19. Starr DA, Fridolfsson HN. Interactions Between Nuclei and the Cytoskeleton Are Mediated by SUN-KASH Nuclear-Envelope Bridges. Annu Rev Cell Dev Bi. 2010;26:421-44. https://doi.org/10.1146/annurev-cellbio-100109-104037
20. Maniotis AJ, Chen CS, Ingber DE. Demonstration of mechanical connections between integrins cytoskeletal filaments, and nucleoplasm that stabilize nuclear structure. P Natl Acad Sci USA. 1997;94(3):849-54. https://doi.org/10.1073/pnas.94.3.849
21. Shin H, Shi Y, Dai C, Tjong H, Gong K, Alber F, et al. TopDom: an efficient and deterministic method for identifying topological domains in genomes. Nucleic Acids Res. 2016;44(7):e70. https://doi.org/10.1093/nar/gkv1505
22. Mohanta TK, Mishra AK, Al-Harrasi A. The 3D Genome: From Structure to Function. Int J Mol Sci. 2021;22(21). https://doi.org/10.3390/ijms222111585
23. Ubelmesser N, Papantonis A. Technologies to study spatial genome organization: beyond 3C. Brief Funct Genomics. 2019;18(6):395-401. https://doi.org/10.1093/bfgp/elz019
24. Lieberman-Aiden E, van Berkum NL, Williams L, Imakaev M, Ragoczy T, Telling A, et al. Comprehensive Mapping of Long-Range Interactions Reveals Folding Principles of the Human Genome. Science. 2009;326(5950):289-93. https://doi.org/10.1126/science.1181369
25. Kong SY, Zhang YB. Deciphering Hi-C: from 3D genome to function. Cell Biol Toxicol. 2019;35(1):15-32. https://doi.org/10.1007/s10565-018-09456-2
26. Shi Y, Guo ZH, Su XB, Meng LM, Zhang MX, Sun J, et al. DeepAntigen: a novel method for neoantigen prioritization via 3D genome and deep sparse learning. Bioinformatics. 2020;36(19):4894-901. https://doi.org/10.1093/bioinformatics/btaa596
27. Meng L, Wang C, Shi Y, Luo Q. Si-C is a method for inferring super-resolution intact genome structure from single-cell Hi-C data. Nat Commun. 2021;12(1):4369. https://doi.org/10.1038/s41467-021-24662-z
28. Yuan YC, Shi Y, Li CY, Kim JM, Cai WD, Han ZG, et al. DeepGene: an advanced cancer type classifier based on deep learning and somatic point mutations. Bmc Bioinformatics. 2016;17. https://doi.org/10.1186/s12859-016-1334-9
29. Wu YT, Zhang CJ, Mol B, Kawai A, Li C, Chen L, et al. Early Prediction of Gestational Diabetes Mellitus in the Chinese Population via Advanced Machine Learning. J Clin Endocr Metab. 2021;106(3):E1191-E205. https://doi.org/10.1210/clinem/dgaa899
30. Wang YF, Shi Y, Zhang CJ, Su KZ, Hu YX, Chen L, et al. Fetal weight estimation based on deep neural network: a retrospective observational study. Bmc Pregnancy Childb. 2023;23(1). https://doi.org/10.1186/s12884-023-05819-8
31. Yuan YC, Shi Y, Su XB, Zou X, Luo Q, Feng DD, et al. Cancer type prediction based on copy number aberration and chromatin 3D structure with convolutional neural networks. Bmc Genomics. 2018;19. https://doi.org/10.1186/s12864-018-4919-z
32. Wang MH, Beckmann ND, Roussos P, Wang EM, Zhou XX, Wang Q, et al. The Mount Sinai cohort of large-scale genomic, transcriptomic and proteomic data in Alzheimer’s disease. Sci Data. 2018;5. https://doi.org/10.1038/sdata.2018.185
33. Morris JC. Clinical dementia rating: a reliable and valid diagnostic and staging measure for dementia of the Alzheimer type. Int Psychogeriatr. 1997;9 Suppl 1:173-6; discussion 7-8. https://doi.org/10.1017/s1041610297004870
34. Zhang XO, Dong R, Zhang Y, Zhang JL, Luo Z, Zhang J, et al. Diverse alternative back-splicing and alternative splicing landscape of circular RNAs. Genome Res. 2016;26(9):1277-87. https://doi.org/10.1101/gr.202895.115
35. Kim D, Pertea G, Trapnell C, Pimentel H, Kelley R, Salzberg SL. TopHat2: accurate alignment of transcriptomes in the presence of insertions, deletions and gene fusions. Genome Biol. 2013;14(4). https://doi.org/10.1186/gb-2013-14-4-r36
36. Glazar P, Papavasileiou P, Rajewsky N. circBase: a database for circular RNAs. Rna. 2014;20(11):1666-70. https://doi.org/10.1261/rna.043687.113
37. Lo I, Hill J, Vilhjalmsson BJ, Kjems J. Linking the association between circRNAs and Alzheimer’s disease progression by multi-tissue circular RNA characterization. Rna Biol. 2020;17(12):1789-97. https://doi.org/10.1080/15476286.2020.1783487
38. Dixon JR, Selvaraj S, Yue F, Kim A, Li Y, Shen Y, et al. Topological domains in mammalian genomes identified by analysis of chromatin interactions. Nature. 2012;485(7398):376-80. https://doi.org/10.1038/nature11082
39. Rao SS, Huntley MH, Durand NC, Stamenova EK, Bochkov ID, Robinson JT, et al. A 3D map of the human genome at kilobase resolution reveals principles of chromatin looping. Cell. 2014;159(7):1665-80. https://doi.org/10.1016/j.cell.2014.11.021
40. Kuhn M. Building Predictive Models in R Using the caret Package. J Stat Softw. 2008;28(5):1-26. https://doi.org/10.18637/jss.v028.i05
41. Virtanen P, Gommers R, Oliphant TE, Haberland M, Reddy T, Cournapeau D, et al. SciPy 1.0: fundamental algorithms for scientific computing in Python (vol 33, pg 219, 2020). Nat Methods. 2020;17(3):352-.https://doi.org/10.1038/s41592-019-0686-2
42. Hahsler M, Piekenbrock M, Doran D. dbscan: Fast Density-Based Clustering with R. J Stat Softw. 2019;91(1):1-30. https://doi.org/10.18637/jss.v091.i01
43. Shi Y, Zhang MX, Meng LM, Su XN, Shang XY, Guo ZH, et al. A novel neoantigen discovery approach based on chromatin high order conformation. Bmc Med Genomics. 2020;13. https://doi.org/10.1186/s12920-020-0708-z
44. Akdemir KC, Le VT, Kim JM, Killcoyne S, King DA, Lin YP, et al. Somatic mutation distributions in cancer genomes vary with three-dimensional chromatin structure. Nat Genet. 2020;52(11):1178-+.https://doi.org/10.1038/s41588-020-0708-0
45. Perkovic MN, Paska AV, Konjevod M, Kouter K, Strac DS, Erjavec GN, et al. Epigenetics of Alzheimer’s Disease. Biomolecules. 2021;11(2). https://doi.org/10.3390/biom11020195
46. Szulwach KE, Li XK, Smrt RD, Li YJ, Luo YP, Lin L, et al. Cross talk between microRNA and epigenetic regulation in adult neurogenesis. J Cell Biol. 2010;189(1):127-U81. https://doi.org/10.1083/jcb.200908151
47. Dudekulay DB, Panda AC, Grammatikakis I, De S, Abdelmohsen K, Gorospe M. CircInteractome: A web tool for exploring circular RNAs and their interacting proteins and microRNAs. Rna Biol. 2016;13(1):34-42. https://doi.org/10.1080/15476286.2015.1128065
48. Yang Y, Wu JH, Guan HY, Cai JC, Fang LS, Li J, et al. MiR-136 promotes apoptosis of glioma cells by targeting AEG-1 and Bcl-2. Febs Lett. 2012;586(20):3608-12. https://doi.org/10.1016/j.febslet.2012.08.003
49. Yang H, Wang H, Shang H, Chen X, Yang S, Qu Y, et al. Circular RNA circ_0000950 promotes neuron apoptosis, suppresses neurite outgrowth and elevates inflammatory cytokines levels via directly sponging miR-103 in Alzheimer’s disease (vol 18, pg 2197, 2019). Cell Cycle. 2019;18(24):3603-.https://doi.org/10.1080/15384101.2019.1629773
50. Shi Y, Su XB, He KY, Wu BH, Zhang BY, Han ZG. Chromatin accessibility contributes to simultaneous mutations of cancer genes. Sci Rep-Uk. 2016;6. https://doi.org/10.1038/srep35270
51. Wu N, Yuan F, Yue SR, Jiang FY, Ren DC, Liu LJ, et al. Effect of exercise and diet intervention in NAFLD and NASH via GAB2 methylation. Cell Biosci. 2021;11(1). https://doi.org/10.1186/s13578-021-00701-6
52. Su XB, Zhao LN, Shi Y, Zhang R, Long Q, Bai SH, et al. Clonal evolution in liver cancer at single-cell and single-variant resolution. J Hematol Oncol. 2021;14(1). https://doi.org/10.1186/s13045-021-01036-y
53. Guo ZH, Liu LJ, Feng MF, Su K, Chi RQ, Li KY, et al. 3D genome assisted protein-protein interaction prediction. Future Gener Comp Sy. 2022;137:87-96. https://doi.org/10.1016/j.future.2022.07.005
54. Chen NF, Zhao G, Yan X, Lv Z, Yin HM, Zhang SL, et al. A novel FLI1 exonic circular RNA promotes metastasis in breast cancer by coordinately regulating TET1 and DNMT1. Genome Biol. 2018;19. https://doi.org/10.1186/s13059-018-1594-y
55. Lee WJ, Moon J, Jeon D, Shin YW, Yoo JS, Park DK, et al. Possible epigenetic regulatory effect of dysregulated circular RNAs in Alzheimer’s disease model. Sci Rep-Uk. 2019;9. https://doi.org/10.1038/s41598-019-48471-z
56. Li Y, Zheng QP, Bao CY, Li SY, Guo WJ, Zhao J, et al. Circular RNA is enriched and stable in exosomes: a promising biomarker for cancer diagnosis. Cell Res. 2015;25(8):981-4. https://doi.org/10.1038/cr.2015.82

© Serdi 2024