Things to try

Some parasites use mimics of host proteins to manipulate host pathways. We’ve developed a mimicry detection pipeline and benchmarked it using well-studied viral mimics. The pipeline successfully recovers known mimics and is ready for deployment at scale.

We create vector graphics of model organisms and emerging biological research organisms to enhance our publications. We’re sharing these editable graphics under a CC0 license for other scientists to use in figures, slides, teaching materials, or other outputs.

Glass Box UMAP transforms UMAP into an interpretable embedding workflow by decomposing each point's coordinates into exact contributions from the original input features, letting users identify which variables drive clusters, gradients, and outliers in high-dimensional data.

Some parasites use mimics of host proteins to manipulate host pathways. We’ve developed a mimicry detection pipeline and benchmarked it using well-studied viral mimics. The pipeline successfully recovers known mimics and is ready for deployment at scale.

We applied a mechanistic interpretability approach to neural networks trained on a large yeast genotype–phenotype dataset. This allowed us to uncover novel evidence of environmentally mediated global epistasis in these data.

We applied a mechanistic interpretability approach to neural networks trained on a large yeast genotype–phenotype dataset. This allowed us to uncover novel evidence of environmentally mediated global epistasis in these data.

A significant portion of a macromolecular crystal is occupied by solvent. Since X-ray scattering arises from all atoms in a crystal, accurate solvent modeling is essential for reliable crystal structure refinement. In most crystallographic model-building protocols, water molecules of the first hydration shells are modeled as discrete, ordered atoms positioned in well-resolved electron density, and the remaining disordered solvent is approximated with a flat, binary mask. However, this approach has well-known shortcomings. Bulk solvent is not truly flat, the boundary between ordered and disordered solvent is nebulous, and regions near partially occupied protein, ligand, or solvent molecules are not appropriately accounted for, as these regions cannot fit cleanly into a binary bulk solvent. The inability to accurately model solvent remains one of the central unsolved challenges in macromolecular crystallography and is widely considered a major driver of the persistent gap between model and experiment. Alternative methods for generating bulk solvent masks include molecular dynamics simulations, 3D-RISM, holographic methods such as EDEN, and neural networks. But integrating these resulting models into standard refinement pipelines has remained a practical barrier. Here, we present a patch to Phenix refinement that allows users to supply their own bulk solvent map in place of the default flat mask. The modification is compact and applied locally to the user's Phenix installation, requiring no changes to the broader software infrastructure.

We post-trained a molecular graph neural network for predicting Raman spectra, extending its capabilities to short-chain peptides and drug-like molecules, and implemented a method to improve its predictions for experimental spectra with a small experimental dataset.

We assembled a comprehensive dataset of proteins from Asgard archaea and giant virus genome assemblies. This dataset lets us explore protein sequence and structure relationships more broadly across the tree of life to better understand protein structure and function.

We used Chlamydomonas reinhardtii motility mutants with disrupted genes homologous to human SPEF2 or DNALI1 to model spermatogenic failure disorders SPGF43 and SPGF83, respectively. We recovered aspects of wild-type motility in these mutants with a small-scale drug screen.

A significant portion of a macromolecular crystal is occupied by solvent. Since X-ray scattering arises from all atoms in a crystal, accurate solvent modeling is essential for reliable crystal structure refinement. In most crystallographic model-building protocols, water molecules of the first hydration shells are modeled as discrete, ordered atoms positioned in well-resolved electron density, and the remaining disordered solvent is approximated with a flat, binary mask. However, this approach has well-known shortcomings. Bulk solvent is not truly flat, the boundary between ordered and disordered solvent is nebulous, and regions near partially occupied protein, ligand, or solvent molecules are not appropriately accounted for, as these regions cannot fit cleanly into a binary bulk solvent. The inability to accurately model solvent remains one of the central unsolved challenges in macromolecular crystallography and is widely considered a major driver of the persistent gap between model and experiment. Alternative methods for generating bulk solvent masks include molecular dynamics simulations, 3D-RISM, holographic methods such as EDEN, and neural networks. But integrating these resulting models into standard refinement pipelines has remained a practical barrier. Here, we present a patch to Phenix refinement that allows users to supply their own bulk solvent map in place of the default flat mask. The modification is compact and applied locally to the user's Phenix installation, requiring no changes to the broader software infrastructure.

We used Chlamydomonas reinhardtii motility mutants with disrupted genes homologous to human SPEF2 or DNALI1 to model spermatogenic failure disorders SPGF43 and SPGF83, respectively. We recovered aspects of wild-type motility in these mutants with a small-scale drug screen.

A significant portion of a macromolecular crystal is occupied by solvent. Since X-ray scattering arises from all atoms in a crystal, accurate solvent modeling is essential for reliable crystal structure refinement. In most crystallographic model-building protocols, water molecules of the first hydration shells are modeled as discrete, ordered atoms positioned in well-resolved electron density, and the remaining disordered solvent is approximated with a flat, binary mask. However, this approach has well-known shortcomings. Bulk solvent is not truly flat, the boundary between ordered and disordered solvent is nebulous, and regions near partially occupied protein, ligand, or solvent molecules are not appropriately accounted for, as these regions cannot fit cleanly into a binary bulk solvent. The inability to accurately model solvent remains one of the central unsolved challenges in macromolecular crystallography and is widely considered a major driver of the persistent gap between model and experiment. Alternative methods for generating bulk solvent masks include molecular dynamics simulations, 3D-RISM, holographic methods such as EDEN, and neural networks. But integrating these resulting models into standard refinement pipelines has remained a practical barrier. Here, we present a patch to Phenix refinement that allows users to supply their own bulk solvent map in place of the default flat mask. The modification is compact and applied locally to the user's Phenix installation, requiring no changes to the broader software infrastructure.

We post-trained a molecular graph neural network for predicting Raman spectra, extending its capabilities to short-chain peptides and drug-like molecules, and implemented a method to improve its predictions for experimental spectra with a small experimental dataset.

We post-trained a molecular graph neural network for predicting Raman spectra, extending its capabilities to short-chain peptides and drug-like molecules, and implemented a method to improve its predictions for experimental spectra with a small experimental dataset.

Growing S. pombe liquid cultures with a bead yields more homogeneous, denser, and faster-growing cultures, without cell sedimentation. We found that, with a bead included, we could grow cultures in larger volumes and narrower vessels without sacrificing cell quality.

The process of deciding whether a candidate actin homolog represents a “true” actin is tricky. We propose clear and data-driven criteria to define actin that highlight the functional importance of this protein while accounting for phylogenetic diversity.

Peptigate predicts bioactive peptides from transcriptomes. It integrates existing tools to predict sORF-encoded peptides, cleavage peptides, and RiPPs, then annotates them for bioactivity and other properties. We welcome feedback on expanding its capabilities.