MethylSPWNet and MethylCapsNet: Biologically Motivated Organization of DNAm Neural Networks, Inspired by Capsule Networks

DNA methylation (DNAm) alterations have been heavily implicated in carcinogenesis and the pathophysiology of diseases through upstream regulation of gene expression. DNAm deep-learning approaches are able to capture features associated with aging, cell type, and disease progression, but lack incorporation of prior biological knowledge. Here, we present modular, user-friendly deep-learning methodology and software, MethylCapsNet and MethylSPWNet, that group CpGs into biologically relevant capsules—such as gene promoter context, CpG island relationship, or user-defined groupings—and relate them to diagnostic and prognostic outcomes. We demonstrate these models’ utility on 3,897 individuals in the classification of central nervous system (CNS) tumors. MethylCapsNet and MethylSPWNet provide an opportunity to increase DNAm deep-learning analyses’ interpretability by enabling a flexible organization of DNAm data into biologically relevant capsules.

Biological knowledge-slanted random forest approach for the classification of calcified aortic valve stenosis

Background Calcific aortic valve stenosis (CAVS) is a fatal disease and there is no pharmacological treatment to prevent the progression of CAVS. This study aims to identify genes potentially implicated with CAVS in patients with congenital bicuspid aortic valve (BAV) and tricuspid aortic valve (TAV) in comparison with patients having normal valves, using a knowledge-slanted random forest (RF). Results This study implemented a knowledge-slanted random forest (RF) using information extracted from a protein-protein interactions network to rank genes in order to modify their selection probability to draw the candidate split-variables. A total of 15,191 genes were assessed in 19 valves with CAVS (BAV, n = 10; TAV, n = 9) and 8 normal valves. The performance of the model was evaluated using accuracy, sensitivity, and specificity to discriminate cases with CAVS. A comparison with conventional RF was also performed. The performance of this proposed approach reported improved accuracy in comparison with conventional RF to classify cases separately with BAV and TAV (Slanted RF: 59.3% versus 40.7%). When patients with BAV and TAV were grouped against patients with normal valves, the addition of prior biological information was not relevant with an accuracy of 92.6%. Conclusion The knowledge-slanted RF approach reflected prior biological knowledge, leading to better precision in distinguishing between cases with BAV, TAV, and normal valves. The results of this study suggest that the integration of biological knowledge can be useful during difficult classification tasks.

Biological Constraints Can Improve Prediction in Precision Oncology

Machine learning (ML) algorithms are used to build predictive models or classifiers for specific disease outcomes using transcriptomic data. However, some of these models show deteriorating performance when tested on unseen data which undermines their clinical utility. In this study, we show the importance of directly embedding prior biological knowledge into the classifier decision rules to build simple and interpretable gene signatures. We tested this in two important classification examples: a) progression in non-muscle invasive bladder cancer; and b) response to neoadjuvant chemotherapy (NACT) in triple-negative breast cancer (TNBC) using different ML algorithms. For each algorithm, we developed two sets of classifiers: agnostic, trained using either individual gene expression values or the corresponding pairwise ranks without biological consideration; and mechanistic, trained by restricting the search to a set of gene pairs capturing important biological relations. Both types were trained on the same training data and their performance was evaluated on unseen testing data using different methodologies and multiple evaluation metrics. Our analysis shows that mechanistic models outperform their agnostic counterparts when tested on independent data and show more consistency to their performance in the training with enhanced interpretability. These findings suggest that using biological constraints in the training process can yield more robust and interpretable gene signatures with high translational potential.

driveR: a novel method for prioritizing cancer driver genes using somatic genomics data

Background Cancer develops due to “driver” alterations. Numerous approaches exist for predicting cancer drivers from cohort-scale genomics data. However, methods for personalized analysis of driver genes are underdeveloped. In this study, we developed a novel personalized/batch analysis approach for driver gene prioritization utilizing somatic genomics data, called driveR. Results Combining genomics information and prior biological knowledge, driveR accurately prioritizes cancer driver genes via a multi-task learning model. Testing on 28 different datasets, this study demonstrates that driveR performs adequately, achieving a median AUC of 0.684 (range 0.651–0.861) on the 28 batch analysis test datasets, and a median AUC of 0.773 (range 0–1) on the 5157 personalized analysis test samples. Moreover, it outperforms existing approaches, achieving a significantly higher median AUC than all of MutSigCV (Wilcoxon rank-sum test p < 0.001), DriverNet (p < 0.001), OncodriveFML (p < 0.001) and MutPanning (p < 0.001) on batch analysis test datasets, and a significantly higher median AUC than DawnRank (p < 0.001) and PRODIGY (p < 0.001) on personalized analysis datasets. Conclusions This study demonstrates that the proposed method is an accurate and easy-to-utilize approach for prioritizing driver genes in cancer genomes in personalized or batch analyses. driveR is available on CRAN: https://cran.r-project.org/package=driveR.

Biological knowledge-slanted random forest approach for the classification of calcified aortic valve stenosis

Background: Calcific aortic valve stenosis (CAVS) is a fatal disease and there is no pharmacological treatment to prevent the progression of CAVS. This study aims to identify genes potentially implicated with CAVS in patients with congenital bicuspid aortic valve (BAV) and tricuspid aortic valve (TAV) in comparison with normal valves, using a knowledge-slanted random forest (RF). Results: This study implemented a knowledge-slanted random forest (RF) using information extracted from a protein-protein interactions network to rank genes in order to modify the selection probability of them to draw the candidate split-variables. A total of 1591 genes were assessed in 19 valves with CAVS (BAV, n=10; TAV, n=9) and 8 normal valves. The performance of the model was evaluated using accuracy, sensitivity, and specificity to discriminate cases with CAVS. A comparison with conventional RF was also performed. The performance of this proposed approach reported better accuracy in comparison with conventional RF to classify cases separately with BAV and TAV (Slanted RF: 59.3% versus 40.7%). When patients with BAV and TAV were grouped against patients with normal valves, the addition of prior biological information was not relevant with an accuracy of 92.6%.Conclusion: The knowledge-slanted RF approach reflected prior biological knowledge, leading to better precision in distinguishing between cases with BAV, TAV, and normal valves. The results of this study suggest that the integration of biological knowledge can be useful during difficult classification tasks.

Prior Biological Knowledge Improves Genomic Prediction of Growth-Related Traits in Arabidopsis thaliana

Prediction of growth-related complex traits is highly important for crop breeding. Photosynthesis efficiency and biomass are direct indicators of overall plant performance and therefore even minor improvements in these traits can result in significant breeding gains. Crop breeding for complex traits has been revolutionized by technological developments in genomics and phenomics. Capitalizing on the growing availability of genomics data, genome-wide marker-based prediction models allow for efficient selection of the best parents for the next generation without the need for phenotypic information. Until now such models mostly predict the phenotype directly from the genotype and fail to make use of relevant biological knowledge. It is an open question to what extent the use of such biological knowledge is beneficial for improving genomic prediction accuracy and reliability. In this study, we explored the use of publicly available biological information for genomic prediction of photosynthetic light use efficiency (ΦPSII) and projected leaf area (PLA) in Arabidopsis thaliana. To explore the use of various types of knowledge, we mapped genomic polymorphisms to Gene Ontology (GO) terms and transcriptomics-based gene clusters, and applied these in a Genomic Feature Best Linear Unbiased Predictor (GFBLUP) model, which is an extension to the traditional Genomic BLUP (GBLUP) benchmark. Our results suggest that incorporation of prior biological knowledge can improve genomic prediction accuracy for both ΦPSII and PLA. The improvement achieved depends on the trait, type of knowledge and trait heritability. Moreover, transcriptomics offers complementary evidence to the Gene Ontology for improvement when used to define functional groups of genes. In conclusion, prior knowledge about trait-specific groups of genes can be directly translated into improved genomic prediction.

driveR: A Novel Method for Prioritizing Cancer Driver Genes Using Somatic Genomics Data

Cancer develops due to “driver” alterations. Numerous approaches exist for predicting cancer drivers from cohort-scale genomic data. However, methods for personalized analysis of driver genes are underdeveloped.In this study, we developed a novel personalized/batch analysis approach for driver gene prioritization utilizing somatic genomic data, called driveR. Combining genomic information and prior biological knowledge, driveR accurately prioritizes cancer driver genes via a multi-task learning model.Testing on 28 different datasets, this study demonstrates that driveR performs adequately, outperforms existing approaches, and is an accurate and easy-to-utilize approach for prioritizing driver genes in cancer genomes. driveR is available on CRAN: https://cran.r-project.org/package=driveR.

PALMER: improving pathway annotation based on the biomedical literature mining with a constrained latent block model

Background In systems biology, it is of great interest to identify previously unreported associations between genes. Recently, biomedical literature has been considered as a valuable resource for this purpose. While classical clustering algorithms have popularly been used to investigate associations among genes, they are not tuned for the literature mining data and are also based on strong assumptions, which are often violated in this type of data. For example, these approaches often assume homogeneity and independence among observations. However, these assumptions are often violated due to both redundancies in functional descriptions and biological functions shared among genes. Latent block models can be alternatives in this case but they also often show suboptimal performances, especially when signals are weak. In addition, they do not allow to utilize valuable prior biological knowledge, such as those available in existing databases. Results In order to address these limitations, here we propose PALMER, a constrained latent block model that allows to identify indirect relationships among genes based on the biomedical literature mining data. By automatically associating relevant Gene Ontology terms, PALMER facilitates biological interpretation of novel findings without laborious downstream analyses. PALMER also allows researchers to utilize prior biological knowledge about known gene-pathway relationships to guide identification of gene–gene associations. We evaluated PALMER with simulation studies and applications to studies of pathway-modulating genes relevant to cancer signaling pathways, while utilizing biological pathway annotations available in the KEGG database as prior knowledge. Conclusions We showed that PALMER outperforms traditional latent block models and it provides reliable identification of novel gene–gene associations by utilizing prior biological knowledge, especially when signals are weak in the biomedical literature mining dataset. We believe that PALMER and its relevant user-friendly software will be powerful tools that can be used to improve existing pathway annotations and identify novel pathway-modulating genes.

Interpretable network-guided epistasis detection

Detecting epistatic interactions at the gene level is essential to understanding the biological mechanisms of complex diseases. Unfortunately, genome-wide interaction association studies (GWAIS) involve many statistical challenges that make such detection hard. We propose a multi-step protocol for epistasis detection along the edges of a gene-gene co-function network. Such an approach reduces the number of tests performed and provides interpretable interactions, while keeping type I error controlled. Yet, mapping gene-interactions into testable SNP-interaction hypotheses, as well as computing gene pair association scores from SNP pair ones, is not trivial. Here we compare three SNP-gene mappings (positional overlap, eQTL and proximity in 3D structure) and used the adaptive truncated product method to compute gene pair scores. This method is non-parametric, does not require a known null distribution, and is fast to compute. We apply multiple variants of this protocol to a GWAS inflammatory bowel disease (IBD) dataset. Different configurations produced different results, highlighting that various mechanisms are implicated in IBD, while at the same time, results overlapped with known disease biology. Importantly, the proposed pipeline also differs from a conventional approach were no network is used, showing the potential for additional discoveries when prior biological knowledge is incorporated into epistasis detection.

Investigation of Capsule-Inspired Neural Network Approaches for DNA Methylation

DNA methylation (DNAm) alterations are implicated with aging and diseases by regulating gene expression. DNAm deep-learning approaches can capture features associated with aging, cell type, and disease progression, but lack incorporation of prior biological knowledge. We present deep-learning software, MethylCapsNet and MethylSPWNet, that group CpGs into user-specified or predefined biologically relevant groupings (eg. gene promoter or CpG island context) related to diagnostic and prognostic outcomes. We train our models on a cohort (n=3,897) to classify central nervous system tumors and compare to existing approaches. Our methodology presents opportunities to increase interpretability of disease mechanisms through utilization of biologically relevant annotations.