Reviews & Analysis

This work involved the design of a multi-view manifold learning algorithm that capitalizes on various types of structure in high-dimensional time-series data to model dynamic signals in low dimensions. The resulting embeddings of human functional brain imaging data unveil trajectories through brain states that predict cognitive processing during diverse experimental tasks.

We present a computational method to generate a single-cell-resolution model of human brain regions starting from microscopy images. The developed method has been benchmarked to reconstruct the CA1 region of a right human hippocampus, including anatomical cell organization, connectivity, and network activity.

We propose a minimal and analytically tractable class of neural networks, the adaptive Ising class. By inferring the model’s parameters from resting-state brain activity recordings, we show that scale-specific oscillations and scale-free avalanches can coexist in resting brains close to a non-equilibrium critical point at the onset of self-sustained oscillations.

Determining whether a drug candidate has sufficient affinity to its target is a critical part of drug development. A purely physics-based computational method was developed that uses non-equilibrium statistical mechanics approaches alongside molecular dynamics simulations. This technique could enable researchers to accurately estimate the binding affinities of potential drug candidates.

We used computational models built using neural networks to predict what brain areas process the new meaning that emerges when words are combined. The brain activity evoked by this composed meaning was detected only with some brain recording modalities, a finding that might have consequences for brain–computer interfaces.

A universal interatomic potential for the periodic table has been developed by combining graph neural networks with three-body interactions. This M3GNet potential can perform structural relaxations, dynamic simulations and property predictions for materials across a diverse chemical space.

The surface energy cannot be assigned to each direction in low-symmetry crystals, making it impossible to predict their shapes by any known methods. Now, combining incomputable energies in an algebraic system, complemented by closure equations, it is possible to predict the equilibrium shape of any crystal.

We developed a computational method to reveal the drug-induced single-cell transcriptomic landscape. This algorithm enabled us to impute unknown drug-induced single-cell gene expression profiles using tensor imputation, predict cell type-specific drug efficacy, detect cell-type-specific marker genes, and identify the trajectories of regulated biological pathways while considering intercellular heterogeneity.

In social interactions, individuals are often tempted to ‘free ride’ (benefit without paying) on other group members’ contributions. A new mathematical framework suggests a strategy based on cumulative reciprocity that can help to sustain mutual cooperation.

Designing efficient bike path networks requires balancing multiple opposing constraints such as cost and safety. An adaptive demand-driven inverse percolation approach is proposed to generate efficient network structures by explicitly taking into account the demands of cyclists and their route choice behavior based on safety preferences.

We developed a machine learning method that consistently and accurately identified dominant patterns of disease progression in amyotrophic later sclerosis (ALS), Alzheimer’s disease and Parkinson’s disease. Of note, the model was able to identify nonlinear progression trajectories in ALS, a finding that has clinical implications for patient stratification and clinical trial design.

A systematic procedure is reported for calculating effective carrier lifetimes in semiconductor crystals from first-principles calculations. Consideration of three major recombination mechanisms and the use of realistic carrier and defect densities is key in resolving the discrepancy between experimental measurements and lifetimes calculated from nonadiabatic molecular dynamics simulations.

A machine learning method is developed and used to predict the adsorption configurations and energies of complex molecules at the surfaces of transition metals and alloys. This method will be useful for investigating complex reaction networks at complex catalyst materials to understand and improve the performance of heterogeneous catalysts.

Inspired by active learning approaches, we have developed a computational method that selects minimal gene sets capable of reliably identifying cell-types and transcriptional states in large sets of single-cell RNA-sequencing data. As the procedure focuses computational resources on poorly classified cells, active support vector machine (ActiveSVM) scales to data sets with over one million cells.

A cell clustering model for spatial transcripts that uses cell embedding obtained by graph neural networks can be applied to datasets from multiple platforms for cell type or subpopulation identification and further analysis of the spatial microenvironment.

A deep neural network method is developed to learn the density functional theory (DFT) Hamiltonian as a function of atomic structure. This approach provides a solution to the accuracy–efficiency dilemma of DFT and opens opportunities to investigate large-scale materials, such as twisted van der Waals materials.

Determining the origin of engineered DNA can help to foster responsible innovation within the biotechnology community. A convolutional neural network approach that learns distances between engineered DNA sequences and various labs that could have created them is used to accurately predict the lab-of-origin.

Biomimetic nanoparticles can form complexes with proteins. Structural descriptors have been identified to predict nanoparticle–protein complex formation and their interaction sites. These descriptors include geometrical and graph-theoretical molecular features that are universally applicable to all nanoscale macromolecules of both organic and inorganic chemistries.

Stochastic modeling of antibody binding dynamics on patterned antigen substrates suggests the separation distance between adjacent antigens could be a control mechanism for the directed bipedal migration of bound antibodies.

A fully automated, high-throughput computational framework accurately predicts stable species in liquid solutions by computing the nuclear magnetic resonance chemical shifts. Data collected from the framework can provide fingerprints to guide the rational design of liquid solutions with optimal properties.