Benjamin Audit

Despite the considerable progress made in recent years thanks to the rise of long-read sequencing, high-resolution, single molecule (SM)-based replication profiling of large genomes has remained out of reach. Here, we present a replication labelling strategy consisting in repeated pulse-labelling of asynchronously growing cells with the thymidine analogue bromodeoxyuridine (BrdU) that greatly enhances the number of detectable replication tracks. We show that the ‘multipulse’ BrdU labelling protocol, in association with ForkML, a machine-learning method translating BrdU signals in nanopore reads into oriented and positioned replication forks, enables the establishment of a SM replication map of the entire human genome at kilobase resolution.

Understanding how gene regulatory networks respond to global cell perturbations remains a central challenge in systems biology and network inference. Modular Response Analysis (MRA) provides a mathematical framework to infer gene-to-gene directed connectivity graphs from perturbation experiments; however, classical MRA captures direct gene-to-gene influences, and does not explicitly account for global stimuli that simultaneously change the graph. Here, we introduce MRA ⁺ , an extension of MRA, that incorporates the effect of global perturbations into gene-to-gene graph inference. MRA ⁺ assumes a sequential experimental design in which targeted gene perturbations are followed by the application of a global stimulus, enabling the separation of connectivity changes from direct gene induction. The method estimates network connectivity under induced conditions and quantifies gene-specific induction strengths, which represent contributions to expression changes arising from mechanisms external to the inferred network. In the case of single-cell expression data, we present a bootstrap strategy to assess the robustness of inferred connectivity coefficients and propose a complementary criterion based on sign stability to interpret weak or non-significant estimates. Together, these developments provide a general framework for robust inference of gene connectivity graphs in the presence of global perturbations, applicable to diverse biological and experimental contexts.

I. We present an analytical framework for modeling eukaryotic DNA replication that, given experimental Replication Fork Directionality (RFD) data, enables Bayesian inference of origin number, activation delay and intrinsic timing ( λ _i ), (the mean replication time if each origin were isolated). By deriving closed-form expressions for RFD and Mean Replication Timing (MRT) under exponential and a specific Weibull firing-time distributions as functions of and ( λ _i ), we eliminate the need for stochastic simulations. These analytical results reveal that RFD, as a ratio of fork directions, is invariant under joint rescaling of intrinsic timing and fork speed; absolute intrinsic timing can nonetheless be inferred when fork speed is independently measured. We demonstrate that under exponential firing distribution for the origin, the observed efficiency ( E _i ), i.e. the probability for an origin to fire which accounts for nearby origin, is simply MRT( x _i ) /λ _i . The closed-form RFD expressions allow to use a Bayesian method that achieves 0.96–0.99 correlation with yeast RFD profiles and resolves ~780 origins in S. cerevisiae . Our framework identifies about 150 origins with biologically significant delays (≥ 3 minute), revealing regulated activation kinetics undetectable by existing methods. By quantifying how origin intrinsic timing and delays shape replication timing landscapes, this work confirms yeast as a paradigm organism for studying DNA replication control mechanisms.

Current approaches to mapping fork progression in the human genome suffer from drastically low throughput. Here, we introduce ForkML, a nanopore sequencing-based method automatically positioning thousands of individual fork velocities by tracking BrdU incorporation into replicating DNA after double pulse-labelling of asynchronous cells. ForkML recovers known human fork speed, accurately detects replication stress, and, crucially, connects replication dynamics to genomic and chromatin contexts, exposing fork slowdown in early-replicating transcribed regions.

We consider a nonlinear inverse problem where the unknown is assumed to be piecewise affine, which is motivated by an application in DNA replication analysis. Since traditional algorithmic and theoretical tools from linear inverse problems do not apply, we propose a novel formalism and computational approach to harness it. In the noiseless case, we establish sufficient identifiability conditions, and prove that the solution is the unique minimizer of a nonconvex optimization problem. The latter is specially challenging because of its multiple local minima. We propose an optimization algorithm that provably finds the global solution in the noiseless case and is shown to be numerically effective for noisy signals. When instantiated in a DNA replication analysis scenario, where the unknown is a so-called timing profile, the approach is shown to be more computationally effective than the state-of-the-art optimization methods by at least 30 orders of magnitude. Besides, it automatically recovers the full configuration of the DNA replication dynamics, which is crucial for DNA replication analysis and was not possible with previous methods.

Nucleosomes are fundamental units of chromatin in which a length of genomic DNA is wrapped around a histone octamer spool in a left-handed superhelix. Large-scale nucleosome maps show a wide distribution of DNA wrapping lengths, which in some cases are tens of base pairs (bp) shorter than the 147 bp canonical wrapping length observed in nucleosome crystal structures. Here, we develop a thermodynamic model that assumes a constant free energy cost of unwrapping a nucleosomal bp. Our model also incorporates linker DNA—short DNA segments between neighboring nucleosomes imposed by the folding of nucleosome arrays into chromatin fibers and other higher-order chromatin structures. We use this model to study nucleosome positioning and occupancy in the presence of nucleosome “breathing”—partial unwrapping and rewrapping of nucleosomal DNA due to interactions with the neighboring particles. We find that, as the unwrapping cost per bp and the chemical potential are varied, the nucleosome arrays are characterized by three distinct states, with low, intermediate, and high densities. The transition between the latter two states proceeds through an equiprobable state in which all nucleosome wrapping lengths are equally likely. We study the equiprobable state theoretically using a mean-field approach, obtaining an excellent agreement with numerical simulations. Finally, we use our model to reproduce S. cerevisiae nucleosome occupancy profiles observed in the vicinity of transcription start sites, as well as genome-wide distributions of nucleosome wrapping lengths. Overall, our results highlight the key role of partial nucleosome unwrapping in shaping the genome-wide patterns of nucleosome positioning and occupancy.

Current temporal studies of DNA replication are either low-resolution or require complex cell synchronisation and/or sorting procedures. Here we introduce Nanotiming, a single-molecule, nanopore sequencing-based method producing high-resolution, telomere-to-telomere replication timing (RT) profiles of eukaryotic genomes by interrogating changes in intracellular dTTP concentration during S phase through competition with its analogue bromodeoxyuridine triphosphate (BrdUTP) for incorporation into replicating DNA. This solely demands the labelling of asynchronously growing cells with an innocuous dose of BrdU during one doubling time followed by BrdU quantification along nanopore reads. We demonstrate in S. cerevisiae model eukaryote that Nanotiming reproduces RT profiles generated by reference methods both in wild-type and mutant cells inactivated for known RT determinants. Nanotiming is simple, accurate, inexpensive, amenable to large-scale analyses, and has the unique ability to access RT of individual telomeres, revealing that Rif1 iconic telomere regulator selectively delays replication of telomeres associated with specific subtelomeric elements.

Motivation: Many machine learning (ML) models developed to classify phenotype from gene expression data provide interpretations for their decisions, with the aim of understanding biological processes. For many models, including neural networks, interpretations are lists of genes ranked by their importance for the predictions, with top-ranked genes likely linked to the phenotype. In this article, we discuss the limitations of such approaches using integrated gradient, an explainability method developed for neural networks, as an example. Results: Experiments are performed on RNA sequencing data from public cancer databases. A collection of ML models, including multilayer perceptrons and graph neural networks, are trained to classify samples by cancer type. Gene rankings from integrated gradients are compared to genes highlighted by statistical feature selection methods such as DESeq2 and other learning methods measuring global feature contribution. Experiments show that a small set of top-ranked genes is sufficient to achieve good classification. However, similar performance is possible with lower-ranked genes, although larger sets are required. Moreover, significant differences in top-ranked genes, especially between statistical and learning methods, prevent a comprehensive biological understanding. In conclusion, while these methods identify pathology-specific biomarkers, the completeness of gene sets selected by explainability techniques for understanding biological processes remains uncertain. Availability and implementation: Python code and datasets are available at https://github.com/mbonto/XAI_in_genomics.

In this work, we focus on the challenging problem of designing an off-the-grid method for dictionaries involving both positional and scale shifts. To tackle this challenge, we introduce a novel algorithm inspired by the Sliding FrankWolfe approach. In our proposed algorithm, positions are treated as continuous variables, whereas scales are discretized. Such a strategy eliminates numerical instabilities inherent to the direct application of Sliding Frank-Wolfe. We successfully apply this algorithm to the study of DNA replication data.

Many machine learning models have been proposed to classify phenotypes from gene expression data. In addition to their good performance, these models can potentially provide some understanding of phenotypes by extracting explanations for their decisions. These explanations often take the form of a list of genes ranked in order of importance for the predictions, the highest-ranked genes being interpreted as linked to the phenotype. We discuss the biological and the methodological limitations of such explanations. Experiments are performed on several datasets gathering cancer and healthy tissue samples from the TCGA, GTEx and TARGET databases. A collection of machine learning models including logistic regression, multilayer perceptron, and graph neural network are trained to classify samples according to their cancer type. Gene rankings are obtained from explainability methods adapted to these models, and compared to the ones from classical statistical feature selection methods such as mutual information, DESeq2, and EdgeR. Interestingly, on simple tasks, we observe that the information learned by black-box neural networks is related to the notion of differential expression. In all cases, a small set containing the best-ranked genes is sufficient to achieve a good classification. However, these genes differ significantly between the methods and similar classification performance can be achieved with numerous lower ranked genes. In conclusion, although these methods enable the identification of biomarkers characteristic of certain pathologies, our results question the completeness of the selected gene sets and thus of explainability by the identification of the underlying biological processes.