Predicting aggregate morphology of sequence-defined macromolecules with recurrent neural networks.

Self-assembly of dilute sequence-defined macromolecules is a complex phenomenon in which the local arrangement of chemical moieties can lead to the formation of long-range structure. The dependence of this structure on the sequence necessarily implies that a mapping between the two exists, yet it has been difficult to model so far. Predicting the aggregation behavior of these macromolecules is challenging due to the lack of effective order parameters, a vast design space, inherent variability, and high computational costs associated with currently available simulation techniques. Here, we accurately predict the morphology of aggregates self-assembled from sequence-defined macromolecules using supervised machine learning. We find that regression models with implicit representation learning perform significantly better than those based on engineered features such as k-mer counting, and a recurrent-neural-network-based regressor performs the best out of nine model architectures we tested. Furthermore, we demonstrate the high-throughput screening of monomer sequences using the regression model to identify candidates for self-assembly into selected morphologies. Our strategy is shown to successfully identify multiple suitable sequences in every test we performed, so we hope the insights gained here can be extended to other increasingly complex design scenarios in the future, such as the design of sequences under polydispersity and at varying environmental conditions.

Unsupervised learning of sequence-specific aggregation behavior for a model copolymer.

We apply a recently developed unsupervised machine learning scheme for local environments [Reinhart, Comput. Mater. Sci., 2021, 196, 110511] to characterize large-scale, disordered aggregates formed by sequence-defined macromolecules. This method provides new insight into the structure of these disordered, dilute aggregates, which has proven difficult to understand using collective variables manually derived from expert knowledge [Statt et al., J. Chem. Phys., 2020, 152, 075101]. In contrast to such conventional order parameters, we are able to classify the global aggregate structure directly using descriptions of the local environments. The resulting characterization provides a deeper understanding of the range of possible self-assembled structures and their relationships to each other. We also provide detailed analysis of the effects of finite system size, stochasticity, and kinetics of these aggregates based on the learned collective variables. Interestingly, we find that the spatiotemporal evolution of systems in the learned latent space is smooth and continuous, despite being derived from only a single snapshot from each of about 1000 monomer sequences. These results demonstrate the insight which can be gained by applying unsupervised machine learning to soft matter systems, especially when suitable order parameters are not known.

Crack Detection in Images of Masonry Using CNNs.

While there is a significant body of research on crack detection by computer vision methods in concrete and asphalt, less attention has been given to masonry. We train a convolutional neural network (CNN) on images of brick walls built in a laboratory environment and test its ability to detect cracks in images of brick-and-mortar structures both in the laboratory and on real-world images taken from the internet. We also compare the performance of the CNN to a variety of simpler classifiers operating on handcrafted features. We find that the CNN performed better on the domain adaptation from laboratory to real-world images than these simple models. However, we also find that performance is significantly better in performing the reverse domain adaptation task, where the simple classifiers are trained on real-world images and tested on the laboratory images. This work demonstrates the ability to detect cracks in images of masonry using a variety of machine learning methods and provides guidance for improving the reliability of such models when performing domain adaptation for crack detection in masonry.

Directed assembly of photonic crystals through simple substrate patterning.

We present molecular dynamics simulations of the epitaxial growth of high quality crystalline films for photonics applications from triblock Janus colloids. With a featureless substrate, the film morphologies were qualitatively similar to previously reported experimental results, with two stacking polymorphs appearing in nearly equal proportion. However, with a patterned substrate deliberately designed to be easy to fabricate by standard photolithography techniques, both the grain size and selectivity towards the photonically active polymorph were greatly improved. We also evaluated the effect of particle flux to find that lower flux led to higher quality crystals, while higher flux led to frustrated films with smaller crystalline domains. Our results suggest that carefully engineered but simple to manufacture patterned substrates could yield self-assembled single crystals of sufficient quality to exhibit a complete photonic bandgap.

Evaporation-induced assembly of colloidal crystals.

Colloidal crystals are often prepared by evaporation from solution, and there is considerable interest to link the processing conditions to the crystal morphology and quality. Here, we study the evaporation-induced assembly of colloidal crystals using massive-scale nonequilibrium molecular dynamics simulations. We apply a recently developed machine-learning technique to characterize the assembling crystal structures with unprecedented microscopic detail. In agreement with previous experiments and simulations, faster evaporation rates lead to earlier onset of crystallization and more disordered surface structures. Surprisingly, we find that collective rearrangements of the bulk crystal during later stages of drying reduce the influence of the initial surface structure, and the final morphology is essentially independent of the evaporation rate. Our structural analysis reveals that the crystallization process is well-described by two time scales, the film drying time and the crystal growth time, with the latter having an unexpected dependence on the evaporation rate due to equilibrium thermodynamic effects at high colloid concentrations. These two time scales may be leveraged to control the relative influence of equilibrium and nonequilibrium growth mechanisms, suggesting a route to rapidly process colloidal crystals while also removing defects. Our analysis additionally reveals that solvent-mediated interactions play a critical role in the crystallization kinetics and that commonly used implicit-solvent models do not faithfully resolve nonequilibrium processes such as drying.

Automated crystal characterization with a fast neighborhood graph analysis method.

We present a significantly improved, very fast implementation of the Neighborhood Graph Analysis technique for template-free characterization of crystal structures [W. F. Reinhart et al., Soft Matter, 2017, 13, 4733]. By comparing local neighborhoods in terms of their relative graphlet frequencies, we reduce the computational cost by four orders of magnitude compared to the original stochastic method. Furthermore, we present protocols for the detection of topologically important structures and assignment of visually informative colors, providing a fully automated procedure for characterization of crystal structures from particle tracking data. We demonstrate the flexibility of our method on a wide range of crystal structures which have proven difficult to classify by previously available techniques.

Crystal growth kinetics of triblock Janus colloids.

We measure the kinetics of crystal growth from a melt of triblock Janus colloids using non-equilibrium molecular dynamics simulations. We assess the impact of interaction anisotropy by systematically varying the size of the attractive patches from 40% to 100% coverage, finding substantially different growth behaviors in the two limits. With isotropic particles, the interface velocity is directly proportional to the subcooling, in agreement with previous studies. With highly anisotropic particles, the growth curves are well approximated by using a power law with exponent and prefactor that depend strongly on the particular surface geometry and patch fraction. This nonlinear growth appears correlated to the roughness of the solid-liquid interface, with the strongest growth inhibition occurring for the smoothest crystal faces. We conclude that crystal growth for patchy particles does not conform to the typical collision-limited mechanism, but is instead an activated process in which the rate-limiting step is the collective rotation of particles into the proper orientation. Finally, we show how differences in the growth kinetics could be leveraged to achieve kinetic control over polymorph growth, either enhancing or suppressing metastable phases near solid-solid coexistence lines.

Multi-atom pattern analysis for binary superlattices.

We present a method for the template-free characterization of binary superlattices. This is an extension of the Neighborhood Graph Analysis method, a technique which evaluates relationships between observed structures based on the topology of their first coordination shell [W. F. Reinhart, et al., Soft Matter, 2017, 13, 4733]. In the present work, we develop a framework for the analysis of multi-atom patterns, which incorporate structural information from the second coordination shell while providing a unified signature for all constituent particles in the superlattice. We construct an efficient metric for making quantitative comparisons between these patterns, making our algorithm the first capable of characterizing partial or defective superlattice structures. As in our previous work, we leverage machine learning techniques to characterize a range of self-assembled crystal structures, discovering a set of emergent collective variables which map each observed pattern into an intuitive global phase space. We demonstrate the method by performing classification of configurations from simulations of binary colloidal self-assembly in two dimensions.

Machine learning for autonomous crystal structure identification.

We present a machine learning technique to discover and distinguish relevant ordered structures from molecular simulation snapshots or particle tracking data. Unlike other popular methods for structural identification, our technique requires no a priori description of the target structures. Instead, we use nonlinear manifold learning to infer structural relationships between particles according to the topology of their local environment. This graph-based approach yields unbiased structural information which allows us to quantify the crystalline character of particles near defects, grain boundaries, and interfaces. We demonstrate the method by classifying particles in a simulation of colloidal crystallization, and show that our method identifies structural features that are missed by standard techniques.

Equilibrium crystal phases of triblock Janus colloids.

Triblock Janus colloids, which are colloidal spheres decorated with attractive patches at each pole, have recently generated significant interest as potential building blocks for functional materials. Their inherent anisotropy is known to induce self-assembly into open structures at moderate temperatures and pressures, where they are stabilized over close-packed crystals by entropic effects. We present a numerical investigation of the equilibrium phases of triblock Janus particles with many different patch geometries in three dimensions, using Monte Carlo simulations combined with free energy calculations. In all cases, we find that the free energy difference between crystal polymorphs is less than 0.2 kBT per particle. By varying the patch fraction and interaction range, we show that large patches stabilize the formation of structures with four bonds per patch over those with three. This transition occurs abruptly above a patch fraction of 0.30 and has a strong dependence on the interaction range. Furthermore, we find that a short interaction range favors four bonds per patch, with longer range increasingly stabilizing structures with only three bonds per patch. By quantifying the effect of patch geometry on the stability of the equilibrium crystal structures, we provide insights into the fundamental design rules for constructing complex colloidal crystals.

Distribution of distances between DNA barcode labels in nanochannels close to the persistence length.

We obtained experimental extension data for barcoded E. coli genomic DNA molecules confined in nanochannels from 40 nm to 51 nm in width. The resulting data set consists of 1 627 779 measurements of the distance between fluorescent probes on 25 407 individual molecules. The probability density for the extension between labels is negatively skewed, and the magnitude of the skewness is relatively insensitive to the distance between labels. The two Odijk theories for DNA confinement bracket the mean extension and its variance, consistent with the scaling arguments underlying the theories. We also find that a harmonic approximation to the free energy, obtained directly from the probability density for the distance between barcode labels, leads to substantial quantitative error in the variance of the extension data. These results suggest that a theory for DNA confinement in such channels must account for the anharmonic nature of the free energy as a function of chain extension.

Entropic depletion of DNA in triangular nanochannels.

Using Monte Carlo simulations of a touching-bead model of double-stranded DNA, we show that DNA extension is enhanced in isosceles triangular nanochannels (relative to a circular nanochannel of the same effective size) due to entropic depletion in the channel corners. The extent of the enhanced extension depends non-monotonically on both the accessible area of the nanochannel and the apex angle of the triangle. We also develop a metric to quantify the extent of entropic depletion, thereby collapsing the extension data for circular, square, and various triangular nanochannels onto a single master curve for channel sizes in the transition between the Odijk and de Gennes regimes.