Talks

Causal reasoning is vital for effective reasoning in many domains, from healthcare to economics. In medical diagnosis, for example, a doctor aims to explain a patient’s symptoms by determining the diseases causing them. This is because causal relations, unlike correlations, allow one to reason about the consequences of possible treatments and to answer counterfactual queries. In this talk I will present some recent work done with my collaborators about how one can learn and reason with counterfactual distributions, and why this is importantly for decision making. In all cases I will strive to motivate and contextualise the results with real word examples.

It is not explicitly obvious that relativity and quantum mechanics are consistent with each other. Extensive research has shown that quantum states are consistent with relativity, in that they do not allow for faster-than-light transferring of information. In contrast, much less research has been done in quantum measurements, and in fact, naive attempts to put together relativity and quantum measurements lead to signaling between space-like separated regions. In this talk I will describe how this same problem arises in non-relativistic quantum physics, where measurements on systems kept spatially separated in general lead to signalling. By giving away the projection postulate, it is possible to alleviate this problem and measure non-local variables without signaling by exploiting pre-shared entanglement as a resource. I will describe a protocol for implementing any joint measurement in a non-signaling manner, and argue that this leads to a complete classification of all joint quantum measurements, based on the required amount of entanglement necessary to measure them.

We propose that Bell correlations are explicable as a combination of (i) collider bias and (ii) a boundary constraint on the collider variable. We show that the proposal is valid for a special class of ('W-shaped') Bell experiments involving delayed-choice entanglement swapping, and argue that it can be extended to the ordinary ('V-shaped') case. The proposal requires no direct causal influence outside lightcones, and may hence offer a way to reconcile Bell nonlocality and relativity.

Given the large number of proposed quantum machine learning (QML) algorithms, it is somewhat surprising that ideas from this field have not yet been extended to causal learning. While deep learning and generative machine learning models have taken centre stage in the industrial application of automated learning on classical data, it is nonetheless well known that these techniques don't reliably capture causal concepts, leading to significant performance vulnerabilities. Increasingly, classical ML experts are taking ideas from causal inference, a field traditionally limited to small data sets of low dimensionality, and injecting modern ML elements to create new algorithms that benefit from the best of both worlds. These hybrid classical approaches provide new opportunity to search for potential quantum advantage. In this talk I explore this new research direction and propose several new quantum algorithms for classical causal inference.

In eukaryotic cells, the nucleosome core particle (NCP) forms the basic unit of the genetic architecture. In NCPs, genomic DNA is tightly wound around an octameric core of histone proteins, much like thread wrapped around a spool. Recent experiments have shown that nucleosomes are highly dynamic, and often unwrap in an asymmetrical fashion at high ionic strengths, or in response to mechanical perturbations. By developing a sequence-specific coarse- grained model for DNA-protein complexes, which recapitulates various aspects of nucleosome structure and dynamics, we show how sequence-specificity of DNA-protein interactions is critical for nucleosome plasticity. Our force-field also captures the two-stage unwrapping observed in single-molecule pulling experiments. We further show that histone tails, which are hotspots for post-translational modifications, play a remarkable role in modulating the extent, as well as the direction of unwrapping. Our observations could set the stage for under...

Human silencers have been shown to exist and regulate developmental gene expression. However, the functional importance of human silencers needs to be elucidated, such as whether they can form “super-silencers” and whether they are linked to cancer progression. Here, through interrogating two putative silencer components of FGF18 gene, we found that two nearby silencers can cooperate via compensatory chromatin interactions to form a “super-silencer”. Furthermore, double knockout of two silencers exhibited synergistic upregulation of FGF18 expression and changes of cell identity. To perturb the “super-silencers”, we applied combinational treatment of an EZH2 inhibitor GSK343, and a REST inhibitor, X5050 (“GR”). We found that GR led to severe loss of TADs and loops, while the use of one inhibitor by itself only showed mild changes. Such changes in TADs and loops were associated with reduced CTCF and TOP2A mRNA levels. Moreover, GSK343 and X5050 synergistically upregulated...

The role of telomeres in cellular and organismal physiology including ageing and cancer is commonly appreciated. However, intriguingly, the molecular impact of telomeres in human cells has been largely limited to the subtelomeres (~10 Mb from telomeres). Questioning this paradigm we found telomere dependent molecular mechanisms affect chromatin across the genome defining functional outcomes ranging from tumor cell immunity to neurogenesis.

We investigate the role of compaction of chromatin domains in modulating search kinetics of proteins. Collapsed conformations of chromatin, characterised by long loops which bring distant regions of the genome into contact, and manifested structurally as Topologically Associated Domains (TADs) affect search kinetics of DNA associated transcription factors and other proteins. In this study, we investigate the role of the compactness of chromatin on the dynamics of proteins using a minimal model. Using analytical theory and simulations, we show that an optimal compaction exists for which the residence time of proteins on a chromatin-like polymer backbone is minimum. We show that while bulk diffusion is an advantageous search strategy for extended polymers, for highly folded polymer domains, intersegmental transfers allow optimal search. We extend these results to more detailed polymer models - using the Freely Rotating Chain model, a Lennard-Jones bead-spring polymer model, which approxi...

The three-dimensional organization of chromatin into domains and compartments leads to specific scaling of contact probability and compaction with genomic distance. However, chromatin is also dynamic, with active loop extrusion playing a crucial role. While extrusion ensures a specific spatial organization, how it affects the dynamic scaling of measurable quantities is an open question. In this work, using polymer simulations with active loop extrusion, we demonstrate that the interplay between the timescales of extrusion processes and polymer relaxation can influence the 3D organization of chromatin polymer. We point out this as a factor contributing to the experimentally observed non-trivial scaling of relaxation time with genomic separation and mean-square displacement with time. We show that the dynamic scaling exponents with loop extrusion are consistent with the experimental observations and can be very different from those predicted by existing fractal-globule models for chromat...

The organization of genome within the cell is essential for survival across all domains of life. The physical principles that govern genome organization remain elusive. Phase separation of protein and DNA has emerged as an attractive mechanism for reshaping and compacting the genome. In vitro studies have shed light on the biophysical principles of protein-DNA condensates driven by protein-protein and protein-DNA interactions. However, the role of DNA sequence and its impact on protein-DNA condensation remains elusive. Guided by experiments, we have developed a simple polymer-based model of protein-mediated DNA condensation that explicitly incorporates the influence of DNA sequence on protein binding. By employing coarse-grained Brownian dynamics simulations, we shed light on how DNA sequence affects the number, size and position of protein-DNA condensates. Comparing our simulation results with experimental data for the nucleoid-associated protein Lsr2 provides new insights into the me...