LDEncoder

<sec> <title>BACKGROUND</title> Prostate cancer progression exhibits significant variability influenced by biological and racial factors. DNA methylation profiling has shown potential in early cancer detection, but its integration with machine learning across racially diverse populations remains limited. </sec> <sec> <title>OBJECTIVE</title> This study aims to develop a race-aware framework using DNA methylation data and a Multi-Layer Perceptron (MLP) model to classify prostate cancer stages into early (I–II) and late (III–IV) stages. </sec> <sec> <title>METHODS</title> Methylation and phenotype data from the TCGA-PRAD dataset were processed using Differentially Methylated Positions (DMP) analysis to identify CpG sites correlated with cancer stages. These features were further refined through Recursive Feature Elimination (RFE) and used to train MLP models. SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) were used to interpret the model and identify key DNA methylation features contributing to model predictions. </sec> <sec> <title>RESULTS</title> The best-performing model achieved ~95% accuracy and up to 99% AUC on the majority race (White) training data using 70 selected features. However, performance declined sharply on minority race groups, revealing the effects of sample imbalance and race-specific methylation patterns. Feature importance examination indicates strong patterns within certain CpG sites driving the models predictions. </sec> <sec> <title>CONCLUSIONS</title> We propose a race-aware MLP model for prostate cancer stage classification using DNA methylation data, optimized through DMP and RFE-based feature selection. SHAP and LIME confirmed the predictive relevance of selected CpG sites, supporting model transparency. Results highlight high performance within the White cohort but reveal poor generalization to minority groups, emphasizing the importance of race-specific modeling strategies. </sec>

Deep learning (DL) is a distinct class of machine learning that has achieved first-class performance in many fields of study. For epigenomics, the application of DL to assist physicians and scientists in human disease-relevant prediction tasks has been relatively unexplored until very recently. In this article, we critically review published studies that employed DL models to predict disease detection, subtype classification, and treatment responses, using epigenomic data. A comprehensive search on PubMed, Scopus, Web of Science, Google Scholar, and arXiv.org was performed following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. Among 1140 initially identified publications, we included 22 articles in our review. DNA methylation and RNA-sequencing data are most frequently used to train the predictive models. The reviewed models achieved a high accuracy ranged from 88.3% to 100.0% for disease detection tasks, from 69.5% to 97.8% for subtype classification tasks, and from 80.0% to 93.0% for treatment response prediction tasks. We generated a workflow to develop a predictive model that encompasses all steps from first defining human disease-related tasks to finally evaluating model performance. DL holds promise for transforming epigenomic big data into valuable knowledge that will enhance the development of translational epigenomics.

DNA methylation plays a critical role in breast cancer development. Previous studies have identified DNA methylation marks in white blood cells as promising biomarkers for breast cancer. However, these studies were limited by low statistical power and potential biases. Using a new methodology, we investigated DNA methylation marks for their associations with breast cancer risk. Statistical models were built to predict levels of DNA methylation marks using genetic data and DNA methylation data from HumanMethylation450 BeadChip from the Framingham Heart Study (n = 1595). The prediction models were validated using data from the Women's Health Initiative (n = 883). We applied these models to genomewide association study (GWAS) data of 122 977 breast cancer patients and 105 974 controls to evaluate if the genetically predicted DNA methylation levels at CpG sites (CpGs) are associated with breast cancer risk. All statistical tests were two-sided. Of the 62 938 CpG sites CpGs investigated, statistically significant associations with breast cancer risk were observed for 450 CpGs at a Bonferroni-corrected threshold of P less than 7.94 × 10-7, including 45 CpGs residing in 18 genomic regions, that have not previously been associated with breast cancer risk. Of the remaining 405 CpGs located within 500 kilobase flaking regions of 70 GWAS-identified breast cancer risk variants, the associations for 11 CpGs were independent of GWAS-identified variants. Integrative analyses of genetic, DNA methylation, and gene expression data found that 38 CpGs may affect breast cancer risk through regulating expression of 21 genes. Our new methodology can identify novel DNA methylation biomarkers for breast cancer risk and can be applied to other diseases.

Analyzing different omics data types independently is often too restrictive to allow for detection of subtle, but consistent, variations that are coherently supported based upon different assays. Integrating multi-omics data in one model can increase statistical power. However, designing such a model is challenging because different omics are measured at different levels. We developed the iNETgrate package (https://bioconductor.org/packages/iNETgrate/) that efficiently integrates transcriptome and DNA methylation data in a single gene network. Applying iNETgrate on five independent datasets improved prognostication compared to common clinical gold standards and a patient similarity network approach.

BackgroundWhile Alzheimer’s disease (AD) remains one of the most challenging diseases to tackle, genome-wide genetic/epigenetic studies reveal many disease-associated risk loci, which sheds new light onto disease heritability, provides novel insights to understand its underlying mechanism and potentially offers easily measurable biomarkers for early diagnosis and intervention.MethodsWe analyzed whole-genome DNA methylation data collected from peripheral blood in a cohort (n = 649) from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and compared the DNA methylation level at baseline among participants diagnosed with AD (n = 87), mild cognitive impairment (MCI, n = 175) and normal controls (n = 162), to identify differentially methylated regions (DMRs). We also leveraged up to 4 years of longitudinal DNA methylation data, sampled at approximately 1 year intervals to model alterations in methylation levels at DMRs to delineate methylation changes associated with aging and disease progression, by linear mixed-effects (LME) modeling for the unchanged diagnosis groups (AD, MCI and control, respectively) and U-shape testing for those with changed diagnosis (converters).ResultsWhen compared with controls, patients with MCI consistently displayed promoter hypomethylation at methylation QTL (mQTL) gene locus PM20D1. This promoter hypomethylation was even more prominent in patients with mild to moderate AD. This is in stark contrast with previously reported hypermethylation in hippocampal and frontal cortex brain tissues in patients with advanced-stage AD at this locus. From longitudinal data, we show that initial promoter hypomethylation of PM20D1 during MCI and early stage AD is reversed to eventual promoter hypermethylation in late stage AD, which helps to complete a fuller picture of methylation dynamics. We also confirm this observation in an independent cohort from the Religious Orders Study and Memory and Aging Project (ROSMAP) Study using DNA methylation and gene expression data from brain tissues as neuropathological staging (Braak score) advances.ConclusionsOur results confirm that PM20D1 is an mQTL in AD and demonstrate that it plays a dynamic role at different stages of the disease. Further in-depth study is thus warranted to fully decipher its role in the evolution of AD and potentially explore its utility as a blood-based biomarker for AD.

Machine Learning-Based Prediction of Survival Outcome in Lower Grade Gliomas With Combined Clinical and DNA Methylation Data

Background:Approximately one-third of breast cancer (BC) patients show poorer cognitive function (CF). Using DNA methylation (DNAm) data here we aimed to identify genes and biological pathways associated with CF in postmenopausal women with early-stage hormone receptor-positive (HR+) BC.Methods:Epigenome-wide association studies (EWAS) and differentially methylated region analyses were performed for each CF phenotype (seven objective domains and one subjective phenotype) using DNAm data from whole blood samples (n=109) taken at the time of enrollment.Results:When adjusting for age, verbal IQ scores, and global DNAm signature, cg10331779 near CTNND2 (p-value=9.65 × 10−9) and cg25906741 in MLIP (p-value=2.01 × 10−8) were associated with processing speed and subjective CF, respectively, while regions in/near SLC6A11, PRKG1/CSTF2T, and FAM3B for processing speed, and regions in/near PI4KB and SGCE/PEG10 for mental flexibility were differentially methylated. In addition, beta-estradiol was identified as a common upstream regulator for all the CF phenotypes, suggesting an essential role of estrogen in explaining variation in CF of HR+ BC patients.Conclusions:In our EWAS of 8 CF phenotypes, we found two epigenome-wide significant signals, one for processing speed and the other for subjective CF. We also found three differentially methylated regions associated with processing speed and two associated with mental flexibility.Clinical trial registration:www.clinicaltrials.gov identifier is NCT02793921.

We show improved accuracy in-silico of inference of gene-regulatory network (GRN) structure, resulting from the use of an epigenomic prior network. We demonstrate important use-cases of our proposed methodology by re-analyzing datasets from 12 different studies, including scRNA-seq, DNA methylation (DNAme), chromatin accessibility, and histone modification data. We find that DNAme data are very effective for inferring the epigenomic prior network, recapitulating known epigenomic network structure found previously from chromatin accessibility data. Furthermore, we find that inferring the epigenomic prior network from DNAme data reveals candidate TF cis-regulations for around eight times as many genes, when compared with chromatin accessibility data. When our proposed methodology is applied to real datasets from human embryonic development and from women at risk of breast cancer, we find patterns of differential cis-regulation that are in line with expectations under appropriate biological models, and that may be used to propose hypotheses about pre-cancerous epigenomic changes.

Ectopic NSD2 Remodels H3K36me2 and DNA Methylation to Promote Oncogenic Gene Expression in Multiple Myeloma

Manifold learning aims to map the original data from a high-dimensional space into a low-dimensional feature space with possible better discriminative structure. In this paper, we propose a supervised manifold learning approach called SubManifold Individuality LEarning (SMILE). In SMILE, the linear subspace derived from the principal component analysis based on data with the same label is named as "the individual subspace", while the linear subspace learned from all data is defined as "the global subspace". For each data sample, the aim of SMILE is to enlarge the diversity between its reconstructed data from individual subspace and global subspace, respectively, so that the intrinsic character of each class can be stimulated in the feature space. SMILE also utilizes the Laplacian matrix to restrict the local structure of data in the low-dimensional feature space in order to preserve the locality of the high-dimensional data. The proposed method is validated in appearance-based face recognition problem on some typical facial image databases via extracting discriminative features. Experimental results show that the proposed approach can obtain the discriminative structure of facial manifold and extract better features for face recognition than other counterparts approaches.

DNA methylation change has been useful for cancer biomarker discovery, classification, and potential treatment development. So far, existing methods use either differentially methylated CpG sites or combined CpG sites, namely differentially methylated regions, that can be mapped to genes. However, such methylation signal mapping has limitations. To address these limitations, in this study, we introduced a combinatorial framework using linear regression, differential expression, deep learning method for accurate biological interpretation of DNA methylation through integrating DNA methylation data and corresponding TCGA gene expression data. We demonstrated it for uterine cervical cancer. First, we pre-filtered outliers from the data set and then determined the predicted gene expression value from the pre-filtered methylation data through linear regression. We identified differentially expressed genes (DEGs) by Empirical Bayes test using . Then we applied a deep learning method, “nnet” to classify the cervical cancer label of those DEGs to determine all classification metrics including accuracy and area under curve (AUC) through 10-fold cross validation. We applied our approach to uterine cervical cancer DNA methylation dataset (NCBI accession ID: GSE30760, 27,578 features covering 63 tumor and 152 matched normal samples). After linear regression and differential expression analysis, we obtained 6287 DEGs with false discovery rate (FDR) . After performing deep learning analysis, we obtained average classification accuracy () of the uterine cervical cancerous labels. This performance is better than that of other peer methods. We performed in-degree and out-degree hub gene network analysis using . We reported five top in-degree genes (, , , and ) and five top out-degree genes (, , , and ). After that, we performed KEGG pathway and Gene Ontology enrichment analysis of DEGs using tool WebGestalt(WEB-based Gene SeT AnaLysis Toolkit). In summary, our proposed framework that integrated linear regression, differential expression, deep learning provides a robust approach to better interpret DNA methylation analysis and gene expression data in disease study.

BackgroundThe Cancer Genome Atlas (TCGA) offers a multilayered view of genomics and epigenomics data of many human cancer types. However, the retrieval of expression and methylation data from TCGA is a cumbersome and time-consuming task.ResultsWanderer is an intuitive Web tool allowing real time access and visualization of gene expression and DNA methylation profiles from TCGA. Given a gene query and selection of a TCGA dataset (e.g., colon adenocarcinomas), the Web resource provides the expression profile, at the single exon level, and the DNA methylation levels of HumanMethylation450 BeadChip loci inside or in the vicinity of the queried gene. Graphic and table outputs include individual and summary data as well as statistical tests, allowing the comparison of tumor and normal profiles and the exploration along the genomic locus and across tumor collections.ConclusionsWanderer offers a simple interface to straightforward access to TCGA data, amenable to experimentalists and clinicians without bioinformatics expertise. Wanderer may be accessed at http://maplab.cat/wanderer.

In the last decade, many approaches have been developed to solve one-class classification (OCC) problems for anomaly detection. Many of them rely on estimating the statistical distribution of the data, find hidden patterns, or remap the data in advantageous feature spaces. This kind of techniques usually needs some a priori knowledge of the data distribution (i.e., Gaussian) or the setting of some parameters to achieve good classification performance, making their use less effective when the data distribution is unknown. In this paper, we propose a novel blind anomaly detection for low dimensional feature spaces, that exploits the flexibility of the neural network (NN) structure to find the class boundaries without any information about the shape of the data distribution. To prove the generality of the solution, we tested many different class shapes, and we applied it to a structural health monitoring (SHM) case study. Without requiring the tuning of hyperparameters, the performance of the proposed algorithm overcomes that of some known approaches like principal component analysis (PCA), kernel principal component analysis (KPCA), Gaussian mixture model (GMM), and autoassociative neural network (ANN) in many cases, and performs well in the specific SHM setting.

Purpose: Current methods of respiratory motion prediction treat each dimension independently. In this study we develop and investigate a low dimensional inference methodology that takes advantage of the correlative relation among various physical dimensions for respiratory motion prediction. Method and Materials: Covariate‐response training pairs are constructed in the original 3D physical space, via state augmentation. Unlike existing methods where independent inference models are built for each of the x,y,z coordinates, we first map these high dimensional paired samples into a lower dimensional feature space, spanned by the principle components. In this feature space, a Gaussian kernel with block‐diagonal covariance is employed for learning the joint covariate‐response pdf, which upon marginalization yields the distribution for the projected prediction value. Finally, the estimate in the feature space is mapped back into the original physical space to recover the prediction representation in the (x,y,z) coordinate. The proposed method is compared against its counterpart with independent prediction, using 159 lung motion traces from the Synchrony respiratory tracking system. Both methods were tested for predicting 160ms and 570ms, spanning the range of measured DMLC tracing system latencies. Performance is quantified with 3D RMSE and 90% percentile 3D error. Results: Under fair setup conditions, the proposed method achieves better performance than its independent counterpart. The case‐wise 3D RMSE was reduced by 30–40%, with p‐values in the order of 10̂(−14) for both lookahead lengths. The 90% percentile 3D error was reduced from 1.8mm to 1.1mm for 160ms prediction and from 2.8mm to 2.0mm for 570ms prediction. Conclusion: We have proposed a general low dimensional feature learning methodology to perform correlated high dimensional inference. Combining it with the already effective kernel density estimation yields a highly efficient approach for real‐time respiratory motion prediction. Research supported by NIH/NCI R01 93626, Varian Medical Systems, and AAPM Research Seed Funding initiative.

In multi-label learning, each object belongs to multiple class labels simultaneously. In the data explosion age, the size of data is often huge, i.e., large number of instances, features and class labels. The high dimension of both the feature and label spaces has posed great challenges to multi-label learning problems, e.g., high time and memory costs. In this paper, we propose a new framework for multi-label learning with a large number of class labels and features, i.e., M ulti- L abel L earning via F eature and L abel S pace D imension R eduction, namely MLL-FLSDR. Specifically, both the feature space and label space are reduced to low dimensional spaces respectively, in which the local structure of data points is utilized to constrain the geometrical structure on both the learned low dimensional spaces and guarantee the qualities of them. Then, an effective multi-label classifier is constructed from the low dimensional feature space to the latent label space. Last, the final prediction for new test data examples can be obtained by recovering from their prediction results in the latent label space with an encoding matrix learned in the previous stage. Extensive comparison experiments with the state-of-the-art approaches manifest the effectiveness of the proposed method MLL-FLSDR.