The recent development of deep learning methods have undoubtedly led to great improvement in various machine learning tasks, especially in prediction tasks. This type of methods have also been adapted to answer various problems in bioinformatics, including automatic genome annotation, artificial genome generation or phenotype prediction. In particular, a specific type of deep learning method, called graph neural network (GNN) has repeatedly been reported as a good candidate to predict phenotypes from gene expression because its ability to embed information on gene regulation or co-expression through the use of a gene network. However, up to date, no complete and reproducible benchmark has ever been performed to analyze the trade-off between cost and benefit of this approach compared to more standard (and simpler) machine learning methods. In this article, we provide such a benchmark, based on clear and comparable policies to evaluate the different methods on several datasets. Our conclusion is that GNN rarely provides a real improvement in prediction performance, especially when compared to the computation effort required by the methods. Our findings on a limited but controlled simulated dataset shows that this could be explained by the limited quality or predictive power of the input biological gene network itself.

Predicting molecular processes using deep learning is a promising approach to provide biological insights for non-coding single nucleotide polymorphisms identified in genome-wide association studies. However, most deep learning methods rely on supervised learning, which requires DNA sequences associated with functional data, and whose amount is severely limited by the finite size of the human genome. Conversely, the amount of mammalian DNA sequences is growing exponentially due to ongoing large-scale sequencing projects, but in most cases without functional data. To alleviate the limitations of supervised learning, we propose a novel semi-supervised learning (SSL) based on pseudo-labeling, which allows to exploit unlabeled DNA sequences from numerous genomes during model pre-training. We further improved it incorporating principles from the Noisy Student algorithm to predict the confidence in pseudo-labeled data used for pre-training, which showed improvements for transcription factor with very few binding (very small training data). The approach is very flexible and can be used to train any neural architecture including state-of-the-art models, and shows in most cases strong predictive performance improvements compared to standard supervised learning. Moreover, small models trained by SSL showed similar or better performance than large language model DNABERT2.

MOTIVATION: Genome-wide association studies have systematically identified thousands of single nucleotide polymorphisms (SNPs) associated with complex genetic diseases. However, the majority of those SNPs were found in non-coding genomic regions, preventing the understanding of the underlying causal mechanism. Predicting molecular processes based on the DNA sequence represents a promising approach to understand the role of those non-coding SNPs. Over the past years, deep learning was successfully applied to regulatory sequence prediction using supervised learning. Supervised learning required DNA sequences associated with functional data for training, whose amount is strongly limited by the finite size of the human genome. Conversely, the amount of mammalian DNA sequences is exponentially increasing due to ongoing large sequencing projects, but without functional data in most cases. RESULTS: To alleviate the limitations of supervised learning, we propose a paradigm shift with semi-supervised learning, which does not only exploit labeled sequences (e.g. human genome with ChIP-seq experiment), but also unlabeled sequences available in much larger amounts (e.g. from other species without ChIP-seq experiment, such as chimpanzee). Our approach is flexible and can be plugged into any neural architecture including shallow and deep networks, and shows strong predictive performance improvements compared to supervised learning in most cases (up to [Formula: see text]). AVAILABILITY AND IMPLEMENTATION: https://forgemia.inra.fr/raphael.mourad/deepgnn .

BACKGROUND/AIM: In higher eukaryotes, the three-dimensional (3D) organization of the genome is intimately related to numerous key biological functions including gene expression, DNA repair and DNA replication regulations. Alteration of 3D organization, in particular topologically associating domains (TADs), is detrimental to the organism and can give rise to a broad range of diseases such as cancers. METHODS: Here, we propose a versatile regression framework which not only identifies TADs in a fast and accurate manner, but also detects differential TAD borders across conditions for which few methods exist, and predicts 3D genome reorganization after chromosomal rearrangement. Moreover, the framework is biologically meaningful, has an intuitive interpretation and is easy to visualize. RESULT AND CONCLUSION: The novel regression ranks among top TAD callers. Moreover, it identifies new features of the genome we called TAD facilitators, and that are enriched with specific transcription factors. It also unveils the importance of cell-type specific transcription factors in establishing novel TAD borders during neuronal differentiation. Lastly, it compares favorably with the state-of-the-art method for predicting rearranged 3D genome.

DNA is a complex molecule carrying the instructions an organism needs to develop, live and reproduce. In 1953, Watson and Crick discovered that DNA is composed of two chains forming a double-helix. Later on, other structures of DNA were discovered and shown to play important roles in the cell, in particular G-quadruplex (G4). Following genome sequencing, several bioinformatic algorithms were developed to map G4s in vitro based on a canonical sequence motif, G-richness and G-skewness or alternatively sequence features including k-mers, and more recently machine/deep learning. Recently, new sequencing techniques were developed to map G4s in vitro (G4-seq) and G4s in vivo (G4 ChIP-seq) at few hundred base resolution. Here, we propose a novel convolutional neural network (DeepG4) to map cell-type specific active G4 regions (e.g. regions within which G4s form both in vitro and in vivo). DeepG4 is very accurate to predict active G4 regions in different cell types. Moreover, DeepG4 identifies key DNA motifs that are predictive of G4 region activity. We found that such motifs do not follow a very flexible sequence pattern as current algorithms seek for. Instead, active G4 regions are determined by numerous specific motifs. Moreover, among those motifs, we identified known transcription factors (TFs) which could play important roles in G4 activity by contributing either directly to G4 structures themselves or indirectly by participating in G4 formation in the vicinity. In addition, we used DeepG4 to predict active G4 regions in a large number of tissues and cancers, thereby providing a comprehensive resource for researchers. Availability: https://github.com/morphos30/DeepG4.

MOTIVATION: The three dimensions (3D) genome is essential to numerous key processes such as the regulation of gene expression and the replication-timing program. In vertebrates, chromatin looping is often mediated by CTCF, and marked by CTCF motif pairs in convergent orientation. Comparative high-throughput sequencing technique (Hi-C) recently revealed that chromatin looping evolves across species. However, Hi-C experiments are complex and costly, which currently limits their use for evolutionary studies over a large number of species. RESULTS: Here, we propose a novel approach to study the 3D genome evolution in vertebrates using the genomic sequence only, e.g. without the need for Hi-C data. The approach is simple and relies on comparing the distances between convergent and divergent CTCF motifs by computing a ratio we named the 3D ratio or '3DR'. We show that 3DR is a powerful statistic to detect CTCF looping encoded in the human genome sequence, thus reflecting strong evolutionary constraints encoded in DNA and associated with the 3D genome. When comparing vertebrate genomes, our results reveal that 3DR which underlies CTCF looping and topologically associating domain organization evolves over time and suggest that ancestral character reconstruction can be used to infer 3DR in ancestral genomes. AVAILABILITY AND IMPLEMENTATION: The R code is available at https://github.com/morphos30/PhyloCTCFLooping. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Chromosomal organization in 3D plays a central role in regulating cell-type specific transcriptional and DNA replication timing programs. Yet it remains unclear to what extent the resulting long-range contacts depend on specific molecular drivers. Here we propose a model that comprehensively assesses the influence on contacts of DNA-binding proteins, cis-regulatory elements and DNA consensus motifs. Using real data, we validate a large number of predictions for long-range contacts involving known architectural proteins and DNA motifs. Our model outperforms existing approaches including enrichment test, random forests and correlation, and it uncovers numerous novel long-range contacts in Drosophila and human. The model uncovers the orientation-dependent specificity for long-range contacts between CTCF motifs in Drosophila, highlighting its conserved property in 3D organization of metazoan genomes. Our model further unravels long-range contacts depending on co-factors recruited to DNA indirectly, as illustrated by the influence of cohesin in stabilizing long-range contacts between CTCF sites. It also reveals asymmetric contacts such as enhancer-promoter contacts that highlight opposite influences of the transcription factors EBF1, EGR1 or MEF2C depending on RNA Polymerase II pausing.

Recent advances in long-range Hi-C contact mapping have revealed the importance of the 3D structure of chromosomes in gene expression. A current challenge is to identify the key molecular drivers of this 3D structure. Several genomic features, such as architectural proteins and functional elements, were shown to be enriched at topological domain borders using classical enrichment tests. Here we propose multiple logistic regression to identify those genomic features that positively or negatively influence domain border establishment or maintenance. The model is flexible, and can account for statistical interactions among multiple genomic features. Using both simulated and real data, we show that our model outperforms enrichment test and non-parametric models, such as random forests, for the identification of genomic features that influence domain borders. Using Drosophila Hi-C data at a very high resolution of 1 kb, our model suggests that, among architectural proteins, BEAF-32 and CP190 are the main positive drivers of 3D domain borders. In humans, our model identifies well-known architectural proteins CTCF and cohesin, as well as ZNF143 and Polycomb group proteins as positive drivers of domain borders. The model also reveals the existence of several negative drivers that counteract the presence of domain borders including P300, RXRA, BCL11A and ELK1.

In data analysis, latent variables play a central role because they help provide powerful insights into a wide variety of phenomena, ranging from biological to human sciences. The latent tree model, a particular type of probabilistic graphical models, deserves attention. Its simple structure - a tree - allows simple and efficient inference, while its latent variables capture complex relationships. In the past decade, the latent tree model has been subject to significant theoretical and methodological developments. In this review, we propose a comprehensive study of this model. First we summarize key ideas underlying the model. Second we explain how it can be efficiently learned from data. Third we illustrate its use within three types of applications: latent structure discovery, multidimensional clustering, and probabilistic inference. Finally, we conclude and give promising directions for future researches in this field.

Probabilistic graphical models have been widely recognized as a powerful formalism in the bioinformatics field, especially in gene expression studies and linkage analysis. Although less well known in association genetics, many successful methods have recently emerged to dissect the genetic architecture of complex diseases. In this review article, we cover the applications of these models to the population association studies' context, such as linkage disequilibrium modeling, fine mapping and candidate gene studies, and genome-scale association studies. Significant breakthroughs of the corresponding methods are highlighted, but emphasis is also given to their current limitations, in particular, to the issue of scalability. Finally, we give promising directions for future research in this field.

BACKGROUND: Discovering the genetic basis of common genetic diseases in the human genome represents a public health issue. However, the dimensionality of the genetic data (up to 1 million genetic markers) and its complexity make the statistical analysis a challenging task. RESULTS: We present an accurate modeling of dependences between genetic markers, based on a forest of hierarchical latent class models which is a particular class of probabilistic graphical models. This model offers an adapted framework to deal with the fuzzy nature of linkage disequilibrium blocks. In addition, the data dimensionality can be reduced through the latent variables of the model which synthesize the information borne by genetic markers. In order to tackle the learning of both forest structure and probability distributions, a generic algorithm has been proposed. A first implementation of our algorithm has been shown to be tractable on benchmarks describing 105 variables for 2000 individuals. CONCLUSIONS: The forest of hierarchical latent class models offers several advantages for genome-wide association studies: accurate modeling of linkage disequilibrium, flexible data dimensionality reduction and biological meaning borne by latent variables.