Enhanced analysis of tabular data through Multi-representation DeepInsight.

Tabular data analysis is a critical task in various domains, enabling us to uncover valuable insights from structured datasets. While traditional machine learning methods can be used for feature engineering and dimensionality reduction, they often struggle to capture the intricate relationships and dependencies within real-world datasets. In this paper, we present Multi-representation DeepInsight (MRep-DeepInsight), a novel extension of the DeepInsight method designed to enhance the analysis of tabular data. By generating multiple representations of samples using diverse feature extraction techniques, our approach is able to capture a broader range of features and reveal deeper insights. We demonstrate the effectiveness of MRep-DeepInsight on single-cell datasets, Alzheimer's data, and artificial data, showcasing an improved accuracy over the original DeepInsight approach and machine learning methods like random forest, XGBoost, LightGBM, FT-Transformer and L2-regularized logistic regression. Our results highlight the value of incorporating multiple representations for robust and accurate tabular data analysis. By leveraging the power of diverse representations, MRep-DeepInsight offers a promising new avenue for advancing decision-making and scientific discovery across a wide range of fields.

Advances in AI and machine learning for predictive medicine.

The field of omics, driven by advances in high-throughput sequencing, faces a data explosion. This abundance of data offers unprecedented opportunities for predictive modeling in precision medicine, but also presents formidable challenges in data analysis and interpretation. Traditional machine learning (ML) techniques have been partly successful in generating predictive models for omics analysis but exhibit limitations in handling potential relationships within the data for more accurate prediction. This review explores a revolutionary shift in predictive modeling through the application of deep learning (DL), specifically convolutional neural networks (CNNs). Using transformation methods such as DeepInsight, omics data with independent variables in tabular (table-like, including vector) form can be turned into image-like representations, enabling CNNs to capture latent features effectively. This approach not only enhances predictive power but also leverages transfer learning, reducing computational time, and improving performance. However, integrating CNNs in predictive omics data analysis is not without challenges, including issues related to model interpretability, data heterogeneity, and data size. Addressing these challenges requires a multidisciplinary approach, involving collaborations between ML experts, bioinformatics researchers, biologists, and medical doctors. This review illuminates these complexities and charts a course for future research to unlock the full predictive potential of CNNs in omics data analysis and related fields.

scDeepInsight: a supervised cell-type identification method for scRNA-seq data with deep learning.

Annotation of cell-types is a critical step in the analysis of single-cell RNA sequencing (scRNA-seq) data that allows the study of heterogeneity across multiple cell populations. Currently, this is most commonly done using unsupervised clustering algorithms, which project single-cell expression data into a lower dimensional space and then cluster cells based on their distances from each other. However, as these methods do not use reference datasets, they can only achieve a rough classification of cell-types, and it is difficult to improve the recognition accuracy further. To effectively solve this issue, we propose a novel supervised annotation method, scDeepInsight. The scDeepInsight method is capable of performing manifold assignments. It is competent in executing data integration through batch normalization, performing supervised training on the reference dataset, doing outlier detection and annotating cell-types on query datasets. Moreover, it can help identify active genes or marker genes related to cell-types. The training of the scDeepInsight model is performed in a unique way. Tabular scRNA-seq data are first converted to corresponding images through the DeepInsight methodology. DeepInsight can create a trainable image transformer to convert non-image RNA data to images by comprehensively comparing interrelationships among multiple genes. Subsequently, the converted images are fed into convolutional neural networks such as EfficientNet-b3. This enables automatic feature extraction to identify the cell-types of scRNA-seq samples. We benchmarked scDeepInsight with six other mainstream cell annotation methods. The average accuracy rate of scDeepInsight reached 87.5%, which is more than 7% higher compared with the state-of-the-art methods.

Topologically associating domain underlies tissue specific expression of long intergenic non-coding RNAs.

Accumulating evidence indicates that long intergenic non-coding RNAs (lincRNAs) show more tissue-specific expression patterns than protein-coding genes (PCGs). However, although lincRNAs are subject to canonical transcriptional regulation like PCGs, the molecular basis for the specificity of their expression patterns remains unclear. Here, using expression data and coordinates of topologically associating domains (TADs) in human tissues, we show that lincRNA loci are significantly enriched in the more internal region of TADs compared to PCGs and that lincRNAs within TADs have higher tissue specificity than those outside TADs. Based on these, we propose an analytical framework to interpret transcriptional status using lincRNA as an indicator. We applied it to hypertrophic cardiomyopathy data and found disease-specific transcriptional regulation: ectopic expression of keratin at the TAD level and derepression of myocyte differentiation-related genes by E2F1 with down-regulation of LINC00881. Our results provide understanding of the function and regulation of lincRNAs according to genomic structure.

DeepInsight-3D architecture for anti-cancer drug response prediction with deep-learning on multi-omics.

Modern oncology offers a wide range of treatments and therefore choosing the best option for particular patient is very important for optimal outcome. Multi-omics profiling in combination with AI-based predictive models have great potential for streamlining these treatment decisions. However, these encouraging developments continue to be hampered by very high dimensionality of the datasets in combination with insufficiently large numbers of annotated samples. Here we proposed a novel deep learning-based method to predict patient-specific anticancer drug response from three types of multi-omics data. The proposed DeepInsight-3D approach relies on structured data-to-image conversion that then allows use of convolutional neural networks, which are particularly robust to high dimensionality of the inputs while retaining capabilities to model highly complex relationships between variables. Of particular note, we demonstrate that in this formalism additional channels of an image can be effectively used to accommodate data from different omics layers while implicitly encoding the connection between them. DeepInsight-3D was able to outperform other state-of-the-art methods applied to this task. The proposed improvements can facilitate the development of better personalized treatment strategies for different cancers in the future.

Association between high immune activity and worse prognosis in uveal melanoma and low-grade glioma in TCGA transcriptomic data.

BACKGROUND: Immune status in the tumor microenvironment is an important determinant of cancer progression and patient prognosis. Although a higher immune activity is often associated with a better prognosis, this trend is not absolute and differs across cancer types. We aimed to give insights into why some cancers do not show better survival despite higher immunity by assessing the relationship between different biological factors, including cytotoxicity, and patient prognosis in various cancer types using RNA-seq data collected by The Cancer Genome Atlas. RESULTS: Results showed that a higher immune activity was associated with worse overall survival in patients with uveal melanoma and low-grade glioma, which are cancers of immune-privileged sites. In these cancers, epithelial or endothelial mesenchymal transition and inflammatory state as well as immune activation had a notable negative correlation with patient survival. Further analysis using additional single-cell data of uveal melanoma and glioma revealed that epithelial or endothelial mesenchymal transition was mainly induced in retinal pigment cells or endothelial cells that comprise the blood-retinal and blood-brain barriers, which are unique structures of the eye and central nervous system, respectively. Inflammation was mainly promoted by macrophages, and their infiltration increased significantly in response to immune activation. Furthermore, we found the expression of inflammatory chemokines, particularly CCL5, was strongly correlated with immune activity and associated with poor survival, particularly in these cancers, suggesting that these inflammatory mediators are potential molecular targets for therapeutics. CONCLUSIONS: In uveal melanoma and low-grade glioma, inflammation from macrophages and epithelial or endothelial mesenchymal transition are particularly associated with a poor prognosis. This implies that they loosen the structures of the blood barrier and impair homeostasis and further recruit immune cells, which could result in a feedback loop of additional inflammatory effects leading to runaway conditions.

Immune subtypes and neoantigen-related immune evasion in advanced colorectal cancer.

Elimination of cancerous cells by the immune system is an important mechanism of protection from cancer, however, its effectiveness can be reduced owing to development of resistance and evasion. To understand the systemic immune response in advanced untreated primary colorectal cancer, we analyze immune subtypes and immune evasion via neoantigen-related mechanisms. We identify a distinctive cancer subtype characterized by immune evasion and very poor overall survival. This subtype has less clonal highly expressed neoantigens and high chromosomal instability, resulting in adaptive immune resistance mediated by the immune checkpoint molecules and neoantigen presentation disorders. We also observe that neoantigen depletion caused by immunoediting and high clonal neoantigen load are correlated with a good overall survival. Our results indicate that the status of the tumor microenvironment and neoantigen composition are promising new prognostic biomarkers with potential relevance for treatment plan decisions in advanced CRC.

DeepFeature: feature selection in nonimage data using convolutional neural network.

Artificial intelligence methods offer exciting new capabilities for the discovery of biological mechanisms from raw data because they are able to detect vastly more complex patterns of association that cannot be captured by classical statistical tests. Among these methods, deep neural networks are currently among the most advanced approaches and, in particular, convolutional neural networks (CNNs) have been shown to perform excellently for a variety of difficult tasks. Despite that applications of this type of networks to high-dimensional omics data and, most importantly, meaningful interpretation of the results returned from such models in a biomedical context remains an open problem. Here we present, an approach applying a CNN to nonimage data for feature selection. Our pipeline, DeepFeature, can both successfully transform omics data into a form that is optimal for fitting a CNN model and can also return sets of the most important genes used internally for computing predictions. Within the framework, the Snowfall compression algorithm is introduced to enable more elements in the fixed pixel framework, and region accumulation and element decoder is developed to find elements or genes from the class activation maps. In comparative tests for cancer type prediction task, DeepFeature simultaneously achieved superior predictive performance and better ability to discover key pathways and biological processes meaningful for this context. Capabilities offered by the proposed framework can enable the effective use of powerful deep learning methods to facilitate the discovery of causal mechanisms in high-dimensional biomedical data.

Prognosis prediction model for conversion from mild cognitive impairment to Alzheimer's disease created by integrative analysis of multi-omics data.

BACKGROUND: Mild cognitive impairment (MCI) is a precursor to Alzheimer's disease (AD), but not all MCI patients develop AD. Biomarkers for early detection of individuals at high risk for MCI-to-AD conversion are urgently required. METHODS: We used blood-based microRNA expression profiles and genomic data of 197 Japanese MCI patients to construct a prognosis prediction model based on a Cox proportional hazard model. We examined the biological significance of our findings with single nucleotide polymorphism-microRNA pairs (miR-eQTLs) by focusing on the target genes of the miRNAs. We investigated functional modules from the target genes with the occurrence of hub genes though a large-scale protein-protein interaction network analysis. We further examined the expression of the genes in 610 blood samples (271 ADs, 248 MCIs, and 91 cognitively normal elderly subjects [CNs]). RESULTS: The final prediction model, composed of 24 miR-eQTLs and three clinical factors (age, sex, and APOE4 alleles), successfully classified MCI patients into low and high risk of MCI-to-AD conversion (log-rank test P = 3.44 × 10-4 and achieved a concordance index of 0.702 on an independent test set. Four important hub genes associated with AD pathogenesis (SHC1, FOXO1, GSK3B, and PTEN) were identified in a network-based meta-analysis of miR-eQTL target genes. RNA-seq data from 610 blood samples showed statistically significant differences in PTEN expression between MCI and AD and in SHC1 expression between CN and AD (PTEN, P = 0.023; SHC1, P = 0.049). CONCLUSIONS: Our proposed model was demonstrated to be effective in MCI-to-AD conversion prediction. A network-based meta-analysis of miR-eQTL target genes identified important hub genes associated with AD pathogenesis. Accurate prediction of MCI-to-AD conversion would enable earlier intervention for MCI patients at high risk, potentially reducing conversion to AD.

Clinical usefulness of multigene screening with phenotype-driven bioinformatics analysis for the diagnosis of patients with monogenic diabetes or severe insulin resistance.

AIMS: Monogenic diabetes is clinically heterogeneous and differs from common forms of diabetes (type 1 and 2). We aimed to investigate the clinical usefulness of a comprehensive genetic testing system, comprised of targeted next-generation sequencing (NGS) with phenotype-driven bioinformatics analysis in patients with monogenic diabetes, which uses patient genotypic and phenotypic data to prioritize potentially causal variants. METHODS: We performed targeted NGS of 383 genes associated with monogenic diabetes or common forms of diabetes in 13 Japanese patients with suspected (n = 10) or previously diagnosed (n = 3) monogenic diabetes or severe insulin resistance. We performed in silico structural analysis and phenotype-driven bioinformatics analysis of candidate variants from NGS data. RESULTS: Among the patients suspected having monogenic diabetes or insulin resistance, we diagnosed 3 patients as subtypes of monogenic diabetes due to disease-associated variants of INSR, LMNA, and HNF1B. Additionally, in 3 other patients, we detected rare variants with potential phenotypic effects. Notably, we identified a novel missense variant in TBC1D4 and an MC4R variant, which together may cause a mixed phenotype of severe insulin resistance. CONCLUSIONS: This comprehensive approach could assist in the early diagnosis of patients with monogenic diabetes and facilitate the provision of tailored therapy.

Quantification of multicellular colonization in tumor metastasis using exome-sequencing data.

Metastasis is a major cause of cancer-related mortality, and it is essential to understand how metastasis occurs in order to overcome it. One relevant question is the origin of a metastatic tumor cell population. Although the hypothesis of a single-cell origin for metastasis from a primary tumor has long been prevalent, several recent studies using mouse models have supported a multicellular origin of metastasis. Human bulk whole-exome sequencing (WES) studies also have demonstrated a multiple "clonal" origin of metastasis, with different mutational compositions. Specifically, there has not yet been strong research to determine how many founder cells colonize a metastatic tumor. To address this question, under the metastatic model of "single bottleneck followed by rapid growth," we developed a method to quantify the "founder cell population size" in a metastasis using paired WES data from primary and metachronous metastatic tumors. Simulation studies demonstrated the proposed method gives unbiased results with sufficient accuracy in the range of realistic settings. Applying the proposed method to real WES data from four colorectal cancer patients, all samples supported a multicellular origin of metastasis and the founder size was quantified, ranging from 3 to 17 cells. Such a wide-range of founder sizes estimated by the proposed method suggests that there are large variations in genetic similarity between primary and metastatic tumors in the same subjects, which may explain the observed (dis)similarity of drug responses between tumors.

Analyses of non-coding somatic drivers in 2,658 cancer whole genomes.

The discovery of drivers of cancer has traditionally focused on protein-coding genes1-4. Here we present analyses of driver point mutations and structural variants in non-coding regions across 2,658 genomes from the Pan-Cancer Analysis of Whole Genomes (PCAWG) Consortium5 of the International Cancer Genome Consortium (ICGC) and The Cancer Genome Atlas (TCGA). For point mutations, we developed a statistically rigorous strategy for combining significance levels from multiple methods of driver discovery that overcomes the limitations of individual methods. For structural variants, we present two methods of driver discovery, and identify regions that are significantly affected by recurrent breakpoints and recurrent somatic juxtapositions. Our analyses confirm previously reported drivers6,7, raise doubts about others and identify novel candidates, including point mutations in the 5' region of TP53, in the 3' untranslated regions of NFKBIZ and TOB1, focal deletions in BRD4 and rearrangements in the loci of AKR1C genes. We show that although point mutations and structural variants that drive cancer are less frequent in non-coding genes and regulatory sequences than in protein-coding genes, additional examples of these drivers will be found as more cancer genomes become available.

A comparison of machine learning classifiers for dementia with Lewy bodies using miRNA expression data.

BACKGROUND: Dementia with Lewy bodies (DLB) is the second most common subtype of neurodegenerative dementia in humans following Alzheimer's disease (AD). Present clinical diagnosis of DLB has high specificity and low sensitivity and finding potential biomarkers of prodromal DLB is still challenging. MicroRNAs (miRNAs) have recently received a lot of attention as a source of novel biomarkers. METHODS: In this study, using serum miRNA expression of 478 Japanese individuals, we investigated potential miRNA biomarkers and constructed an optimal risk prediction model based on several machine learning methods: penalized regression, random forest, support vector machine, and gradient boosting decision tree. RESULTS: The final risk prediction model, constructed via a gradient boosting decision tree using 180 miRNAs and two clinical features, achieved an accuracy of 0.829 on an independent test set. We further predicted candidate target genes from the miRNAs. Gene set enrichment analysis of the miRNA target genes revealed 6 functional genes included in the DHA signaling pathway associated with DLB pathology. Two of them were further supported by gene-based association studies using a large number of single nucleotide polymorphism markers (BCL2L1: P = 0.012, PIK3R2: P = 0.021). CONCLUSIONS: Our proposed prediction model provides an effective tool for DLB classification. Also, a gene-based association test of rare variants revealed that BCL2L1 and PIK3R2 were statistically significantly associated with DLB.

DeepInsight: A methodology to transform a non-image data to an image for convolution neural network architecture.

It is critical, but difficult, to catch the small variation in genomic or other kinds of data that differentiates phenotypes or categories. A plethora of data is available, but the information from its genes or elements is spread over arbitrarily, making it challenging to extract relevant details for identification. However, an arrangement of similar genes into clusters makes these differences more accessible and allows for robust identification of hidden mechanisms (e.g. pathways) than dealing with elements individually. Here we propose, DeepInsight, which converts non-image samples into a well-organized image-form. Thereby, the power of convolution neural network (CNN), including GPU utilization, can be realized for non-image samples. Furthermore, DeepInsight enables feature extraction through the application of CNN for non-image samples to seize imperative information and shown promising results. To our knowledge, this is the first work to apply CNN simultaneously on different kinds of non-image datasets: RNA-seq, vowels, text, and artificial.

Risk prediction models for dementia constructed by supervised principal component analysis using miRNA expression data.

Alzheimer's disease (AD) is the most common subtype of dementia, followed by Vascular Dementia (VaD), and Dementia with Lewy Bodies (DLB). Recently, microRNAs (miRNAs) have received a lot of attention as the novel biomarkers for dementia. Here, using serum miRNA expression of 1,601 Japanese individuals, we investigated potential miRNA biomarkers and constructed risk prediction models, based on a supervised principal component analysis (PCA) logistic regression method, according to the subtype of dementia. The final risk prediction model achieved a high accuracy of 0.873 on a validation cohort in AD, when using 78 miRNAs: Accuracy = 0.836 with 86 miRNAs in VaD; Accuracy = 0.825 with 110 miRNAs in DLB. To our knowledge, this is the first report applying miRNA-based risk prediction models to a dementia prospective cohort. Our study demonstrates our models to be effective in prospective disease risk prediction, and with further improvement may contribute to practical clinical use in dementia.

An integrative machine learning approach for prediction of toxicity-related drug safety.

Recent trends in drug development have been marked by diminishing returns caused by the escalating costs and falling rates of new drug approval. Unacceptable drug toxicity is a substantial cause of drug failure during clinical trials and the leading cause of drug withdraws after release to the market. Computational methods capable of predicting these failures can reduce the waste of resources and time devoted to the investigation of compounds that ultimately fail. We propose an original machine learning method that leverages identity of drug targets and off-targets, functional impact score computed from Gene Ontology annotations, and biological network data to predict drug toxicity. We demonstrate that our method (TargeTox) can distinguish potentially idiosyncratically toxic drugs from safe drugs and is also suitable for speculative evaluation of different target sets to support the design of optimal low-toxicity combinations.

Integrated analysis of human genetic association study and mouse transcriptome suggests LBH and SHF genes as novel susceptible genes for amyloid-ß accumulation in Alzheimer's disease.

Alzheimer's disease (AD) is a common neurological disease that causes dementia in humans. Although the reports of associated pathological genes have been increasing, the molecular mechanism leading to the accumulation of amyloid-ß (Aß) in human brain is still not well understood. To identify novel genes that cause accumulation of Aß in AD patients, we conducted an integrative analysis by combining a human genetic association study and transcriptome analysis in mouse brain. First, we examined genome-wide gene expression levels in the hippocampus, comparing them to amyloid Aß level in mice with mixed genetic backgrounds. Next, based on a GWAS statistics obtained by a previous study with human AD subjects, we obtained gene-based statistics from the SNP-based statistics. We combined p values from the two types of analysis across orthologous gene pairs in human and mouse into one p value for each gene to evaluate AD susceptibility. As a result, we found five genes with significant p values in this integrated analysis among the 373 genes analyzed. We also examined the gene expression level of these five genes in the hippocampus of independent human AD cases and control subjects. Two genes, LBH and SHF, showed lower expression levels in AD cases than control subjects. This is consistent with the gene expression levels of both the genes in mouse which were negatively correlated with Aß accumulation. These results, obtained from the integrative approach, suggest that LBH and SHF are associated with the AD pathogenesis.

Publisher Correction: IMSindel: An accurate intermediate-size indel detection tool incorporating de novo assembly and gapped global-local alignment with split read analysis.

A correction to this article has been published and is linked from the HTML and PDF versions of this paper. The error has not been fixed in the paper.

A novel homozygous missense mutation in the SH3-binding motif of STAMBP causing microcephaly-capillary malformation syndrome.

Microcephaly-capillary malformation syndrome is a congenital and neurodevelopmental disorder caused by biallelic mutations in the STAMBP gene. Here we identify the novel homozygous mutation located in the SH3 binding motif of STAMBP (NM_006463.4) (c.707C>T: p.Ser236Phe) through whole-exome sequencing. The case patient was a 2-year-old boy showing severe global developmental delay, progressive microcephaly, refractory seizures, dysmorphic facial features, and multiple capillary malformations. Immunoblot analysis of patient-derived lymphoblastoid cell lines (LCLs) revealed a severe reduction in STAMBP expression, indicating that Ser236Phe induces protein instability. STAMBP interacts with the SH3 domain of STAM and transduces downstream signals from the Jaks-STAM complex. The substitution of Ser236Phe found in the case patient was located in the SH3-binding motif, and we propose the mutation may block STAM binding and subsequently induce STAMBP degradation. Contrary to previously reported STAMBP mutations, the Ser236Phe mutation did not lead to constitutive activation of the PI3K-AKT-mTOR pathway in patient-derived LCLs, as indicated by the expression of phosphorylated S6 ribosomal protein, suggesting that it is not the major pathomechanism underlying the disorder in this patient.