Search

Deep neural networks (DNNs) exhibit a remarkable ability to automatically learn data representations, finding appropriate features without human input. Here we present a method for analysing feature learning by decomposing DNNs into 1) a forward feature-map $\Phi$ that maps the input dataspace to the post-activations of the penultimate layer, and 2) a final linear layer that classifies the data. We diagonalize $\Phi$ with respect to the gradient descent operator and track feature learning by measuring how the eigenfunctions and eigenvalues of $\Phi$ change during training. Across many popular architectures and classification datasets, we find that DNNs converge, after just a few epochs, to a minimal feature (MF) regime dominated by a number of eigenfunctions equal to the number of classes. This behaviour resembles the neural collapse phenomenon studied at longer training times. For other DNN-data combinations, such as a fully connected network on CIFAR10, we find an extended feature (EF) regime where significantly more features are used. Optimal generalisation performance upon hyperparameter tuning typically coincides with the MF regime, but we also find examples of poor performance within the MF regime. Finally, we recast the phenomenon of neural collapse into a kernel picture which can be extended to broader tasks such as regression.

Quantum machine learning models based on parametrized quantum circuits, also called quantum neural networks (QNNs), are considered to be among the most promising candidates for applications on near-term quantum devices. Here we explore the expressivity and inductive bias of QNNs by exploiting an exact mapping from QNNs with inputs $x$ to classical perceptrons acting on $x \otimes x$ (generalised to complex inputs). The simplicity of the perceptron architecture allows us to provide clear examples of the shortcomings of current QNN models, and the many barriers they face to becoming useful general-purpose learning algorithms. For example, a QNN with amplitude encoding cannot express the Boolean parity function for $n\geq 3$, which is but one of an exponential number of data structures that such a QNN is unable to express. Mapping a QNN to a classical perceptron simplifies training, allowing us to systematically study the inductive biases of other, more expressive embeddings on Boolean data. Several popular embeddings primarily produce an inductive bias towards functions with low class balance, reducing their generalisation performance compared to deep neural network architectures which exhibit much richer inductive biases. We explore two alternate strategies that move beyond standard QNNs. In the first, we use a QNN to help generate a classical DNN-inspired kernel. In the second we draw an analogy to the hierarchical structure of deep neural networks and construct a layered non-linear QNN that is provably fully expressive on Boolean data, while also exhibiting a richer inductive bias than simple QNNs. Finally, we discuss characteristics of the QNN literature that may obscure how hard it is to achieve quantum advantage over deep learning algorithms on classical data.

Deep learning models can exhibit what appears to be a sudden ability to solve a new problem as training time, training data, or model size increases, a phenomenon known as emergence. In this paper, we present a framework where each new ability (a skill) is represented as a basis function. We solve a simple multi-linear model in this skill-basis, finding analytic expressions for the emergence of new skills, as well as for scaling laws of the loss with training time, data size, model size, and optimal compute. We compare our detailed calculations to direct simulations of a two-layer neural network trained on multitask sparse parity, where the tasks in the dataset are distributed according to a power-law. Our simple model captures, using a single fit parameter, the sigmoidal emergence of multiple new skills as training time, data size or model size increases in the neural network., Comment: Accepted at NeurIPS 2024 Conference

Arguments inspired by algorithmic information theory predict an inverse relation between the probability and complexity of output patterns in a wide range of input-output maps. This phenomenon is known as \emph{simplicity bias}. By viewing the parameters of dynamical systems as inputs, and resulting (digitised) trajectories as outputs, we study simplicity bias in the logistic map, Gauss map, sine map, Bernoulli map, and tent map. We find that the logistic map, Gauss map, and sine map all exhibit simplicity bias upon sampling of map initial values and parameter values, but the Bernoulli map and tent map do not. The simplicity bias upper bound on output pattern probability is used to make \emph{a priori} predictions for the probability of output patterns. In some cases, the predictions are surprisingly accurate, given that almost no details of the underlying dynamical systems are assumed. More generally, we argue that studying probability-complexity relationships may be a useful tool in studying patterns in dynamical systems., Comment: Preliminary version submitted on Researchgate in 11/2023, https://www.researchgate.net/publication/375121377_Exploring_simplicity_bias_in_1D_dynamical_systems

We introduce oxNA, a new model for the simulation of DNA-RNA hybrids which is based on two previously developed coarse-grained models$\unicode{x2014}$oxDNA and oxRNA. The model naturally reproduces the physical properties of hybrid duplexes including their structure, persistence length and force-extension characteristics. By parameterising the DNA-RNA hydrogen bonding interaction we fit the model's thermodynamic properties to experimental data using both average-sequence and sequence-dependent parameters. To demonstrate the model's applicability we provide three examples of its use$\unicode{x2014}$calculating the free energy profiles of hybrid strand displacement reactions, studying the resolution of a short R-loop and simulating RNA-scaffolded wireframe origami., Comment: 15 pages, 10 figures

The remarkable performance of overparameterized deep neural networks (DNNs) must arise from an interplay between network architecture, training algorithms, and structure in the data. To disentangle these three components, we apply a Bayesian picture, based on the functions expressed by a DNN, to supervised learning. The prior over functions is determined by the network, and is varied by exploiting a transition between ordered and chaotic regimes. For Boolean function classification, we approximate the likelihood using the error spectrum of functions on data. When combined with the prior, this accurately predicts the posterior, measured for DNNs trained with stochastic gradient descent. This analysis reveals that structured data, combined with an intrinsic Occam's razor-like inductive bias towards (Kolmogorov) simple functions that is strong enough to counteract the exponential growth of the number of functions with complexity, is a key to the success of DNNs.

The design space for a self-assembled multicomponent objects ranges from a solution in which every building block is unique to one with the minimum number of distinct building blocks that unambiguously define the target structure. Using a novel pipeline, we explore the design spaces for a set of structures of various sizes and complexities. To understand the implications of the different solutions, we analyse their assembly dynamics using patchy particle simulations and study the influence of the number of distinct building blocks and the angular and spatial tolerances on their interactions on the kinetics and yield of the target assembly. We show that the resource-saving solution with minimum number of distinct blocks can often assemble just as well (or faster) than designs where each building block is unique. We further use our methods to design multifarious structures, where building blocks are shared between different target structures. Finally, we use coarse-grained DNA simulations to investigate the realisation of multicomponent shapes using DNA nanostructures as building blocks., Comment: 12 pages

This chapter will provide an overview of how characterizing free-energy landscapes can provide insights into the biophysical properties of DNA, as well as into the behaviour of the DNA assemblies used in the field of DNA nanotechnology. The landscapes for these complex systems are accessible through the use of accurate coarse-grained descriptions of DNA. Particular foci will be the landscapes associated with DNA self-assembly and mechanical deformation, where the latter can arise from either externally imposed forces or internal stresses., Comment: 20 pages, 5 figures

The intuition that local flatness of the loss landscape is correlated with better generalization for deep neural networks (DNNs) has been explored for decades, spawning many different flatness measures. Recently, this link with generalization has been called into question by a demonstration that many measures of flatness are vulnerable to parameter re-scaling which arbitrarily changes their value without changing neural network outputs. Here we show that, in addition, some popular variants of SGD such as Adam and Entropy-SGD, can also break the flatness-generalization correlation. As an alternative to flatness measures, we use a function based picture and propose using the log of Bayesian prior upon initialization, $\log P(f)$, as a predictor of the generalization when a DNN converges on function $f$ after training to zero error. The prior is directly proportional to the Bayesian posterior for functions that give zero error on a test set. For the case of image classification, we show that $\log P(f)$ is a significantly more robust predictor of generalization than flatness measures are. Whilst local flatness measures fail under parameter re-scaling, the prior/posterior, which is global quantity, remains invariant under re-scaling. Moreover, the correlation with generalization as a function of data complexity remains good for different variants of SGD.

Double-descent curves in neural networks describe the phenomenon that the generalisation error initially descends with increasing parameters, then grows after reaching an optimal number of parameters which is less than the number of data points, but then descends again in the overparameterized regime. In this paper, we use techniques from random matrix theory to characterize the spectral distribution of the empirical feature covariance matrix as a width-dependent perturbation of the spectrum of the neural network Gaussian process (NNGP) kernel, thus establishing a novel connection between the NNGP literature and the random matrix theory literature in the context of neural networks. Our analytical expression allows us to study the generalisation behavior of the corresponding kernel and GP regression, and provides a new interpretation of the double-descent phenomenon, namely as governed by the discrepancy between the width-dependent empirical kernel and the width-independent NNGP kernel.

Across many scientific and engineering disciplines, it is important to consider how much the output of a given system changes due to perturbations of the input. Here, we investigate the glassy phase of $\pm J$ spin glasses at zero temperature by calculating the robustness of the ground states to flips in the sign of single interactions. For random graphs and the Sherrington-Kirkpatrick model, we find relatively large sets of bond configurations that generate the same ground state. These sets can themselves be analyzed as subgraphs of the interaction domain, and we compute many of their topological properties. In particular, we find that the robustness, equivalent to the average degree, of these subgraphs is much higher than one would expect from a random model. Most notably, it scales in the same logarithmic way with the size of the subgraph as has been found in genotype-phenotype maps for RNA secondary structure folding, protein quaternary structure, gene regulatory networks, as well as for models for genetic programming. The similarity between these disparate systems suggests that this scaling may have a more universal origin.

Generalization in deep learning has been the topic of much recent theoretical and empirical research. Here we introduce desiderata for techniques that predict generalization errors for deep learning models in supervised learning. Such predictions should 1) scale correctly with data complexity; 2) scale correctly with training set size; 3) capture differences between architectures; 4) capture differences between optimization algorithms; 5) be quantitatively not too far from the true error (in particular, be non-vacuous); 6) be efficiently computable; and 7) be rigorous. We focus on generalization error upper bounds, and introduce a categorisation of bounds depending on assumptions on the algorithm and data. We review a wide range of existing approaches, from classical VC dimension to recent PAC-Bayesian bounds, commenting on how well they perform against the desiderata. We next use a function-based picture to derive a marginal-likelihood PAC-Bayesian bound. This bound is, by one definition, optimal up to a multiplicative constant in the asymptotic limit of large training sets, as long as the learning curve follows a power law, which is typically found in practice for deep learning problems. Extensive empirical analysis demonstrates that our marginal-likelihood PAC-Bayes bound fulfills desiderata 1-3 and 5. The results for 6 and 7 are promising, but not yet fully conclusive, while only desideratum 4 is currently beyond the scope of our bound. Finally, we comment on why this function-based bound performs significantly better than current parameter-based PAC-Bayes bounds.

We present a new method for calculating internal forces in DNA structures using coarse-grained models and demonstrate its utility with the oxDNA model. The instantaneous forces on individual nucleotides are explored and related to model potentials, and using our framework, internal forces are calculated for two simple DNA systems and for a recently-published nanoscopic force clamp. Our results highlight some pitfalls associated with conventional methods for estimating internal forces, which are based on elastic polymer models, and emphasise the importance of carefully considering secondary structure and ionic conditions when modelling the elastic behaviour of single-stranded DNA. Beyond its relevance to the DNA nanotechnological community, we expect our approach to be broadly applicable to calculations of internal force in a variety of structures -- from DNA to protein -- and across other coarse-grained simulation models., Comment: 39 pages, 9 figures

Overparameterised deep neural networks (DNNs) are highly expressive and so can, in principle, generate almost any function that fits a training dataset with zero error. The vast majority of these functions will perform poorly on unseen data, and yet in practice DNNs often generalise remarkably well. This success suggests that a trained DNN must have a strong inductive bias towards functions with low generalisation error. Here we empirically investigate this inductive bias by calculating, for a range of architectures and datasets, the probability $P_{SGD}(f\mid S)$ that an overparameterised DNN, trained with stochastic gradient descent (SGD) or one of its variants, converges on a function $f$ consistent with a training set $S$. We also use Gaussian processes to estimate the Bayesian posterior probability $P_B(f\mid S)$ that the DNN expresses $f$ upon random sampling of its parameters, conditioned on $S$. Our main findings are that $P_{SGD}(f\mid S)$ correlates remarkably well with $P_B(f\mid S)$ and that $P_B(f\mid S)$ is strongly biased towards low-error and low complexity functions. These results imply that strong inductive bias in the parameter-function map (which determines $P_B(f\mid S)$), rather than a special property of SGD, is the primary explanation for why DNNs generalise so well in the overparameterised regime. While our results suggest that the Bayesian posterior $P_B(f\mid S)$ is the first order determinant of $P_{SGD}(f\mid S)$, there remain second order differences that are sensitive to hyperparameter tuning. A function probability picture, based on $P_{SGD}(f\mid S)$ and/or $P_B(f\mid S)$, can shed new light on the way that variations in architecture or hyperparameter settings such as batch size, learning rate, and optimiser choice, affect DNN performance.

The success of DNA nanotechnology has been driven by the discovery of novel structural motifs with a wide range of shapes and uses. We present a comprehensive study of the T-motif, a 3-armed, planar, right-angled junction that has been used in the self-assembly of DNA polyhedra and periodic structures. The motif is formed through the interaction of a bulge loop in one duplex and a sticky end of another. The polarity of the sticky end has significant consequences for the thermodynamic and geometrical properties of the T-motif: different polarities create junctions spanning different grooves of the duplex. We compare experimental binding strengths with predictions of oxDNA, a coarse-grained model of DNA, for various loop sizes. We find that, although both sticky-end polarities can create stable junctions, junctions resulting from 5$'$ sticky ends are stable over a wider range of bulge loop sizes. We highlight the importance of possible coaxial stacking interactions within the motif and investigate how each coaxial stacking interaction stabilises the structure and favours a particular geometry., Comment: 8 pages; 8 figures

This chapter introduces how to run molecular dynamics simulations for DNA origami using the oxDNA coarse-grained model., Comment: 17 pages, 5 figures

Understanding how genotypes map onto phenotypes, fitness, and eventually organisms is arguably the next major missing piece in a fully predictive theory of evolution. We refer to this generally as the problem of the genotype-phenotype map. Though we are still far from achieving a complete picture of these relationships, our current understanding of simpler questions, such as the structure induced in the space of genotypes by sequences mapped to molecular structures, has revealed important facts that deeply affect the dynamical description of evolutionary processes. Empirical evidence supporting the fundamental relevance of features such as phenotypic bias is mounting as well, while the synthesis of conceptual and experimental progress leads to questioning current assumptions on the nature of evolutionary dynamics-cancer progression models or synthetic biology approaches being notable examples. This work delves into a critical and constructive attitude in our current knowledge of how genotypes map onto molecular phenotypes and organismal functions, and discusses theoretical and empirical avenues to broaden and improve this comprehension. As a final goal, this community should aim at deriving an updated picture of evolutionary processes soundly relying on the structural properties of genotype spaces, as revealed by modern techniques of molecular and functional analysis., Comment: 111 pages, 11 figures uses elsarticle latex class

For a broad class of input-output maps, arguments based on the coding theorem from algorithmic information theory (AIT) predict that simple (low Kolmogorov complexity) outputs are exponentially more likely to occur upon uniform random sampling of inputs than complex outputs are. Here, we derive probability bounds that are based on the complexities of the inputs as well as the outputs, rather than just on the complexities of the outputs. The more that outputs deviate from the coding theorem bound, the lower the complexity of their inputs. Our new bounds are tested for an RNA sequence to structure map, a finite state transducer and a perceptron. These results open avenues for AIT to be more widely used in physics., Comment: 6 pages plus supplementary materials

Understanding the inductive bias of neural networks is critical to explaining their ability to generalise. Here, for one of the simplest neural networks -- a single-layer perceptron with n input neurons, one output neuron, and no threshold bias term -- we prove that upon random initialisation of weights, the a priori probability $P(t)$ that it represents a Boolean function that classifies t points in ${0,1}^n$ as 1 has a remarkably simple form: $P(t) = 2^{-n}$ for $0\leq t < 2^n$. Since a perceptron can express far fewer Boolean functions with small or large values of t (low entropy) than with intermediate values of t (high entropy) there is, on average, a strong intrinsic a-priori bias towards individual functions with low entropy. Furthermore, within a class of functions with fixed t, we often observe a further intrinsic bias towards functions of lower complexity. Finally, we prove that, regardless of the distribution of inputs, the bias towards low entropy becomes monotonically stronger upon adding ReLU layers, and empirically show that increasing the variance of the bias term has a similar effect.

We use the oxDNA coarse-grained model to provide a detailed characterization of the fundamental structural properties of DNA origamis, focussing on archetypal 2D and 3D origamis. The model reproduces well the characteristic pattern of helix bending in a 2D origami, showing that it stems from the intrinsic tendency of anti-parallel four-way junctions to splay apart, a tendency that is enhanced both by less screened electrostatic interactions and by increased thermal motion. We also compare to the structure of a 3D origami whose structure has been determined by cryo-electron microscopy. The oxDNA average structure has a root-mean-square deviation from the experimental structure of 8.4 Angstrom, which is of the order of the experimental resolution. These results illustrate that the oxDNA model is capable of providing detailed and accurate insights into the structure of DNA origamis, and has the potential to be used to routinely pre-screen putative origami designs., Comment: 14 pages, 10 figures

Deep neural networks (DNNs) generalize remarkably well without explicit regularization even in the strongly over-parametrized regime where classical learning theory would instead predict that they would severely overfit. While many proposals for some kind of implicit regularization have been made to rationalise this success, there is no consensus for the fundamental reason why DNNs do not strongly overfit. In this paper, we provide a new explanation. By applying a very general probability-complexity bound recently derived from algorithmic information theory (AIT), we argue that the parameter-function map of many DNNs should be exponentially biased towards simple functions. We then provide clear evidence for this strong simplicity bias in a model DNN for Boolean functions, as well as in much larger fully connected and convolutional networks applied to CIFAR10 and MNIST. As the target functions in many real problems are expected to be highly structured, this intrinsic simplicity bias helps explain why deep networks generalize well on real world problems. This picture also facilitates a novel PAC-Bayes approach where the prior is taken over the DNN input-output function space, rather than the more conventional prior over parameter space. If we assume that the training algorithm samples parameters close to uniformly within the zero-error region then the PAC-Bayes theorem can be used to guarantee good expected generalization for target functions producing high-likelihood training sets. By exploiting recently discovered connections between DNNs and Gaussian processes to estimate the marginal likelihood, we produce relatively tight generalization PAC-Bayes error bounds which correlate well with the true error on realistic datasets such as MNIST and CIFAR10 and for architectures including convolutional and fully connected networks., Comment: Published as a conference paper at ICLR 2019

Inspired by recent successes using single-stranded DNA tiles to produce complex structures, we develop a two-step coarse-graining approach that uses detailed thermodynamic calculations with oxDNA, a nucleotide-based model of DNA, to parametrize a coarser kinetic model that can reach the time and length scales needed to study the assembly mechanisms of these structures. We test the model by performing a detailed study of the assembly pathways for a two-dimensional target structure made up of 334 unique strands each of which are 42 nucleotides long. Without adjustable parameters, the model reproduces a critical temperature for the formation of the assembly that is close to the temperature at which assembly first occurs in experiments. Furthermore, the model allows us to investigate in detail the nucleation barriers and the distribution of critical nucleus shapes for the assembly of a single target structure. The assembly intermediates are compact and highly connected (although not maximally so) and classical nucleation theory provides a good fit to the height and shape of the nucleation barrier at temperatures close to where assembly first occurs.

DNA bending is biologically important for genome regulation and is relevant to a range of nanotechnological systems. Recent results suggest that sharp bending is much easier than implied by the widely-used worm-like chain model; many of these studies, however, remain controversial. We use a coarse-grained model, previously fitted to DNA's basic thermodynamic and mechanical properties, to explore strongly bent systems. We find that as the end-to-end distance is decreased sufficiently short duplexes undergo a transition to a state in which the bending strain is localized at a flexible kink that involves disruption of base-pairing and stacking. This kinked state, which is not well-described by the worm-like chain model, allows the duplex to more easily be sharply bent. It is not completely flexible, however, due to constraints arising from the connectivity of both DNA backbones. We also perform a detailed comparison to recent experiments on a "molecular vice" that probes highly bent DNA. Close agreement between simulations and experiments strengthens the hypothesis that localised bending via kinking occurs in the molecular vice and causes enhanced flexibility of duplex DNA. Our calculations therefore suggests that the cost of kinking implied by this experiment is consistent with the known thermodynamic and mechanical properties of DNA.

DNA cyclization is a powerful technique to gain insight into the nature of DNA bending. The worm-like chain model provides a good description of small to moderate bending fluctuations, but some experiments on strongly-bent shorter molecules suggest enhanced flexibility over and above that expected from the worm-like chain. Here, we use a coarse-grained model of DNA to investigate the thermodynamics of DNA cyclization for molecules with less than 210 base pairs. As the molecules get shorter we find increasing deviations between our computed equilibrium j-factor and the worm-like chain predictions of Shimada and Yamakawa. These deviations are due to sharp kinking, first at nicks, and only subsequently in the body of the duplex. At the shortest lengths, substantial fraying at the ends of duplex domains is the dominant method of relaxation. We also estimate the dynamic j-factor measured in recent FRET experiments. We find that the dynamic j-factor is systematically larger than its equilibrium counterpart, with the deviation larger for shorter molecules, because not all the stress present in the fully cyclized state is present in the transition state. These observations are important for the interpretation of recent experiments, as only kinking within the body of the duplex is genuinely indicative of non-worm-like chain behaviour.

The prevalence of neutral mutations implies that biological systems typically have many more genotypes than phenotypes. But can the way that genotypes are distributed over phenotypes determine evolutionary outcomes? Answering such questions is difficult because the number of genotypes can be hyper-astronomically large. By solving the genotype-phenotype (GP) map for RNA secondary structure for systems up to length $L=126$ nucleotides (where the set of all possible RNA strands would weigh more than the mass of the visible universe) we show that the GP map strongly constrains the evolution of non-coding RNA (ncRNA). Simple random sampling over genotypes predicts the distribution of properties such as the mutational robustness or the number of stems per secondary structure found in naturally occurring ncRNA with surprising accuracy. Since we ignore natural selection, this strikingly close correspondence with the mapping suggests that structures allowing for functionality are easily discovered, despite the enormous size of the genetic spaces. The mapping is extremely biased: the majority of genotypes map to an exponentially small portion of the morphospace of all biophysically possible structures. Such strong constraints provide a non-adaptive explanation for the convergent evolution of structures such as the hammerhead ribozyme. These results presents a particularly clear example of bias in the arrival of variation strongly shaping evolutionary outcomes and may be relevant to Mayr's distinction between proximate and ultimate causes in evolutionary biology., Comment: new version with some small tweaks to clarify the conclusions and some minor corrections

We study the behaviour of double-stranded RNA under twist and tension using oxRNA, a recently developed coarse-grained model of RNA. Introducing explicit salt-dependence into the model allows us to directly compare our results to data from recent single-molecule experiments. The model reproduces extension curves as a function of twist and stretching force, including the buckling transition and the behaviour of plectoneme structures. For negative supercoiling, we predict denaturation bubble formation in plectoneme end-loops, suggesting preferential plectoneme localisation in weak base sequences. OxRNA exhibits a positive twist-stretch coupling constant, in agreement with recent experimental observations., Comment: 8 pages + 5 pages Supplementary Material

Mutational neighbourhoods in genotype-phenotype (GP) maps are widely believed to be more likely to share characteristics than expected from random chance. Such genetic correlations should, as John Maynard Smith famously pointed out, strongly influence evolutionary dynamics. We explore and quantify these intuitions by comparing three GP maps - RNA SS, HP for tertiary, Polyominoes for protein quaternary structure - to a simple random null model that maintains the number of genotypes mapping to each phenotype, but assigns genotypes randomly. The mutational neighbourhood of a genotype in these GP maps is much more likely to contain (mutationally neutral) genotypes mapping to the same phenotype than in the random null model. These neutral correlations can increase the robustness to mutations by orders of magnitude over that of the null model, raising robustness above the critical threshold for the formation of large neutral networks that enhance the capacity for neutral exploration. We also study {\em non-neutral correlations}: Compared to the null model, i) If a particular (non-neutral) phenotype is found once in the 1-mutation neighbourhood of a genotype, then the chance of finding that phenotype multiple times in this neighbourhood is larger than expected; ii) If two genotypes are connected by a single neutral mutation, then their respective non-neutral 1-mutation neighbourhoods are more likely to be similar; iii) If a genotype maps to a folding or self-assembling phenotype, then its non-neutral neighbours are less likely to be a potentially deleterious non-folding or non-assembling phenotype. Non-neutral correlations of type i) and ii) reduce the rate at which new phenotypes can be found by neutral exploration, and so may diminish evolvability, while non-neutral correlations of type iii) may instead facilitate evolutionary exploration and so increase evolvability., Comment: corrected manuscript

We introduce an extended version of oxDNA, a coarse-grained model of DNA designed to capture the thermodynamic, structural and mechanical properties of single- and double-stranded DNA. By including explicit major and minor grooves, and by slightly modifying the coaxial stacking and backbone-backbone interactions, we improve the ability of the model to treat large (kilobase-pair) structures such as DNA origami which are sensitive to these geometric features. Further, we extend the model, which was previously parameterised to just one salt concentration ([Na$^+$]=0.5M), so that it can be used for a range of salt concentrations including those corresponding to physiological conditions. Finally, we use new experimental data to parameterise the oxDNA potential so that consecutive adenine bases stack with a different strength to consecutive thymine bases, a feature which allows a more accurate treatment of systems where the flexibility of single-stranded regions is important. We illustrate the new possibilities opened up by the updated model, oxDNA2, by presenting results from simulations of the structure of large DNA objects and by using the model to investigate some salt-dependent properties of DNA.

The rupture of double-stranded DNA under stress is a key process in biophysics and nanotechnology. In this article we consider the shear-induced rupture of short DNA duplexes, a system that has been given new importance by recently designed force sensors and nanotechnological devices. We argue that rupture must be understood as an activated process, where the duplex state is metastable and the strands will separate in a finite time that depends on the duplex length and the force applied. Thus, the critical shearing force required to rupture a duplex within a given experiment depends strongly on the time scale of observation. We use simple models of DNA to demonstrate that this approach naturally captures the experimentally observed dependence of the critical force on duplex length for a given observation time. In particular, the critical force is zero for the shortest duplexes, before rising sharply and then plateauing in the long length limit. The prevailing approach, based on identifying when the presence of each additional base pair within the duplex is thermodynamically unfavorable rather than allowing for metastability, does not predict a time-scale-dependent critical force and does not naturally incorporate a critical force of zero for the shortest duplexes. Additionally, motivated by a recently proposed force sensor, we investigate application of stress to a duplex in a mixed mode that interpolates between shearing and unzipping. As with pure shearing, the critical force depends on the time scale of observation; at a fixed time scale and duplex length, the critical force exhibits a sigmoidal dependence on the fraction of the duplex that is subject to shearing., Comment: 10 pages, 6 figures

Advances in DNA nanotechnology have stimulated the search for simple motifs that can be used to control the properties of DNA nanostructures. One such motif, which has been used extensively in structures such as polyhedral cages, two-dimensional arrays, and ribbons, is a bulged duplex, that is two helical segments that connect at a bulge loop. We use a coarse-grained model of DNA to characterize such bulged duplexes. We find that this motif can adopt structures belonging to two main classes: one where the stacking of the helices at the center of the system is preserved, the geometry is roughly straight and the bulge is on one side of the duplex, and the other where the stacking at the center is broken, thus allowing this junction to act as a hinge and increasing flexibility. Small loops favor states where stacking at the center of the duplex is preserved, with loop bases either flipped out or incorporated into the duplex. Duplexes with longer loops show more of a tendency to unstack at the bulge and adopt an open structure. The unstacking probability, however, is highest for loops of intermediate lengths, when the rigidity of single-stranded DNA is significant and the loop resists compression. The properties of this basic structural motif clearly correlate with the structural behavior of certain nano-scale objects, where the enhanced flexibility associated with larger bulges has been used to tune the self-assembly product as well as the detailed geometry of the resulting nanostructures., Comment: 12 pages + 4 pages of supplemental materials

We study the thermodynamics and kinetics of an RNA toehold-mediated strand displacement reaction with a recently developed coarse-grained model of RNA. Strand displacement, during which a single strand displaces a different strand previously bound to a complementary substrate strand, is an essential mechanism in active nucleic acid nanotechnology and has also been hypothesized to occur in vivo. We study the rate of displacement reactions as a function of the length of the toehold and temperature and make two experimentally testable predictions: that the displacement is faster if the toehold is placed at the 5' end of the substrate and that the displacement slows down with increasing temperature for longer toeholds.

We study the dynamics of DNA hairpin formation using oxDNA, a nucleotide-level coarse-grained model of DNA. In particular, we explore the effects of the loop stacking interactions and non-native base pairing on the hairpin closing times. We find a non-monotonic variation of the hairpin closing time with temperature, in agreement with the experimental work of Wallace et al. [Proc. Nat. Acad. Sci. USA 2001, 98, 5584-5589]. The hairpin closing process involves the formation of an initial nucleus of one or two bonds between the stems followed by a rapid zippering of the stem. At high temperatures, typically above the hairpin melting temperature, an effective negative activation enthalpy is observed because the nucleus has a lower enthalpy than the open state. By contrast, at low temperatures, the activation enthalpy becomes positive mainly due to the increasing energetic cost of bending a loop that becomes increasingly highly stacked as the temperature decreases. We show that stacking must be very strong to induce this experimentally observed behavior, and that the existence of just a few weak stacking points along the loop can substantially suppress it. Non-native base pairs are observed to have only a small effect, slightly accelerating hairpin formation., Comment: 24 pages, 9 figures

The effect of secondary structure on DNA duplex formation is poorly understood. We use a coarse-grained model of DNA to show that specific 3- and 4-base pair hairpins reduce hybridization rates by factors of 2 and 10 respectively, in good agreement with experiment. By contrast, melting rates are accelerated by factors of ~100 and ~2000. This surprisingly large speed-up occurs because hairpins form during the melting process, stabilizing partially melted states, and facilitating dissociation. These results may help guide the design of DNA devices that use hairpins to modulate hybridization and dissociation pathways and rates., Comment: 5 pages + 14 pages of appendices

Biological information is not only stored in the digital chemical sequence of double helical DNA, but is also encoded in the mechanical properties of the DNA strands, which can influence biochemical processes involving its readout. For example, loop formation in the Lac operon can regulate the expression of key genes, and DNA supercoiling is closely correlated to rhythmic circardian gene expression in cyanobacteria. Supercoiling is also important for large scale organisation of the genome in both eukaryotic and prokaryotic cells. DNA can respond to torsional stress by writhing to form looped structures called plectonemes, thus transferring energy stored as twist into energy stored in bending. Denaturation bubbles can also relax torsional stress, with the enthalpic cost of breaking bonds being compensated by their ability to absorb undertwist. Here we predict a novel regime where bubbles form at the tips of plectonemes, and study its properties using coarse-grained simulations. These tip bubbles can occur for both positive and negative supercoiling and greatly reduce plectoneme diffusion by a pinning mechanism. They can cause plectonemes to preferentially localise to AT rich regions, because bubbles more easily form there. The tip-bubble regime occurs for supercoiling densities and forces that are typically encountered for DNA in vivo, and may be exploited for biological control of genomic processes., Comment: 5 pages + 26 pages Supplementary Material

We present a new, nucleotide-level model for RNA, oxRNA, based on the coarse-graining methodology recently developed for the oxDNA model of DNA. The model is designed to reproduce structural, mechanical and thermodynamic properties of RNA, and the coarse-graining level aims to retain the relevant physics for RNA hybridization and the structure of single- and double-stranded RNA. In order to explore its strengths and weaknesses, we test the model in a range of nanotechnological and biological settings. Applications explored include the folding thermodynamics of a pseudoknot, the formation of a kissing loop complex, the structure of a hexagonal RNA nanoring, and the unzipping of a hairpin motif. We argue that the model can be used for efficient simulations of the structure of systems with thousands of base pairs, and for the assembly of systems of up to hundreds of base pairs. The source code implementing the model is released for public use., Comment: 25 pages

Genotype-phenotype (GP) maps specify how the random mutations that change genotypes generate variation by altering phenotypes, which, in turn, can trigger selection. Many GP maps share the following general properties: 1) The number of genotypes $N_G$ is much larger than the number of selectable phenotypes; 2) Neutral exploration changes the variation that is accessible to the population; 3) The distribution of phenotype frequencies $F_p=N_p/N_G$, with $N_p$ the number of genotypes mapping onto phenotype $p$, is highly biased: the majority of genotypes map to only a small minority of the phenotypes. Here we explore how these properties affect the evolutionary dynamics of haploid Wright-Fisher models that are coupled to a simplified and general random GP map or to a more complex RNA sequence to secondary structure map. For both maps the probability of a mutation leading to a phenotype $p$ scales to first order as $F_p$, although for the RNA map there are further correlations as well. By using mean-field theory, supported by computer simulations, we show that the discovery time $T_p$ of a phenotype $p$ similarly scales to first order as $1/F_p$ for a wide range of population sizes and mutation rates in both the monomorphic and polymorphic regimes. These differences in the rate at which variation arises can vary over many orders of magnitude. Phenotypic variation with a larger $F_p$ is therefore be much more likely to arise than variation with a small $F_p$. We show, using the RNA model, that frequent phenotypes (with larger $F_p$) can fix in a population even when alternative, but less frequent, phenotypes with much higher fitness are potentially accessible. In other words, if the fittest never `arrive' on the timescales of evolutionary change, then they can't fix. We call this highly non-ergodic effect the `arrival of the frequent'., Comment: full paper plus supplementary materials

The mapping between biological genotypes and phenotypes is central to the study of biological evolution. Here we introduce a rich, intuitive, and biologically realistic genotype-phenotype (GP) map, that serves as a model of self-assembling biological structures, such as protein complexes, and remains computationally and analytically tractable. Our GP map arises naturally from the self-assembly of polyomino structures on a 2D lattice and exhibits a number of properties: $\textit{redundancy}$ (genotypes vastly outnumber phenotypes), $\textit{phenotype bias}$ (genotypic redundancy varies greatly between phenotypes), $\textit{genotype component disconnectivity}$ (phenotypes consist of disconnected mutational networks) and $\textit{shape space covering}$ (most phenotypes can be reached in a small number of mutations). We also show that the mutational robustness of phenotypes scales very roughly logarithmically with phenotype redundancy and is positively correlated with phenotypic evolvability. Although our GP map describes the assembly of disconnected objects, it shares many properties with other popular GP maps for connected units, such as models for RNA secondary structure or the HP lattice model for protein tertiary structure. The remarkable fact that these important properties similarly emerge from such different models suggests the possibility that universal features underlie a much wider class of biologically realistic GP maps., Comment: 12 pages, 6 figures

Paramagnetic colloidal particles that are optically trapped in a linear array can form a zigzag pattern when an external magnetic field induces repulsive interparticle interactions. When the traps are abruptly turned off, the particles form a nonequilibrium expanding pattern with a zigzag symmetry, even when the strength of the magnetic interaction is weaker than that required to break the linear symmetry of the equilibrium state. We show that the transition to the equilibrium zigzag state is always potentially possible for purely harmonic traps. For anharmonic traps that have a finite height, the equilibrium zigzag state becomes unstable above a critical anharmonicity. A normal mode analysis of the equilibrium line configuration demonstrates that increasing the magnetic field leads to a hardening and softening of the spring constants in the longitudinal and transverse directions, respectively. The mode that first becomes unstable is the mode with the zigzag symmetry, which explains the symmetry of nonequilibrium patterns. Our analytically tractable models help to give further insight into the way that the interplay of such factors as the length of the chain, hydrodynamic interactions, thermal fluctuations affect the formation and evolution of the experimentally observed nonequilibrium patterns., Comment: 16 pages, 8 figures; to appear in the Journal of Chemical Physics

We experimentally study the viscous fingering instability in a fluid-fluid phase separated colloid-polymer mixture by means of laser scanning confocal microscopy and microfluidics. We focus on three aspects of the instability. (i) The interface between the two demixed phases has an ultralow surface tension, such that we can address the role of thermal interface fluctuations. (ii) We image the interface in three dimensions allowing us to study the interplay between interface curvature and flow. (iii) The displacing fluid wets all walls completely, in contrast to traditional viscous fingering experiments, in which the displaced fluid wets the walls. We also perform lattice Boltzmann simulations, which help to interpret the experimental observations.

To simulate long time and length scale processes involving DNA it is necessary to use a coarse-grained description. Here we provide an overview of different approaches to such coarse graining, focussing on those at the nucleotide level that allow the self-assembly processes associated with DNA nanotechnology to be studied. OxDNA, our recently-developed coarse-grained DNA model, is particularly suited to this task, and has opened up this field to systematic study by simulations. We illustrate some of the range of DNA nanotechnology systems to which the model is being applied, as well as the insights it can provide into fundamental biophysical properties of DNA., Comment: 20 pages, 9 figures