The questions.
The science.

Molecular recognition begins with a specific interaction: a guest entering a host cavity, a probe responding to a chemical reaction, or a gas moving between airspaces and blood. We investigate how these processes can distinguish chemical environments and report on tissue function.

Our collaborations on metal-organic frameworks use the local environment of xenon to produce distinguishable chemical shifts. The 2026 ZIF-8 study extends this idea to polymeric surfaces, demonstrating spectrally selective coatings in phantoms, a flow simulator, and a rat airway. Related work examines pulmonary gas exchange and glucose metabolism. These are distinct biological questions, connected by the chemistry of molecular exchange.

Approaches

Host-guest chemistry; metal-organic frameworks; hyperpolarized ¹²⁹Xe MRI; chemical exchange saturation transfer; dynamic glucose-enhanced magnetic resonance fingerprinting.

Questions to explore

Relate pore chemistry to molecular exchange and spectral selectivity. Test the stability and specificity of reporters in biological environments. Separate changes in tissue metabolism or gas exchange from changes in probe delivery.

Transport through a confined or heterogeneous material depends on more than a single diffusion coefficient. Molecular collisions, boundaries, and memory of earlier motion can all influence the response. We study how these dynamics connect microscopic structure to measurable behavior.

Our line-shape studies couple molecular dynamics to generalized Langevin descriptions. In simulations of xenon confined in carbon nanotubes, the resulting signal depends on temperature and nanotube geometry. This provides a route to structural inference using weak gradients, while making the underlying transport assumptions explicit. Complementary mathematical work develops weak formulations for spin dynamics with long-range dipolar interactions.

Approaches

Molecular dynamics; stochastic differential equations; generalized Langevin models; Green-Kubo correlations; NMR line-shape analysis; Bloch equations with distant dipolar fields.

Questions to explore

Determine which geometric features are identifiable from a signal. Extend transport models to realistic boundaries and mixtures. Quantify the limits of a continuum description when molecular correlations matter.

Immune cells respond to the mechanics of their surroundings as well as to biochemical signals. Our collaborative research examines how three-dimensional architecture, mechanical stimulation, antigen presentation, and local delivery of immune modulators influence T-cell behavior.

Studies of engineered microenvironments connect material mechanics to T-cell activation. In preclinical tumor models, peritumoral macroporous scaffolds provide a local setting for immunomodulation with systemic antitumor effects. Related biomaterial work examines memory-associated responses in aged mice. The goal is to understand and control local cues, not to assume that a response observed in a model will translate directly to patients.

Approaches

Three-dimensional biomaterials; macroporous scaffolds; engineered antigen-presenting cells; controlled cytokine delivery; mechanical stimulation; cell-state and functional measurements.

Questions to explore

Separate the effects of scaffold mechanics and biochemical presentation. Connect local cell-state changes to longer-term immune responses. Quantify cell viability and density independently in three-dimensional cultures.

Chirality-induced spin selectivity raises fundamental questions about the connection between molecular geometry, electronic states, and magnetic response. We develop theories that distinguish a microscopic mechanism from the signal that a particular experiment or contact arrangement can actually detect.

Our work explores electronic mediation of nuclear-spin interactions, chirality-encoded wavefunctions, and relativistic and many-body contributions. The 2026 effective-medium treatment asks how microscopic spin-charge response survives disorder and orientational averaging. For the rank-one longitudinal model studied, isotropic and head-tail-symmetric orientational distributions cancel that contribution even in an enantiopure material. Predicted bulk conversion and a measured terminal voltage are not interchangeable: boundary conditions and contacts matter.

Approaches

Effective spin Hamiltonians; density-functional theory; indirect nuclear-spin coupling; relativistic and electron-phonon interactions; homogenization; open-system linear response.

Questions to explore

Identify symmetry constraints on observable responses. Connect molecular calculations to disordered materials. Design tests that distinguish chirality-dependent mechanisms from sample preparation and measurement artifacts.

Spins can retain quantum information, but interactions that enable control can also introduce noise and unwanted coupling. We study this balance in individual spins, spin ensembles, and multilevel quantum systems.

Our 2026 rare-earth-qubit theory proposes local ligand symmetries that suppress anisotropic exchange while retaining isotropic coupling; the reported protection is a theoretical and computational design result. Other work addresses entanglement in interacting ensembles and geometric advantages in quantum control. Nanoscale magnetic textures offer another route to spatially localized control, while qudit encodings explore ways of mitigating decoherence at the logical level.

Approaches

Spin Hamiltonians; first-principles calculations; quantum-control geometry; numerical gate optimization; nanomagnet and skyrmion field models; qudit encodings; noise and leakage analysis.

Questions to explore

Test how symmetry protection tolerates placement disorder. Quantify control fidelity, leakage, and cross-talk. Compare the resources required by alternative encodings and control paths.

Age-associated changes are distributed across many genes and methylation sites, and their trajectories need not be linear. We investigate representations and statistical models that capture this structure without treating prediction accuracy as proof of a biological mechanism.

BayesAge 2.0 models RNA-sequencing counts with a Poisson likelihood and nonlinear gene-age relationships; MicroBayesAge adapts the framework to methylation microarrays. A multi-tissue killifish atlas examines how transcriptomic aging differs across tissues. A 2026 preprint adds conformal prediction intervals to BayesAge, with coverage guarantees dependent on exchangeability. In parallel, Embed-Search-Align studies learned DNA representations for sequence alignment.

Approaches

Maximum-likelihood age estimation; nonlinear smoothing; RNA-seq and methylation data; conformal prediction; transformer-based sequence embeddings; vector search and alignment.

Questions to explore

Evaluate age bias and uncertainty under cohort shifts. Compare shared and tissue-specific aging signatures. Develop sequence representations that preserve biologically meaningful similarity.

Landslides emerge from interacting controls that vary across space and time. Working with Earth scientists, we investigate how terrain, precipitation, and their interactions shape susceptibility, and how to build models whose predictions can be examined in physical terms.

Our interpretable-neural-network study examined landslide inventories in three easternmost Himalayan regions, identifying slope-climate coupling and aspect as important contributors. Susceptibility describes where failure is more likely; it is not an event-time forecast. Current directions extend this work toward geomorphic world models, self-supervised spatial representations, and tests of transfer to geographically separated regions. Related theoretical work studies whether solitary-wave modes are admissible in models of long-runout debris flows.

Approaches

Interpretable additive neural networks; geospatial sensor data; terrain and climate covariates; spatially blocked evaluation; representation learning; continuum models of debris flows.

Questions to explore

Separate genuine spatial transfer from information leakage. Evaluate what changes when training and test regions differ. Connect model representations to geomorphic processes and quantify uncertainty in dynamic environments.