MOTIVATION: Variational autoencoders (VAEs) have rapidly increased in popularity in biological applications and have already successfully been used on many omic datasets. Their latent space provides a low-dimensional representation of input data, and VAEs have been applied, e.g. for clustering of single-cell transcriptomic data. However, due to their non-linear nature, the patterns that VAEs learn in the latent space remain obscure. Hence, the lower-dimensional data embedding cannot directly be related to input features. RESULTS: To shed light on the inner workings of VAE and enable direct interpretability of the model through its structure, we designed a novel VAE, OntoVAE (Ontology guided VAE) that can incorporate any ontology in its latent space and decoder part and, thus, provide pathway or phenotype activities for the ontology terms. In this work, we demonstrate that OntoVAE can be applied in the context of predictive modeling and show its ability to predict the effects of genetic or drug-induced perturbations using different ontologies and both, bulk and single-cell transcriptomic datasets. Finally, we provide a flexible framework, which can be easily adapted to any ontology and dataset. AVAILABILITY AND IMPLEMENTATION: OntoVAE is available as a python package under https://github.com/hdsu-bioquant/onto-vae.

MOTIVATION: In multi-cohort machine learning studies, it is critical to differentiate between effects that are reproducible across cohorts and those that are cohort-specific. Multi-task learning (MTL) is a machine learning approach that facilitates this differentiation through the simultaneous learning of prediction tasks across cohorts. Since multi-cohort data can often not be combined into a single storage solution, there would be the substantial utility of an MTL application for geographically distributed data sources. RESULTS: Here, we describe the development of 'dsMTL', a computational framework for privacy-preserving, distributed multi-task machine learning that includes three supervised and one unsupervised algorithms. First, we derive the theoretical properties of these methods and the relevant machine learning workflows to ensure the validity of the software implementation. Second, we implement dsMTL as a library for the R programming language, building on the DataSHIELD platform that supports the federated analysis of sensitive individual-level data. Third, we demonstrate the applicability of dsMTL for comorbidity modeling in distributed data. We show that comorbidity modeling using dsMTL outperformed conventional, federated machine learning, as well as the aggregation of multiple models built on the distributed datasets individually. The application of dsMTL was computationally efficient and highly scalable when applied to moderate-size (n < 500), real expression data given the actual network latency. AVAILABILITY AND IMPLEMENTATION: dsMTL is freely available at https://github.com/transbioZI/dsMTLBase (server-side package) and https://github.com/transbioZI/dsMTLClient (client-side package). SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Since the first whole-genome sequencing, the biomedical research community has made significant steps towards a more precise, predictive and personalized medicine. Genomic data is nowadays widely considered privacy-sensitive and consequently protected by strict regulations and released only after careful consideration. Various additional types of biomedical data, however, are not shielded by any dedicated legal means and consequently disseminated much less thoughtfully. This in particular holds true for DNA methylation data as one of the most important and well-understood epigenetic element influencing human health. In this paper, we show that, in contrast to the aforementioned belief, releasing one's DNA methylation data causes privacy issues akin to releasing one's actual genome. We show that already a small subset of methylation regions influenced by genomic variants are sufficient to infer parts of someone's genome, and to further map this DNA methylation profile to the corresponding genome. Notably, we show that such re-identification is possible with 97.5% accuracy, relying on a dataset of more than 2500 genomes, and that we can reject all wrongly matched genomes using an appropriate statistical test. We provide means for countering this threat by proposing a novel cryptographic scheme for privately classifying tumors that enables a privacy-respecting medical diagnosis in a common clinical setting. The scheme relies on a combination of random forests and homomorphic encryption, and it is proven secure in the honest-but-curious model. We evaluate this scheme on real DNA methylation data, and show that we can keep the computational overhead to acceptable values for our application scenario.

ChIP-seq has become a widely adopted genomic assay in recent years to determine binding sites for transcription factors or enrichments for specific histone modifications. Beside detection of enriched or bound regions, an important question is to determine differences between conditions. While this is a common analysis for gene expression, for which a large number of computational approaches have been validated, the same question for ChIP-seq is particularly challenging owing to the complexity of ChIP-seq data in terms of noisiness and variability. Many different tools have been developed and published in recent years. However, a comprehensive comparison and review of these tools is still missing. Here, we have reviewed 14 tools, which have been developed to determine differential enrichment between two conditions. They differ in their algorithmic setups, and also in the range of applicability. Hence, we have benchmarked these tools on real data sets for transcription factors and histone modifications, as well as on simulated data sets to quantitatively evaluate their performance. Overall, there is a great variety in the type of signal detected by these tools with a surprisingly low level of agreement. Depending on the type of analysis performed, the choice of method will crucially impact the outcome.

UNLABELLED: Supervised classification based on support vector machines (SVMs) has successfully been used for the prediction of cis-regulatory modules (CRMs). However, no integrated tool using such heterogeneous data as position-specific scoring matrices, ChIP-seq data or conservation scores is currently available. Here, we present LedPred, a flexible SVM workflow that predicts new regulatory sequences based on the annotation of known CRMs, which are associated to a large variety of feature types. LedPred is provided as an R/Bioconductor package connected to an online server to avoid installation of non-R software. Due to the heterogeneous CRM feature integration, LedPred excels at the prediction of regulatory sequences in Drosophila and mouse datasets compared with similar SVM-based software. AVAILABILITY AND IMPLEMENTATION: LedPred is available on GitHub: https://github.com/aitgon/LedPred and Bioconductor: http://bioconductor.org/packages/release/bioc/html/LedPred.html under the MIT license. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

UNLABELLED: We present a web-based service, SimCT, which allows to graphically display the relationships between biological objects (e.g. genes or proteins) based on their annotations to a biomedical ontology. The result is presented as a tree of these objects, which can be viewed and explored through a specific java applet designed to highlight relevant features. Unlike the numerous tools that search for overrepresented terms, SimCT draws a simplified representation of biological terms present in the set of objects, and can be applied to any ontology for which annotation data is available. Being web-based, it does not require prior installation, and provides an intuitive, easy-to-use service. AVAILABILITY: http://tagc.univ-mrs.fr/SimCT. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Understanding transcriptional regulation of gene expression is one of the greatest challenges of modern molecular biology. A central role in this mechanism is played by transcription factors, which typically bind to specific, short DNA sequence motifs usually located in the upstream region of the regulated genes. We discuss here a simple and powerful approach for the ab initio identification of these cis-regulatory motifs. The method we present integrates several elements: human-mouse comparison, statistical analysis of genomic sequences and the concept of coregulation. We apply it to a complete scan of the human genome. RESULTS: By using the catalogue of conserved upstream sequences collected in the CORG database we construct sets of genes sharing the same overrepresented motif (short DNA sequence) in their upstream regions both in human and in mouse. We perform this construction for all possible motifs from 5 to 8 nucleotides in length and then filter the resulting sets looking for two types of evidence of coregulation: first, we analyze the Gene Ontology annotation of the genes in the set, searching for statistically significant common annotations; second, we analyze the expression profiles of the genes in the set as measured by microarray experiments, searching for evidence of coexpression. The sets which pass one or both filters are conjectured to contain a significant fraction of coregulated genes, and the upstream motifs characterizing the sets are thus good candidates to be the binding sites of the TF's involved in such regulation. In this way we find various known motifs and also some new candidate binding sites. CONCLUSION: We have discussed a new integrated algorithm for the "ab initio" identification of transcription factor binding sites in the human genome. The method is based on three ingredients: comparative genomics, overrepresentation, different types of coregulation. The method is applied to a full-scan of the human genome, giving satisfactory results.

BACKGROUND: Developing reliable and efficient strategies allowing to infer a function to yet uncharacterized proteins based on interaction networks is of crucial interest in the current context of high-throughput data generation. In this paper, we develop a new algorithm for clustering vertices of a protein-protein interaction network using a density function, providing disjoint classes. RESULTS: Applied to the yeast interaction network, the classes obtained appear to be biological significant. The partitions are then used to make functional predictions for uncharacterized yeast proteins, using an annotation procedure that takes into account the binary interactions between proteins inside the classes. We show that this procedure is able to enhance the performances with respect to previous approaches. Finally, we propose a new annotation for 37 previously uncharacterized yeast proteins. CONCLUSION: We believe that our results represent a significant improvement for the inference of cellular functions, that can be applied to other organism as well as to other type of interaction graph, such as genetic interactions.