Matt W. Thomson

The paper uses lineage tracing to show that fate restriction in mouse embryos and stem cell-derived models is asynchronous and not solely determined by cell identity.

This paper introduces CIFM, a generative virtual tissue model that simulates transcriptional dynamics under therapeutic conditions to design immunotherapy strategies for cancer and autoimmune diseases.

This study identifies second-order spatial features, such as ligand-receptor interactions, as predictive of glioma malignant transformation using integrated spatial DNA and RNA profiling.

The paper introduces a method for high-fidelity, multiplexed DNA construction from oligo pools, enabling efficient assembly of multiple defined constructs.

The paper measures ATP gradients in motor-microtubule assemblies and shows they are sustained by energy consumption, with implications for cytoskeletal organization.

The paper decodes BMP/TGFβ signaling from single-cell transcriptomes by identifying a low-dimensional transcriptional code and training a model to infer ligand concentrations.

CONCORD is a self-supervised framework that integrates single-cell datasets across batches, technologies, and species, producing high-resolution cell atlases with improved biological resolution and reduced noise.

This paper uses protein language models to discover new-to-nature protein domains with structures outside known CATH superfamilies, suggesting natural protein structure-space is incomplete.

This paper shows that protein language models prioritize sequence novelty over structural diversity, leading to a tradeoff in generating realistic protein structures.

The paper identifies a noise-driven differentiation switch mechanism that controls cell type diversity in neural progenitor cells through combinatorial signaling inputs.

The paper develops a cell-free platform to study how kinesin motor sequences influence microtubule dynamics, identifying distinct phenotypes and showing that recombination can alter dynamics.

This paper shows that increasing microtubule bundle length in active networks triggers a percolation transition, enabling global force generation and mechanical work.

D-SPIN constructs regulatory network models from single-cell perturbation data by integrating multiple conditions and inferring gene or program-level interactions.

The paper investigates motor-driven microtubule network contraction and identifies a diffusive-like spread within the network, showing a conserved relationship between contraction rates and diffusion constants.

This paper introduces CircuiTree, a Monte Carlo Tree Search framework for designing biochemical circuits, finding robust designs for oscillators and identifying novel fault-tolerant strategies.

This study benchmarks protein language models for predicting protein crystallization propensity and identifies novel potentially crystallizable proteins.

Protein CREATE generates and tests synthetic protein binders using a closed-loop computational-experimental pipeline, identifying novel binders and binding features.

High school students designed novel dehalogenases using large language models to degrade PFOA, a persistent PFAS compound, and plan to test their enzymatic activity.

This paper demonstrates that predictive coding can construct spatial maps from sensory data, enabling agents to navigate using internal representations of environments.

This study identifies an active mechanical mechanism involving PDGFRA+ mesenchymal cells that pattern and fold intestinal villi through myosin II-dependent forces and matrix metalloproteinase-mediated tissue fluidization.

The paper uses spatial transcriptomics to identify injury-specific microenvironments and molecular interactions in the mouse kidney during injury and regeneration.

TRILL enables scalable, modular deep-learning workflows for protein analysis and engineering by integrating models like ESM-2 and DiffDock.

The paper addresses missing gene expression data in scRNA-seq by improving read mapping, revealing more genes and biological insights in mouse and human datasets.

The paper introduces TopGen, a framework using persistent homology to analyze topological structures in single-cell transcriptomes, revealing non-trivial topological motifs in developmental processes.

This study shows how cell density affects signal transduction in synthetic gene circuits, enabling spatial and temporal control of multicellular patterning.

This study identifies how changes in epithelial proportions and transcriptional states influenced by pregnancy and BMI affect hormone signaling and breast cancer risk in premenopausal women.

The paper shows that optimizing receptor localization in cells enhances information acquisition in spatially structured environments, improving navigation efficiency by 30-fold through dynamic redistribution.

The paper identifies DNA base-excision repair as a mechanism that amplifies transcriptional noise, enhancing cellular plasticity and reprogramming efficiency.

ActiveSVM identifies minimal gene sets for accurate cell-type and disease state classification in single-cell mRNA-seq data, achieving 90% accuracy across diverse datasets.