Docking protocol development and virtual screening often require researchers to assemble separate preparation, execution, and analysis tools into custom workflows. DockMate-VS is an open-source Python application that integrates these tasks through spreadsheet-driven graphical and headless workflows. Its protocol-development stage compares docking engines, binding-site definitions, water treatment, search effort, and rescoring methods, with separate reporting of pose generation, pose ranking, and runtime. Selected protocols can then be applied in resumable, multi-target screening campaigns with labelled controls, per-receptor enrichment analysis, physicochemical diagnostics, and explicit failure reporting. Machine-readable outputs and run manifests retain the information needed to inspect, reproduce, and extend analyses. Worked examples using an ESR1 antagonist dataset demonstrate protocol comparison and the interpretation of screening metrics in the presence of molecular-size confounding. A matched ACHE–DEKOIS 2.0 screening experiment reproduces archived Smina scores, rankings and enrichment metrics using the authors’ prepared inputs. By connecting protocol configuration, evaluation and campaign execution, DockMate-VS reduces the need for custom orchestration scripts and supports research and training across different levels of computational experience.
An open-source software framework, Program for Structural Design (P4Struct), is developed for topology optimization. P4Struct provides an interactive interface and data management system, allowing flexible configuration of design parameters. Taking meshed models as input, P4Struct provides an integrated workflow that encompasses preprocessing, finite element analysis, topology optimization, and post-processing, ultimately generating STL models. It supports linear elastic analysis of 2D and 3D structures, upon which single-objective optimization can be performed based on arbitrary combinations of strain energy, volume, and von Mises stress. The accuracy of P4Struct is verified via numerical tests, and several examples are presented to demonstrate its applicability.
Relational event models analyse ordered interaction data in which event timing, sequence, and available alternatives contain information that is not preserved in an aggregated network. Applying these models usually requires several connected steps, including event-history preparation, risk-set construction, statistic calculation, estimation, and model checking. REMFlow is an open-source Python library that integrates these stages into one workflow for sender-receiver event data. It supports selected tie-oriented, actor-oriented, exact-time, ordinal, and duration models. NumPy provides the default CPU backend, while selected likelihood calculations can use JAX on a CPU or GPU. The library also provides readable result tables and basic diagnostic plots. Its use is demonstrated through a classroom interaction tutorial, a controlled exact-time simulation, and a typed-event workflow. The classroom analysis presents nested model comparison, coefficient interpretation, and event-rank diagnostics. The simulation shows that numerical convergence does not guarantee parameter identification. The typed-event example explains how action types extend the risk set and how supplied actor attributes enter descriptive statistics. Benchmark results show numerical agreement between NumPy and JAX and substantial GPU speedups for repeated likelihood-and-gradient evaluations. Full directed risk sets grow quadratically with actor count and rapidly increase dense storage. REMFlow supports case-control sampling for selected tie-oriented models while documenting the remaining preprocessing, computational, and model-family limits. An independent comparison with relevent evaluates all eight classroom fits and the same simulated event history under explicitly matched specifications.
This article presents the new major release of flow-models, our framework for analyzing IP network flows. The main addition is an interactive web interface, available at www.flow-models.org, which turns the repository into a browsable online resource. It provides hierarchical model selection, interactive plots with tooltips, zooming and curve isolation, histogram and time-series viewers, previews of binary flow records, and visualization of experimental results, covering both theoretical analysis and machine learning-based elephant detection. Supporting changes include a documented dataset convention, new output formats, and modules for series charts and record previews.
In this article, we present a major update to Parallel JPEG with RDOQ (PJRD). PJRD is a parallel JPEG encoder with rate-distortion-optimized quantization (RDOQ) and Huffman table optimization, accelerated with platform-specific SIMD instructions. The new version 3.0 introduces an arithmetic coding path compliant with ISO/IEC 10918–1. Moreover, to enable RDOQ for arithmetic coding, a bit-count estimator is proposed. Therefore, the new version of the software is a unique combination of JPEG arithmetic coding, RDOQ, and parallel processing. Compared with Huffman coding, the arithmetic path reduces the bitrate by 8–25%, and on a 16-core processor the encoder and decoder achieve speedups of up to 23 × and 19 ×, respectively.
