Journal of Open Source Software

UniRep is a recurrent neural network model trained on 24 million protein sequences, and has shown utility in protein engineering. The original model, however, has rough spots in its implementation, and a convenient API is not available for certain tasks. To rectify this, we reimplemented the model in JAX/NumPy, achieving near-100X speedups in forward pass performance, and implemented a convenient API for specialized tasks. In this article, we wish to document our model reimplementation process with the goal of educating others interested in learning how to dissect a deep learning model, and engineer it for robustness and ease of use.

The Internet of Things (IoT) provides a simple framework to easily control online devices. IoT is now a commonplace tool used by technology companies, but it is rarely used in biology experiments. IoT can benefit research through alarm notifications, automation, and the real-time monitoring of experiments. We developed and implemented an IoT architecture to control biological devices used in experiments. We developed our own electrophysiology, microscopy, and microfluidic devices so that may be controlled through a unified IoT architecture. The system allows each device to be monitored and controlled through an online web tool. We present our IoT architecture so other labs may replicate it for their own experiments.

Clearance of toxic species that may cause neurodegenerative diseases relies on convection and diffusion of mass around cells in the central nervous system. In this manuscript, methods that allow the nanoscale modeling of mass transport around cells in the brain at a resolution of 8 nm are described. The cells are modeled as surface meshes and a parallelization scheme is used to solve the diffusion equation directly on the surface mesh of each biological cell independently. This is followed by mass exchange between biological cells that are in direct contact. The analysis volume size is fixed at 4 m x 4 m x 4 m but arrays of analysis volumes of arbitrary size may be analyzed, with fluxes across analysis volume boundaries updated at each time step by a semi-implicit formulation. Setup of the discretized equations is described, along with the face matching that allows mass transfer between cells. Parallelization is via a manager/worker framework. One-sided RMA communication in MPI is used to coordinate the efforts of multiple workers, with common information handled by the manager. This specialized framework is suitable for analyzing diffusional transport around thousands of interacting objects simultaneously. The framework is implemented with custom code developed in Julia, which is used here as a rapid prototyping language. Using model geometries, the accuracy of the discretization methods are demonstrated, with limitations and next steps described. Authors summaryThe methods that enable modeling diffusional mass transport around thousands of objects in parallel are described. The accuracy of the method with model objects is demonstrated.

DNA nanotechnology is evolving rapidly, paralleling the historic trajectory of the 1970s electronics industry. However, current DNA nanostructure (DN) design software limits users to either manual design with minimal automation or a constrained range of automated designs. inS[e]quio Design Studio, developed by Parabon(R) NanoLabs, bridges this gap as a programmable 3D computer-aided design (CAD) application, integrating a domain-specific graphical editor with a Python API for versatile DN design. Developed in C++ for Windows(R) and Macintosh(R) systems, inS[e]quio features a user-friendly GUI with extensive CAD tools, capable of managing complex designs and offloading computational tasks to the cloud. It supports various DNA design formats, PDB molecule integration, residue modifications, and includes preloaded designs and thorough documentation. With its combination of features, inS[e]quio enables a code-centric design (CCD) approach, enhancing DN construction with improved precision, scalability, and efficiency. This approach is elucidated through a streptavidin barrel cage designed via Python notebook and a spheroid origami case study. Marking a significant advance in DN design automation, inS[e]quio, the first fully programmable 3D CAD tool for DN design, enables both manual and programmatic 3D editing. This fusion of features establishes inS[e]quio as a transformative tool, poised to significantly enhance designer productivity and expand the scope of possible designs. Extended AbstractAdvances in DNA nanotechnology have positioned the field at a juncture reminiscent of the pivotal growth phase of the electronics industry in the 1970s. The evolution of software for designing DNA nanostructures (DNs) is following a similar historical trajectory and dozens of software packages have been developed for creating them. Existing software options, however, require users to choose between manual design with minimal automation support or selecting from a limited set of designs, typically wireframe, that can be generated from a high-level structural description. Here, we introduce the inS[e]quio Design Studio, a programmable 3D computer-aided design (CAD) application that effectively bridges this gap. By integrating a domain-specific, freeform graphical editor with a Python application programming interface (API), inS[e]quio provides a comprehensive and extensible platform for designing complex nucleic acid (NA) nanostructures. The inS[e]quio desktop application, developed in C++, runs on Windows(R) and Macintosh(R) operating systems. Its graphical user interface (GUI) features multiple synchronized view panels and a diverse set of CAD and NA-specific editing tools. Its optimized graphics pipeline enables editing of designs with >2M nucleotides, and it includes an integrated service infrastructure for offloading heavy computations to cloud servers. The software also supports import and export of various DNA design file formats, integration of arbitrary PDB molecules, and specification of residue modifications. Additionally, it includes preloaded sample designs, scripts, and comprehensive documentation. Parabon has used evolving versions of inS[e]quio for over a decade to design a variety of proprietary DNs and have now transitioned it into a commercially available product. This paper summarizes inS[e]quios features, discusses its strengths and limitations, and outlines planned enhancements. Although freeform 3D design is well supported in inS[e]quio, the integration of its CAD environment with its API facilitates a code-centric design (CCD) approach for DN construction that offers notable productivity advantages over traditional methods, including enhanced precision, scalability, and efficiency. Here we describe CCD, outline its benefits and demonstrate its use through a well-documented Python notebook, included with the product, which generates a sample design within the inS[e]quio application. A spheroid origami created using CCD is also presented. As the first commercial fully programmable 3D CAD application specifically created for DN design, the release of inS[e]quio represents a milestone in the field of DN design automation. It introduces a new dimension to the discipline by enabling both manual and programmatic 3D editing, thereby facilitating an innovative CCD approach. The availability of extensive documentation and technical support enables designers to efficiently adopt and utilize these capabilities. This combination of features establishes inS[e]quio as a noteworthy addition to the tools available for DN design, with the potential to significantly increase designer productivity and broaden the scope of designs that can be developed by practitioners of all skill levels. Windows and Mac versions of the inS[e]quio desktop application are available for download at https://parabon.com/insequio. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=119 SRC="FIGDIR/small/586810v1_ufig1.gif" ALT="Figure 1"> View larger version (64K): org.highwire.dtl.DTLVardef@16a2714org.highwire.dtl.DTLVardef@2bc610org.highwire.dtl.DTLVardef@1d852e2org.highwire.dtl.DTLVardef@129b3fa_HPS_FORMAT_FIGEXP M_FIG C_FIG An illustration of the inS[e]quio Design Studio desktop application interoperating with a Python Jupyter notebook and molecular dynamics (MD) simulation tools to support an iterative code-centric design (CCD) process. The design cycle includes (a) programmatic and/or manual creation of objects in the inS[e]quio editors; (b) visual inspection and manipulation of objects via user interface; (c) in silico evaluation of designs via MD simulation using native or external tools; repeating a-c as necessary; and (d) procurement of strands and synthesis of DNA nanostructures (DNs).

Voltage-gated sodium ion channels allow for the initiation and transmission of action potentials. There is a high interest in research and drug development to selectively target these ion channels to treat epilepsy and other disorders such as pain. Scientific literature and presentations often incorporate maps of these integral membrane proteins with markers indicating gene mutations to highlight genotype/phenotype correlations. There is a need for automated tools to create high quality figures with mutation (variant) locations displayed on these channel maps. This manuscript introduces a simple application to create visualization for mutations on alpha voltage-gated sodium channels, created using the D3.js library. The application allows for mapping of variant sequences, as well as important properties like the type of variant and the phenotypes linked to the variant. It also allows for customizability and the production of high-quality images for publication. This application and code base can further be extrapolated to other ion channels as well.

Protein language models (pLMs) are ubiquitous across biological machine learning research, but state-of-the-art models like ESM2 take hundreds of thousands of GPU hours to pre-train on the vast protein universe. Resource requirements for scaling up pLMs prevent fundamental investigations into how optimal modeling choices might differ from those used in natural language. Here, we define a "cramming" challenge for pLMs and train performant models in 24 hours on a single GPU. By re-examining many aspects of pLM training, we are able to train a 67 million parameter model in a single day that achieves comparable performance on downstream protein fitness landscape inference tasks to ESM-3B, a model trained for over 15, 000x more GPU hours than ours. We open source our library1 for training and inference, LBSTER: Language models for Biological Sequence Transformation and Evolutionary Representation.

High throughput sequencing technologies have become essential in the fields of evolutionary biology and genomics. When dealing with non-model organisms or genomic gigantism, sequencing whole genomes is still relatively costly and therefore reduced-genome representations are frequently obtained, for instance by target capture approaches. While computational tools exist that can handle target capture data and identify small-scale variants such as single nucleotide polymorphisms and micro-indels, options to identify large scale structural variants are limited. To meet this need, we introduce PAV-spotter: a tool that can identify presence/absence variation (PAV) in target capture data. PAV-spotter conducts a signal cross-correlation calculation, in which the distribution of read counts per target between samples of different a priori defined classes - e.g. male versus female, or diseased versus healthy - are compared. We apply and test our methodology by studying Triturus newts: salamanders with gigantic genomes that currently lack an annotated reference genome. Triturus newts suffer from a hereditary disease that kills half their offspring during embryogenesis. We compare the target capture data of two different types of diseased embryos, characterized by unique deletions, with those of healthy embryos. Our findings show that PAV-spotter helps to expose such structural variants, even in the face of medium to low sequencing coverage levels, low sample sizes, and background noise due to mis-mapped reads. PAV-spotter can be used to study the structural variation underlying supergene systems in the absence of whole genome assemblies. The code, including further explanation, is available through the PAV-spotter GitHub repository: https://github.com/Wielstra-Lab/PAVspotter.

Turning SNP data into biologically meaningful results requires considerable computational acrobatics, including importing, exporting, and manipulating data among different analytical packages and programming environments, and finding ways to visualise results for data exploration and presentation. I present GENOMALICIOUS, an R package designed to provide a selection of functions for population genomicists to simply, intuitively, and flexibly, guide SNP data through their analytical pipelines, within and outside R. At the core of the original GENOMALICIOUS workflow is the conversion of genomic variant data into a data.table object. This provides a useful way of storing large amounts of data in an intuitive format that can be easily manipulated using methods unique to this object class. Over time, GENOMALICIOUS has grown to cater to a range of analyses in population structure and demography, adaptive evolution, quantitative traits, and phylogenetics. Researchers using pooled allele frequencies, or individually sequenced genotypes, are sure to find functions that accommodate their tastes in GENOMALICIOUS. The simplicity and accessibility of pipelines in GENOMALICIOUS may also serve as a useful tool for teaching basic population genetics and genomics in an R environment. The source code and a series of tutorials for this package are freely available on GitHub

The Digital Pedon (DP) is an open-source Python framework that represents a soil profile as a continuously updated digital twin, bridging three persistent gaps in soil science: disconnected models and observations, cross-database interoperability, and the inference gap between raw sensor signals and agronomically meaningful variables. Integrating real-time sensor streams, model-based solver chains (Model-Zoo), GLOSIS-compliant ontology mapping, and a novel LLM agentic interface layer enabling natural language soil queries, the DP supports applications spanning precision agriculture, digital soil mapping, and environmental sustainability assessment. Four proof-of-concept experiments confirm automatic profile initialisation fidelity, solver chain consistency, ontology compliance, and user-defined solver extensibility.

Microbioreactor (MBR) devices have emerged as powerful cultivation tools for tasks of microbial phenotyping and bioprocess characterization and provide a wealth of online process data in a highly parallelized manner. Such datasets are difficult to interpret in short time by manual workflows. In this study, we present the Python package bletl and show how it enables robust data analyses and the application of machine learning techniques without tedious data parsing and preprocessing. bletl reads raw result files from BioLector I, II and Pro devices to make all the contained information available to Python-based data analysis workflows. Together with standard tooling from the Python scientific computing ecosystem, interactive visualizations and spline-based derivative calculations can be performed. Additionally, we present a new method for unbiased quantification of time-variable specific growth rate [Formula] based on a novel method of unsupervised switchpoint detection with Student-t distributed random walks. With an adequate calibration model, this method enables practitioners to quantify time-variable growth rate with Bayesian uncertainty quantification and automatically detect switch-points that indicate relevant metabolic changes. Finally, we show how time series feature extraction enables the application of machine learning methods to MBR data, resulting in unsupervised phenotype characterization. As an example, t-distributed Stochastic Neighbor Embedding (t-SNE) is performed to visualize datasets comprising a variety of growth/DO/pH phenotypes. Practical ApplicationThe bletl package can be used to analyze microbioreactor datasets in both data analysis and autonomous experimentation workflows. Using the example of BioLector datasets, we show that loading such datasets into commonly used data structures with one line of Python code is a significant improvement over spreadsheet or hand-crafted scripting approaches. On top of established standard data structures, practitioners may continue with their favorite data analysis routines, or make use of the additional analysis functions that we specifically tailored to the analysis of microbioreactor time series. Particularly our function to fit cross-validated smoothing splines can be used for on-line signals from any microbioreactor system and has the potential to improve robustness and objectivity of many data analyses. Likewise, our random walk based [Formula] method for inferring growth rates under uncertainty, but also the time-series feature extraction may be applied to on-line data from other cultivation systems as well.

Transmission electron microscopy of brain tissue yields high resolution maps of the spatial organization of cells in the central nervous system. Automated segmentation identifies distinct biological cells and the resulting segmented images are easily converted into surface meshes. The surface meshes are structural models of cell surfaces, providing a framework for modeling of mass transport within the interstitial fluid of brain tissue that surrounds the cells. Our goal is to model the production and clearance of proteins implicated in the development of Alzheimers Disease. This work introduces a new custom computational framework to allow massive parallelization of the solution of the mathematical equations of mass transfer. The diffusion equation was solved directly on the surface meshes of multiple biological cells in parallel, with exchange of mass across co-localized faces at the end of each time step. Mass transfer across the interfaces of multiple analysis volumes was incorporated using a semi-implicit approach. To demonstrate the capabilities of the framework, unsteady mass transfer along cell surfaces was modeled in an array of eight 4 x 4 x 4 m analysis volumes containing 2175 biological cells at 8 nm resolution, consisting of 313 million face elements. Areas of enhanced and hindered diffusive transport were identified, suggesting structural motifs that may contribute to the development of insoluble plaques. The tortuosity and fractional anisotropy were consistent with Diffusion Tensor Imaging (DTI) measurements in cortical tissue. This new framework allows modeling of diffusive mass transport while preserving anatomical details at <10 nm resolution. Author summaryA framework was developed to model mass transport around thousands of biological cells in parallel. This may be useful to identify structural features in cortical tissue that are more prone to the development of amyloid plaques.

A multitude of methods have been developed to morph one image into another. In scientific and medical imaging, it is common to have lower resolution in one imaging direction. To improve the quality of surface meshes generated from these images, morphing is used to produce intermediate images to achieve equal resolution in all directions. The most common method is based on interpolating signed distance functions that describe the shapes of objects in the images. Most methods rely on identification and recording of object boundaries. The situation becomes more complex in the case of interpolating images containing multiple objects of interest wherein the morphing method must not introduce gaps or overlaps of objects in the intermediate images. The goal of the present work was to develop an interpolation method for biological cells in transmission electron microscopy images meeting the constraints listed above. A heat diffusion method was developed that produces steady state isotherms in a 3D model of the intervening space between two images. The isotherms are then used to produce paths from pixels that are unique to a biological cell in one frame towards the overlap region. Instead of identifying boundaries, only membership in the overlap region is tested. For each pixel that terminates paths, the path lengths of all members are normalized to the maximum path length. The path lengths are then used to produce intermediate frames. The method succeeds in avoiding the introduction of gaps or overlap between neighboring biological cells without the need for boundary identification.

Neuroscientists now have the opportunity to analyze synaptic resolution connectomes that are larger than the memory on single consumer workstations. As dataset size and tissue diversity have grown, there is increasing interest in conducting comparative connectomics research, including rapidly querying and searching for recurring patterns of connectivity across brain regions and species. There is also a demand for algorithm reuse -- applying methods developed for one dataset to another volume. A key technological hurdle is enabling researchers to efficiently and effectively query these diverse datasets, especially as the raw image volumes grow beyond terabyte sizes. Existing community tools can perform such queries and analysis on smaller scale datasets, which can fit locally in memory, but the path to scaling remains unclear. Existing solutions such as neuPrint or FlyBrainLab enable these queries for specific datasets, but there remains a need to generalize algorithms and standards across datasets. To overcome this challenge, we present a software framework for comparative connectomics and graph discovery to make connectomes easy to analyze, even when larger-than-RAM, and even when stored in disparate datastores. This software suite includes visualization tools, a web portal, a connectivity and annotation query engine, and the ability to interface with a variety of data sources and community tools from the neuroscience community. These tools include MossDB (an immutable datastore for metadata and rich annotations); Grand (for prototyping larger-than-RAM graphs); GrandIso-Cloud (for querying existing graphs that exceed the capabilities of a single work-station); and Motif Studio (for enabling the public to query across connectomes). These tools interface with existing frameworks such as neuPrint, graph databases such as Neo4j, and standard data analysis tools such as Pandas or NetworkX. Together, these tools enable tool and algorithm reuse, standardization, and neuroscience discovery.

SignificanceComputational modeling and the use of simulation software tools are essential for biomedical optics research. Designing effective simulations often requires in-depth understanding of the underlying physical problems and proper configuration of the software settings, which often constitute key barriers for novice users including students. The rapid emergence of large language models (LLMs) offers new opportunities for natural-language-based interaction, but integrating them with technical software remains challenging because of their limited output reproducibility. Overcoming these limitations would allow more intuitive, efficient, and reproducible interaction between scientists and scientific software. AimWe investigate the use of LLMs in quantitative biophotonics simulation tools, with a goal of enabling novice users to build complex photon simulations using intuitive natural-language-based problem descriptions. ApproachWe have explored prompt engineering strategies that enable LLMs to bridge the gap between natural language descriptions and advanced simulation software by constraining LLM outputs using a data schema (i.e., format) and a modular component architecture, followed by deterministic validation to ensure correctness and reproducibility of the outputs. ResultsUsing Monte Carlo eXtreme (MCX) - a widely used photon transport simulator - as an example, we showcase the capability of the proposed framework to convert user descriptions to structured simulation inputs. Benchmarked using 33 diverse natural language simulation descriptions, our LLM interface, MCX-LLM, achieves 98% accuracy and 99% repeatability, with an average processing time of 8.96 seconds per prompt. The framework also successfully handles various linguistic styles and diverse simulation settings, achieving a 100% success rate on 20 unconstrained real-world prompts. With only minor adjustments, our LLM interface also produces valid inputs for a finite-element-based diffusion solver to demonstrate generality towards other optical simulators. ConclusionsBy combining LLMs capability for textual data comprehension with structured constraints, this work provides a pathway to making complex scientific tools accessible while ensuring the reliability and technical correctness required for rigorous scientific research. MCX-LLM has been integrated with MCX Cloud accessible at https://mcx.space/cloud.

Visualisations facilitate the interpretation of geometrically structured data and results. However, heterogeneous geometries--such as volumes, surfaces, and networks--have traditionally mandated different software approaches. We introduce hyve, a Python library that uses a compositional functional framework to enable parametric implementation of custom visualisations for different brain geometries. Under this framework, users compose a reusable visualisation protocol from geometric primitives for representing data geometries, input primitives for common data formats and research objectives, and output primitives for producing interactive displays or configurable snapshots. hyve also writes documentation for user-constructed protocols, automates serial production of multiple visualisations, and includes an API for semantically organising an editable multi-panel figure. Through the seamless composition of input, output, and geometric primitives, hyve supports creating visualisations for a range of neuroimaging research objectives.

Human immunogenetic variation in the form of HLA and KIR types has been shown to be strongly associated with a multitude of immune-related phenotypes. We present MiDAS, an R package enabling statistical association analysis and using immunogenetic data transformation functions for HLA amino acid fine mapping, analysis of HLA evolutionary divergence as well as HLA-KIR interactions. MiDAS closes the gap between inference of immunogenetic variation and its efficient utilization to make meaningful discoveries.

Computational models are central to cognitive neuroscience, but their rigorous application to experimental datasets is often constrained to a narrow set of canonical models that afford tractable analytical computations. We introduce the HSSM (Hierarchical Sequential Sampling Model) ecosystem, a Python toolbox that democratizes access to a broad, extensible array of neurocognitive process models through hierarchical Bayesian inference. Naturally leveraging simulation-based inference via likelihood surrogates, HSSM enables fast parameter estimation for models lacking closed-form likelihoods. Built atop PyMC and Bambi, HSSM provides a user-friendly formula syntax for specifying hierarchical mixed-effects regressions on model parameters, incorporating trial-by-trial neural or physiological covariates. The ecosystem allows fast model simulation and training data generation, as well as the neural network training utilities to deploy surrogate likelihood networks via HuggingFace. Contributions are designed to benefit not only the single researcher working on a problem, but organically, the entire research community. Together, the tools in the HSSM ecosystem bridge the interests of computational theorists as well as experimentalists, accelerating the cycle from model development to rigorous empirical testing.

MotivationExecution of code is critical for computational biology, but technical requirements can prevent others from running it. Public web-apps and services thus remain the most effective way to make code accessible, but no fully reusable infrastructure exists to help researchers do this. ResultsWe developed Slivka to enable easy provision of robust HTTP-based execution services backed by local or distributed hardware; accessible via curl and dedicated clients. We demonstrate it with Slivka-bio, which provides semantically annotated services for Jalview 2.12 (https://www.jalview.org/development/jalview_develop/) and includes 15+ tools for protein and RNA analysis. Slivka has been in production in academic and industry environments for 5 years and ran more than 1.5M jobs. Availability and ImplementationSlivka and Slivka-bio are released under the Apache 2.0 License. Slivka-bio public instance at https://www.compbio.dundee.ac.uk/slivka with links to documentation, docker containers, and github repositories for Slivka-bio and Slivka. Contactj.procter@dundee.ac.uk and g.j.barton@dundee.ac.uk.

SummaryHigh-throughput analysis of behaviour is a pivotal instrument in modern neuroscience, allowing researchers to combine modern genetics breakthrough to unbiased, objective, reproducible experimental approaches. To this extent, we recently created an open-source hardware platform (ethoscope (Geissmann et al., 2017)) that allows for inexpensive, accessible, high-throughput analysis of behaviour in Drosophila or other animal models. Here we equip ethoscopes with a Python framework for data analysis, ethoscopy, designed to be a user-friendly yet powerful platform, meeting the requirements of researchers with limited coding expertise as well as experienced data scientists. Ethoscopy is best consumed in a prebaked Jupyter-based docker container, ethoscope-lab, to improve accessibility and to encourage the use of notebooks as a natural platform to share post-publication data analysis. Availability and implementationEthoscopy is a Python package available on GitHub and PyPi. Ethoscope-lab is a docker container available on DockerHub. A landing page aggregating all the code and documentation is available at https://lab.gilest.ro/ethoscopy.

MotivationIllustrative renderings of proteins are useful aids for scientific communication and education. Nevertheless, few software packages exist to automate the generation of these visualizations. ResultsWe introduce CellScape, a tool designed to generate 2D molecular cartoons from atomic coordinates and combine them into larger cellular scenes. These illustrations can outline protein regions in different levels of detail. Unlike most molecular visualization tools which use raster image formats, these illustrations are represented as vector graphics, making them easily editable and composable with other graphics. Availability and ImplementationCellScape is implemented in Python 3 and freely available at https://github.com/jordisr/cellscape. It can be run as a command-line tool or interactively in a Jupyter notebook. Contactjordisr@berkeley.edu