Our Work

Large Language Models have extended beyond natural language processing tasks such as language generation, becoming transformative tools in advancing the field of structural biology. However, tradition…

Large Language Models have extended beyond natural language processing tasks such as language generation, becoming transformative tools in advancing the field of structural biology. However, traditional large language models (LLMs) predominantly employ subword tokenization techniques (e.g., Byte-Pair Encoding), followed by complex pre- tokenization schemes. In contrast, most protein language models (PLMs) typically rely on character-level tokenization, treating each amino acid individually. This design choice is often a byproduct of the downstream tasks of interest, which are predominantly defined at the residue level. Our key findings reveal a divergence between the results of intrinsic and extrinsic evaluations. Amino acid-level PLMs perform better in terms of normalized perplexity on a test partition of the pretraining data. Among the intrinsic metrics, minimum description length is the one that most closely aligns with downstream performance, favouring a vocabulary size of around 4,000 tokens for proteins and larger vocabularies for natural language. Meanwhile, an intermediate vocabulary size of 1,000 is optimal for Deeploc’s downstream classification tasks. The results suggest that tokenization in PLMs is task-dependent. We further demonstrate that BPE tokenization recovers biologically meaningful sequence patterns such as nuclear localization signals and hydrophobic repeats. Additionally, we examine the impact of pre-tokenisation on the pre-training of LLMs and present evidence that the pre-tokeniser is an effective regularization tool, a prior that is currently absent in PLMs.

This article examines the development of Euroscepticism in contemporary France. France’s attitude toward the European Union has always been ambivalent. Concerns regarding the EU’s impact o…

This article examines the development of Euroscepticism in contemporary France. France’s attitude toward the European Union has always been ambivalent. Concerns regarding the EU’s impact on France’s socio-economic model, as well as fears of losing sovereignty and cultural identity, have been present and clearly articulated at key moments of European integration. Opposition to the EU in France is not monolithic; it manifests in various forms rooted in diverse political ideologies. Contemporary France is characterized by the rising prominence of both far-right and far-left Eurosceptics, represented by the National Rally (*Rassemblement National*) and France Unbowed (*La France Insoumise*). Despite their ideological polarization, both parties are experiencing steady growth in voter support - a trend directly linked to a profound crisis of confidence among the French public regarding European Union institutions.

R scripts used for statistics and figure creation in Staudt, Tran et al

Prefață Trăim într-o lume plină de tipare ascunse. De la felul în care corpul își reglează temperatura, până la modul în care învățarea, prieteniile, ecosistemele s…

Prefață Trăim într-o lume plină de tipare ascunse. De la felul în care corpul își reglează temperatura, până la modul în care învățarea, prieteniile, ecosistemele sau piețele economice se transformă în timp, aceleași structuri simple se repetă peste tot. Aceste structuri se numesc bucle — uneori mențin echilibrul, alteori creează schimbare. Cartea de față reunește două părţi complementare dezvoltate în cadrul Dr Mihăilă Graphic System®, un sistem vizual conceput pentru a face gândirea sistemică accesibilă copiilor, adolescenților și educatorilor. Fiecare parte prezintă un set de diagrame intuitive care dezvăluie „codurile secrete” ale lumii: buclele de echilibru, care stabilizează sistemele, și buclele de amplificare, care generează spirale ascendente sau descendente. Aceste modele nu sunt doar instrumente educaționale. Ele sunt o lentilă prin care cititorii pot înțelege mai clar dinamica vieții — de la biologie la economie, de la comportamente personale la fenomene sociale. Într-o epocă în care complexitatea pare copleșitoare, Dr Mihăilă Graphic System® oferă un limbaj simplu, vizual și universal pentru a naviga lumea. Această lucrare este o invitație la curiozitate, la observarea tiparelor și la descifrarea mecanismelor invizibile care ne modelează viața. Este un ghid pentru viitorii oameni de știință, pentru educatorii care cultivă minți curioase și pentru părinții care vor să le ofere copiilor lor o înțelegere profundă a lumii.   Preface We live in a world filled with hidden patterns. From the way the body regulates its temperature to the way learning, friendships, ecosystems, or economic markets change over time, the same simple structures repeat themselves everywhere. These structures are called loops — sometimes they maintain balance, other times they create change. This book brings together two complementary parts developed within the Dr Mihăilă Graphic System®, a visual system designed to make systems thinking accessible to children, teenagers, and educators. Each part presents a set of intuitive diagrams that reveal the world’s “secret codes”: balancing loops, which stabilize systems, and reinforcing loops, which generate upward or downward spirals. These models are more than educational tools. They are a lens through which readers can better understand the dynamics of life — from biology to economics, from personal behaviors to social phenomena. In an age when complexity can feel overwhelming, the Dr Mihăilă Graphic System® offers a simple, visual, and universal language for navigating the world. This work is an invitation to curiosity, to observing patterns, and to uncovering the invisible mechanisms that shape our lives. It is a guide for future scientists, for educators who nurture curious minds, and for parents who want to offer their children a deeper understanding of the world.  

This dataset contains the scripts and data required to reproduce Figures 1–7 for the study of the extreme rainfall and flood event in Bahía Blanca (Argentina) that occurred on 6–7 M…

This dataset contains the scripts and data required to reproduce Figures 1–7 for the study of the extreme rainfall and flood event in Bahía Blanca (Argentina) that occurred on 6–7 March 2025. Included data: IMERG precipitation series, Integrated Vapor Transport (IVT) anomalies, vertical profiles of omega, and divergence fields in NetCDF and CSV formats.Scripts: Python code for data processing and visualization.

Two Masters students at Swansea University, Jess and Ellie, demonstrate how to count in binary using hand-signals. This activity is designed to demonstrate how to develop multiple intelligences in one…

Two Masters students at Swansea University, Jess and Ellie, demonstrate how to count in binary using hand-signals. This activity is designed to demonstrate how to develop multiple intelligences in one activity. Students are employing and building social and collaborative skills in person, to count together, they are problem-solving using  skills and insight in mathematics and gaining insight into computational science.   Contact Berry Billingsley, Professor of AI and Education at Swansea University for more resources to support the activity  - berry.billingsley@swansea.ac.uk

Numerical data, publication figures, calculation and plotting scripts, symbolic derivation notebooks, and convergence/provenance records supporting Figures 1–14 and S1–S3 and Tables I and …

Numerical data, publication figures, calculation and plotting scripts, symbolic derivation notebooks, and convergence/provenance records supporting Figures 1–14 and S1–S3 and Tables I and S1–S3 of "Quasiparticle-resolved variational theory of Andreev spin qubits". Replotting uses supplied numerical data; fresh numerical regeneration is available for the documented calculation-backed results. Symbolic regeneration requires separately installed Wolfram and SNEG. Supplied NRG and DMRG references are archived numerical inputs. Data, figures, and documentation are licensed under CC BY 4.0; source code is licensed under BSD-3-Clause. See LICENSE.md inside the archive for applicability.

1.302.2 (2026-10-02) Bug Fixes pipeline-form: tidy composer cards scroll and reorder (#1460) (48854fc6) Code Refactoring webapp: drop @forkbombeu/temporal-ui (#1465) (2a6d5338) Chores release: 🚀 …

1.302.2 (2026-10-02) Bug Fixes pipeline-form: tidy composer cards scroll and reorder (#1460) (48854fc6) Code Refactoring webapp: drop @forkbombeu/temporal-ui (#1465) (2a6d5338) Chores release: 🚀 update VERSION to 1.302.2 (2dcaff6f)

This small repository contains materials to include outgassing in a spacecraft charging tool, to allow reproductability of the study : Outgassing modelling and impacts on surface charging during BepiC…

This small repository contains materials to include outgassing in a spacecraft charging tool, to allow reproductability of the study : Outgassing modelling and impacts on surface charging during BepiColombo cruise phase.  It includes a tutorial to set up the software. A device and code modification to include in the software package.  This repository is created before the While an adequate updated version of SPIS is about to be released to the public, in the meantime, this can be used for SPIS version above 6.2.0.