Single-cell multiplex chromatin and RNA interactions in ageing human brain.

Dynamically organized chromatin complexes often involve multiplex chromatin interactions and sometimes chromatin-associated RNA1-3. Chromatin complex compositions change during cellular differentiation and ageing, and are expected to be highly heterogeneous among terminally differentiated single cells4-7. Here we introduce the multinucleic acid interaction mapping in single cells (MUSIC) technique for concurrent profiling of multiplex chromatin interactions, gene expression and RNA-chromatin associations within individual nuclei. When applied to 14 human frontal cortex samples from older donors, MUSIC delineated diverse cortical cell types and states. We observed that nuclei exhibiting fewer short-range chromatin interactions were correlated with both an 'older' transcriptomic signature and Alzheimer's disease pathology. Furthermore, the cell type exhibiting chromatin contacts between cis expression quantitative trait loci and a promoter tends to be that in which these cis expression quantitative trait loci specifically affect the expression of their target gene. In addition, female cortical cells exhibit highly heterogeneous interactions between XIST non-coding RNA and chromosome X, along with diverse spatial organizations of the X chromosomes. MUSIC presents a potent tool for exploration of chromatin architecture and transcription at cellular resolution in complex tissues.

Single-cell multiplex chromatin and RNA interactions in aging human brain.

The dynamically organized chromatin complexes often involve multiplex chromatin interactions and sometimes chromatin-associated RNA (caRNA) 1-3. Chromatin complex compositions change during cellular differentiation and aging, and are expected to be highly heterogeneous among terminally differentiated single cells 4-7. Here we introduce the Multi-Nucleic Acid Interaction Mapping in Single Cell (MUSIC) technique for concurrent profiling of multiplex chromatin interactions, gene expression, and RNA-chromatin associations within individual nuclei. Applied to 14 human frontal cortex samples from elderly donors, MUSIC delineates diverse cortical cell types and states. We observed the nuclei exhibiting fewer short-range chromatin interactions are correlated with an "older" transcriptomic signature and with Alzheimer's pathology. Furthermore, the cell type exhibiting chromatin contacts between cis expression quantitative trait loci (cis eQTLs) and a promoter tends to be the cell type where these cis eQTLs specifically affect their target gene's expression. Additionally, the female cortical cells exhibit highly heterogeneous interactions between the XIST non-coding RNA and Chromosome X, along with diverse spatial organizations of the X chromosomes. MUSIC presents a potent tool for exploring chromatin architecture and transcription at cellular resolution in complex tissues.

CircIRAK3 exerts negative feedback regulation on inflammation by binding to HNRNP U and destabilizing proinflammatory cytokine mRNA in osteoarthritis and chondrogenesis.

Osteoarthritis (OA) is the most prevalent age-related and degenerative joint disease with limited treatment options. Previous studies have identified the therapeutic effects of mesenchymal stem cells (MSCs) therapy. Nevertheless, chronic inflammation impedes MSCs therapeutic effect. There have been reports suggesting that circular RNAs (circRNAs) are involved in OA and chondrogenesis. The combination of MSCs and circRNAs in therapies appears to be a promising option. In this study, we identified circIRAK3 as a significant regulator in cartilage degeneration and chondrogenesis through high-throughput sequencing analyses. We observed increased circIRAK3 in OA cartilage and during MSCs chondrogenesis. Knockdown of circIRAK3 resulted in excessive apoptosis, inhibited proliferation, and degradation of chondrocytes, along with the inhibition of MSCs chondrogenesis. Mechanistically, circIRAK3 bound to HNRNP U and competitively prevented its binding to IL-1ß, TNFα, and IL6 mRNA, thereby promoting mRNA degradation. Notably, circIRAK3 expression in plasma increased with higher OARSI scores. Intra-articular injection of adeno-associated virus-circIRAK3 delayed cartilage degeneration and reduced inflammation in DMM mouse model. Our study highlights a compensatory regulation network of circIRAK3 in chondrocytes in response to inflammation. CircIRAK3 has the potential to serve as a new therapeutic target for OA. Furthermore, therapies targeting circIRAK3 combined with MSCs hold promise.

Genome-wide analysis of the interplay between chromatin-associated RNA and 3D genome organization in human cells.

The interphase genome is dynamically organized in the nucleus and decorated with chromatin-associated RNA (caRNA). It remains unclear whether the genome architecture modulates the spatial distribution of caRNA and vice versa. Here, we generate a resource of genome-wide RNA-DNA and DNA-DNA contact maps in human cells. These maps reveal the chromosomal domains demarcated by locally transcribed RNA, hereafter termed RNA-defined chromosomal domains. Further, the spreading of caRNA is constrained by the boundaries of topologically associating domains (TADs), demonstrating the role of the 3D genome structure in modulating the spatial distribution of RNA. Conversely, stopping transcription or acute depletion of RNA induces thousands of chromatin loops genome-wide. Activation or suppression of the transcription of specific genes suppresses or creates chromatin loops straddling these genes. Deletion of a specific caRNA-producing genomic sequence promotes chromatin loops that straddle the interchromosomal target sequences of this caRNA. These data suggest a feedback loop where the 3D genome modulates the spatial distribution of RNA, which in turn affects the dynamic 3D genome organization.

FUNDC1/PFKP-mediated mitophagy induced by KD025 ameliorates cartilage degeneration in osteoarthritis.

Osteoarthritis (OA) is the most common joint disease, but no disease-modifying drugs have been approved for OA treatment. Mitophagy participates in mitochondrial homeostasis regulation by selectively clearing dysfunctional mitochondria, which might contribute to cartilage degeneration in OA. Here, we provide evidence of impaired mitophagy in OA chondrocytes, which exacerbates chondrocyte degeneration. Among the several classic mitophagy-regulating pathways and receptors, we found that FUNDC1 plays a key role in preserving chondrocyte homeostasis by inducing mitophagy. FUNDC1 knockdown in vitro and knockout in vivo decreased mitophagy and exacerbated mitochondrial dysfunction, exacerbating chondrocyte degeneration and OA progression. FUNDC1 overexpression via intra-articular injection of adeno-associated virus alleviated cartilage degeneration in OA. Mechanistically, our study demonstrated that PFKP interacts with and dephosphorylates FUNDC1 to induce mitophagy in chondrocytes. Further analysis identified KD025 as a candidate drug for restoring chondrocyte mitophagy by increasing the FUNDC1-PFKP interaction and thus alleviating cartilage degeneration in mice with DMM-induced OA. Our study highlights the role of the FUNDC1-PFKP interaction in chondrocyte homeostasis via mitophagy induction and identifies KD025 as a promising agent for treating OA by increasing chondrocyte mitophagy.

Alu transposable elements rewire enhancer-promoter network through RNA pairing.

A recent study by Liang et al.1 reveals that interacting enhancer RNAs (eRNAs) and promoter-transcribed upstream antisense RNAs (uaRNAs) can identify enhancer-promoter interactions. Complementary sequences within the interacting eRNAs and uaRNAs, predominantly Alu sequences, confer the specificity for eRNA-uaRNA pairing and hence enhancer-promoter recognition.

Silencing of NOTCH3 Signaling in Meniscus Smooth Muscle Cells Inhibits Fibrosis and Exacerbates Degeneration in a HEYL-Dependent Manner.

The mechanisms of meniscus fibrosis and novel ways to enhance fibrosis is unclear. This work reveals human meniscus fibrosis initiated at E24 weeks. Smooth muscle cell cluster is identified in embryonic meniscus, and the combined analysis with previous data suggests smooth muscle cell in embryonic meniscus as precursors of progenitor cells in the mature meniscus. NOTCH3 is constantly expressed in smooth muscle cells throughout embryogenesis to adulthood. Inhibition of NOTCH3 signaling in vivo inhibits meniscus fibrosis and exacerbates degeneration. Continuous histological sections show that HEYL, NOTCH3 downstream target gene, is expressed consistently with NOTCH3. HEYL knockdown in meniscus cells attenuated the COL1A1 upregulation by CTGF and TGF-ß stimulation. Thus, this study discovers the existence of smooth muscle cells and fibers in the meniscus. Inhibition of NOTCH3 signaling in meniscus smooth muscle cells in a HEYL-dependent manner prevented meniscus fibrosis and exacerbated degeneration. Therefore, NOTCH3/HEYL signaling might be a potential therapeutic target for meniscus fibrosis.

Higher Pelvic Incidence Was Associated with a Higher Risk of Sagittal Malposition of Femoral Component and Poor Outcomes of Primary Total Knee Arthroplasty: A Retrospective Cohort Analysis.

Spine-pelvis-lower extremity sagittal alignment is regarded as a global sagittal balance. Currently, there are few studies evaluating the pelvic and femoral sagittal alignment during total knee arthroplasty (TKA). This retrospective study aims to elucidate how pelvic and femoral sagittal alignment affect clinical outcomes of primary TKA for osteoarthritis (OA) and determine the proper range of femoral sagittal alignment. Patient-reported outcome measures (PROMs), including the Knee Society Score (KSS), Western Ontario and McMaster Universities (WOMAC), and patient satisfaction scores, and clinician-reported outcomes (CROs), including range of motion (ROM) and pelvic and femoral sagittal parameters, of 67 cases were evaluated (89 knees) before and 1 year after TKA. The angle between the distal femur anterior cortex line and flange of the femoral component (FC) was defined as the α angle. Correlations between the α angle and PROM and CRO were investigated using multivariate and secondary regression analyses. Patients were further divided into four cohorts (A, B, C, and D) according to the α angle, and comparisons of their postoperative PROM and ROM scores were performed. Postoperative PROM and ROM scores improved significantly compared with the preoperative scores (p < 0.01). Only the α angle was significantly associated with postoperative knee extension among all PROM and CRO indexes (p = 0.001). Secondary regression demonstrated a convex upward function, and the scores were the highest at α angles of 0.57, 0.96, and -1.42 degrees for postoperative KSS, satisfaction, and range of knee extension, respectively (p < 0.01). However, the concave upward degree was the lowest at an α angle of 0.33 degrees for pelvic incidence (p < 0.001). Bonferroni's paired comparisons indicated that postoperative KSS and satisfaction of the cohort B (0 degrees ≤ α angle ≤ 3 degrees) were better than those of other cohorts (p < 0.0125). The results indicate that surgeons should pay more attention to the sagittal alignment of FC in patients with increased pelvic incidence, the distal femoral anterior cortex is recommended as an anatomic landmark, and 0 to 3 degrees might be "safe zones" of the sagittal flexion of FC in TKA. This study reflects the level of evidence III.

Hyaluronan size alters chondrogenesis of mesenchymal stem cells cultured on tricalcium phosphate-collagen-hyaluronan scaffolds.

Hyaluronan (HA) provides a favorable environment for chondrogenesis of bone marrow mesenchymal stem cells (BMSCs). A previous report from our group indicated that addition of HA increases the chondro-inductive capacity of scaffolds. Therefore, this study aimed to investigate whether the Mw of the HA could affect chondrogenesis of BMSCs seeded on TCP-COL-HA scaffolds. Human BMSCs (hBMSCs) and rabbit BMSCs (rBMSCs) were isolated and expanded. TCP-COL scaffolds and TCP-COL-HA scaffolds with two different HA Mws were assessed for their capacity to induce cartilage regeneration from hBMSCs in vitro and in vivo. The results showed that about 96.96% of hBMSCs expressed CD44. Moreover, Hyal-1 and chondrogenic marker genes expressions were increased in hMSCs seeded on TCP-COL-HA scaffolds, and blocking the HA-CD44 interaction with an anti-CD44 antibody reduced the expression levels of Hyal-1 and chondrogenic marker genes. Additionally, TCP-COL-HA scaffolds with 2000 kDa Mw showed greater induction of BMSC chondrogenesis induction compared with those with 80 kDa Mw. Similar results were observed in an ectopic implantation nude mouse model. In a rabbit osteochondral defect repair model, rBMSCs seeded on TCP-COL-HA scaffolds with 2000 kDa Mw showed greater cartilage regeneration than those seeded with 80 kDa Mw. In addition, hBMSC-seeded TCP-COL-HA scaffolds with 2000 kDa Mw showed a significantly higher mechanical strength than those with 80 kDa Mw. Collectively, these results indicate that the Mw of HA could affect chondrogenesis of BMSCs seeded on TCP-COL-HA scaffolds. The TCP-COL-HA scaffolds might be used as allogenic off the shelf products in cartilage tissue engineering in future.

Functional organization of the maternal and paternal human 4D Nucleome.

Every human somatic cell inherits a maternal and a paternal genome, which work together to give rise to cellular phenotypes. However, the allele-specific relationship between gene expression and genome structure through the cell cycle is largely unknown. By integrating haplotype-resolved genome-wide chromosome conformation capture, mature and nascent mRNA, and protein binding data from a B lymphoblastoid cell line, we investigate this relationship both globally and locally. We introduce the maternal and paternal 4D Nucleome, enabling detailed analysis of the mechanisms and dynamics of genome structure and gene function for diploid organisms. Our analyses find significant coordination between allelic expression biases and local genome conformation, and notably absent expression bias in universally essential cell cycle and glycolysis genes. We propose a model in which coordinated biallelic expression reflects prioritized preservation of essential gene sets.

Whole-Transcriptome Sequence of Degenerative Meniscus Cells Unveiling Diagnostic Markers and Therapeutic Targets for Osteoarthritis.

Meniscus plays an important role in joint homeostasis. Tear or degeneration of meniscus might facilitate the process of knee osteoarthritis (OA). Hence, to investigate the transcriptome change during meniscus degeneration, we reveal the alterations of messenger RNA (mRNA), microRNA (miRNA), long noncoding RNA (lncRNA), and circular RNA (circRNA) in meniscus during OA by whole-transcriptome sequence. A total of 375 mRNAs, 15 miRNAs, 56 lncRNAs, and 90 circRNAs were significantly altered in the degenerative meniscus treated with interleukin-1ß (IL-1ß). More importantly, highly specific co-expression RNA (ceRNA) networks regulated by lncRNA LOC107986251-miR-212-5p-SESN3 and hsa_circ_0018069-miR-147b-3p-TJP2 were screened out during IL-induced meniscus degeneration, unveiling potential therapeutic targets for meniscus degeneration during the OA process. Furthermore, lipocalin-2 (LCN2) and RAB27B were identified as potential biomarkers in meniscus degeneration by overlapping three previously constructed databases of OA menisci. LCN2 and RAB27B were both upregulated in osteoarthritic menisci and IL-1ß-treated menisci and were highly associated with the severity of OA. This could introduce potential novel molecules into the database of clinical diagnostic biomarkers and possible therapeutic targets for early-stage OA treatment.

Revealing protein-protein interactions at the transcriptome scale by sequencing.

We describe PROPER-seq (protein-protein interaction sequencing) to map protein-protein interactions (PPIs) en masse. PROPER-seq first converts transcriptomes of input cells into RNA-barcoded protein libraries, in which all interacting protein pairs are captured through nucleotide barcode ligation, recorded as chimeric DNA sequences, and decoded at once by sequencing and mapping. We applied PROPER-seq to human embryonic kidney cells, T lymphocytes, and endothelial cells and identified 210,518 human PPIs (collected in the PROPER v.1.0 database). Among these, 1,365 and 2,480 PPIs are supported by published co-immunoprecipitation (coIP) and affinity purification-mass spectrometry (AP-MS) data, 17,638 PPIs are predicted by the prePPI algorithm without previous experimental validation, and 100 PPIs overlap human synthetic lethal gene pairs. In addition, four previously uncharacterized interaction partners with poly(ADP-ribose) polymerase 1 (PARP1) (a critical protein in DNA repair) known as XPO1, MATR3, IPO5, and LEO1 are validated in vivo. PROPER-seq presents a time-effective technology to map PPIs at the transcriptome scale, and PROPER v.1.0 provides a rich resource for studying PPIs.

Inferring single cell expression profiles from overlapped pooling sequencing data with compressed sensing strategy.

Though single cell RNA sequencing (scRNA-seq) technologies have been well developed, the acquisition of large-scale single cell expression data may still lead to high costs. Single cell expression profile has its inherent sparse properties, which makes it compressible, thus providing opportunities for solutions. Here, by computational simulation as well as experiment of 54 single cells, we propose that expression profiles can be compressed from the dimension of samples by overlapped assigning each cell into plenty of pools. And we prove that expression profiles can be inferred from these pool expression data with overlapped pooling design and compressed sensing strategy. We also show that by combining this approach with plate-based scRNA-seq measurement, it can maintain its superiorities in gene detection sensitivity and individual identity and recover the expression profile with high precision, while saving about half of the library cost. This method can inspire novel conceptions on the measurement, storage or computation improvements for other compressible signals in many biological areas.

Inhibition of miR-490-5p Promotes Human Adipose-Derived Stem Cells Chondrogenesis and Protects Chondrocytes via the PITPNM1/PI3K/AKT Axis.

MicroRNAs (miRNAs) play a pivotal role in cartilage development and homeostasis in osteoarthritis (OA). While the fundamental roles of miRNAs in cartilage degeneration have been extensively studied, their effects on chondrogenic differentiation induced by human adipose-derived stem cells (hADSCs) and the underlying mechanisms remain largely elusive. Here, we investigated the roles and mechanisms of miRNAs in hADSC chondrogenic differentiation and chondrocyte homeostasis. Using microarray analysis, we screened miRNAs expressed in the chondrogenic differentiated hADSCs and identified miR-490-5p as the most significantly down-regulated miRNA. We analyzed its expression patterns during chondrogenesis in vivo and in vitro. Our study showed that miR-490-5p overexpression promoted the transition of hADSCs from chondrogenesis to osteogenesis. In addition, based on miRNA-mRNA prediction analysis and dual-luciferase reporter assay, we proposed and proved that miR-490-5p targeted PITPNM1 by binding to its 3'-UTR and inhibiting its translation. Moreover, loss- and gain-of-function experiments identified the involvement of the PI3K/AKT signaling pathway, and a rescue experiment determined the effect and specific mechanism of the miR-490-5p/PITPNM1/PI3K/AKT axis in hADSC chondrogenic differentiation and chondrocyte homeostasis. Inhibition of miR-490-5p alleviated cartilage injury in vivo as demonstrated using the destabilization of the medial meniscus (DMM) OA model. We identified miR-490-5p as a novel modulator of hADSC-mediated chondrogenesis and chondrocyte phenotype. This study highlighted that miR-490-5p attenuated hADSC chondrogenesis and accelerated cartilage degradation through activation of the PI3K/AKT signaling pathway by targeting PITPNM1. Inhibition of miR-490-5p facilitated hADSC chondrogenic differentiation and protected chondrocyte phenotype via the PITPNM1/PI3K/AKT axis, thus providing a novel stem cell potential therapeutic target for OA treatment.

The Synovium Attenuates Cartilage Degeneration in KOA through Activation of the Smad2/3-Runx1 Cascade and Chondrogenesis-related miRNAs.

Knee osteoarthritis (KOA) is a highly prevalent disabling joint disease in aged people. Progressive cartilage degradation is the hallmark of KOA, but its deeper mechanism remains unclear. Substantial evidence indicates the importance of the synovium for joint homeostasis. The present study aimed to determine whether the synovium regulates cartilage metabolism through chondrogenesis-related microRNAs (miRNAs) in the KOA microenvironment. Clinical sample testing and in vitro cell experiments screened out miR-455 and miR-210 as effective miRNAs. The levels of both were significantly reduced in KOA cartilage but increased in KOA synovial fluid compared with controls. We further revealed that transforming growth factor ß1 (TGF-ß1) can significantly upregulate miR-455 and miR-210 expression in synoviocytes. The upregulated miRNAs can be secreted into the extracellular environment and prevent cartilage degeneration. Through bioinformatics and in vitro experiments, we found that Runx1 can bind to the promoter regions of miR-455 and miR-210 and enhance their transcription in TGF-ß1-treated synoviocytes. Collectively, our findings demonstrate a protective effect of the synovium against cartilage degeneration mediated by chondrogenesis-related miRNAs, which suggests that Runx1 is a potential target for KOA therapy.

MiR-455-3p reduces apoptosis and alleviates degeneration of chondrocyte through regulating PI3K/AKT pathway.

AIMS: This study aimed to explore the functions of miR-455-3p, PTEN, and PI3K/AKT pathway in osteoarthritis. MATERIALS AND METHODS: We used the human bone marrow stem cell (BMSC), healthy chondrocytes, osteoarthritis chondrocytes (OA), and the IL-1ß/TNF-α-treated chondrocyte model to explore the relationship between miR-455-3p and PTEN. Mimic or inhibitor was used to transfect chondrocytes to determine whether miR-455-3p can regulate PTEN and influence COL2A1 and MMP13. Apoptosis was detected by flow cytometry. A luciferase report was applied to verify the targeted binding. KO mice were applied to investigate PTEN and pAKT expression and the effect on chondrocytes in vivo. KEY FINDINGS: MiR-455-3p and PTEN were reverse in chondrogenesis and healthy cartilage versus OA cartilage. Similar trends were noted in IL-1ß model. PTEN and MMP13 decreased and COL2A1 increased after overexpressing miR-455-3p, whereas the inhibition showed opposite results. Flow cytometry showed that miR-455-3p could reduce the apoptosis of chondrocytes. The results of luciferase revealed that miR-455-3p could affect fluorescence activity of PTEN by targeting its 3'-UTR. Finally, we found a marked increased in the expression of PTEN in KO mice relative to WT mice, while pAKT levels decreased. SIGNIFICANCE: It can be supported that miR-455-3p can reduce the apoptosis of chondrocytes and alleviate OA through regulating PI3K/AKT pathway, which may be expected to be a target for the treatment of osteoarthritis.

Single-cell RNA-seq analysis identifies meniscus progenitors and reveals the progression of meniscus degeneration.

OBJECTIVES: The heterogeneity of meniscus cells and the mechanism of meniscus degeneration is not well understood. Here, single-cell RNA sequencing (scRNA-seq) was used to identify various meniscus cell subsets and investigate the mechanism of meniscus degeneration. METHODS: scRNA-seq was used to identify cell subsets and their gene signatures in healthy human and degenerated meniscus cells to determine their differentiation relationships and characterise the diversity within specific cell types. Colony-forming, multi-differentiation assays and a mice meniscus injury model were used to identify meniscus progenitor cells. We investigated the role of degenerated meniscus progenitor (DegP) cell clusters during meniscus degeneration using computational analysis and experimental verification. RESULTS: We identified seven clusters in healthy human meniscus, including five empirically defined populations and two novel populations. Pseudotime analysis showed endothelial cells and fibrochondrocyte progenitors (FCP) existed at the pseudospace trajectory start. Melanoma cell adhesion molecule ((MCAM)/CD146) was highly expressed in two clusters. CD146+ meniscus cells differentiated into osteoblasts and adipocytes and formed colonies. We identified changes in the proportions of degenerated meniscus cell clusters and found a cluster specific to degenerative meniscus with progenitor cell characteristics. The reconstruction of four progenitor cell clusters indicated that FCP differentiation into DegP was an aberrant process. Interleukin 1ß stimulation in healthy human meniscus cells increased CD318+ cells, while TGFß1 attenuated the increase in CD318+ cells in degenerated meniscus cells. CONCLUSIONS: The identification of meniscus progenitor cells provided new insights into cell-based meniscus tissue engineering, demonstrating a novel mechanism of meniscus degeneration, which contributes to the development of a novel therapeutic strategy.

MicroRNA-455-3p promotes TGF-ß signaling and inhibits osteoarthritis development by directly targeting PAK2.

MicroRNAs (miRNAs, miR) play a key role in the pathogenesis of osteoarthritis (OA). Few studies have examined the regulatory role of P21-activated kinases (PAKs), a family of serine/threonine kinases, in OA. The aim of this study was to determine whether miR-455-3p can regulate cartilage degeneration in OA by targeting PAK2. MiR-455-3p knockout mice showed significant degeneration of the knee cartilage. MiR-455-3p expression increased and PAK2 expression decreased in the late stage of human adipose-derived stem cell (hADSC) chondrogenesis and in chondrocytes affected by OA. Furthermore, in both miR-455-3p-overexpressing chondrocytes and PAK2-suppressing chondrocytes, cartilage-specific genes were upregulated, and hypertrophy-related genes were downregulated. A luciferase reporter assay confirmed that miR-455-3p regulates PAK2 expression by directly targeting the 3'-untranslated regions (3'UTRs) of PAK2 mRNA. IPA-3, a PAK inhibitor, inhibited cartilage degeneration due to OA. Moreover, suppressing PAK2 promoted R-Smad activation in the TGF/Smad signaling pathway in chondrocytes. Altogether, our results suggest that miR-455-3p promotes TGF-ß/Smad signaling in chondrocytes and inhibits cartilage degeneration by directly suppressing PAK2. These results thus indicate that miR-455-3p and PAK2 are novel potential therapeutic agents and targets, respectively, for the treatment of OA.

D-GPM: A Deep Learning Method for Gene Promoter Methylation Inference.

Whole-genome bisulfite sequencing generates a comprehensive profiling of the gene methylation levels, but is limited by a high cost. Recent studies have partitioned the genes into landmark genes and target genes and suggested that the landmark gene expression levels capture adequate information to reconstruct the target gene expression levels. This inspired us to propose that the methylation level of the promoters in landmark genes might be adequate to reconstruct the promoter methylation level of target genes, which would eventually reduce the cost of promoter methylation profiling. Here, we propose a deep learning model called Deep-Gene Promoter Methylation (D-GPM) to predict the whole-genome promoter methylation level based on the promoter methylation profile of the landmark genes from The Cancer Genome Atlas (TCGA). D-GPM-15%-7000 × 5, the optimal architecture of D-GPM, acquires the least overall mean absolute error (MAE) and the highest overall Pearson correlation coefficient (PCC), with values of 0.0329 and 0.8186, respectively, when testing data. Additionally, the D-GPM outperforms the regression tree (RT), linear regression (LR), and the support vector machine (SVM) in 95.66%, 92.65%, and 85.49% of the target genes by virtue of its relatively lower MAE and in 98.25%, 91.00%, and 81.56% of the target genes based on its relatively higher PCC, respectively. More importantly, the D-GPM predominates in predicting 79.86% and 78.34% of the target genes according to the model distribution of the least MAE and the highest PCC, respectively.