Leiden University Scholary Publications
Not a member yet
131438 research outputs found
Sort by
Evaluating bindcraft for generative design of high-affinity peptides
Full text linkDiscovering high-affinity ligands directly from protein structures remains a key challenge in drug discovery. BindCraft is a structure-guided generative modeling platform able to de novo design miniproteins with a high affinity for a large set of targets. While miniproteins are valuable research tools, short peptides offer substantially greater therapeutic potential. However, given their lack of stabilized tertiary structures, de novo generation of functional peptides is a remarkable challenge. Here, we show that BindCraft is able to generate high affinity peptides, solely based on target structure, with remarkable success rates. For the oncoprotein MDM2, BindCraft generated 70 unique peptides; 15 were synthesized, and 7 showed specific binding with nanomolar affinities. Competition assays confirmed site-specific binding for the intended target site. For another oncology target, WDR5, six out of nine candidates bound the MYC binding WBM site with submicromolar affinity. Bindcraft's high fidelity structure prediction enabled one shot peptide optimization via rational chemical modification, improving the potency of one WDR5 binder by 6-fold to a KD of 39 nM. BindCraft also generated candidate peptides for targeting PD-1 and PD-L1. However, none of the tested peptides showed detectable binding. Together, these results establish a first evaluation of BindCraft for peptide binder prediction. Despite remaining limitations, this tool shows the potential to rival display technologies in delivering high-affinity ligands for therapeutic development.Medicinal Chemistr
Clinical reasoning by pharmacists: fostering clinical decision-making and interprofessional collaboration in pharmacy practice and education
Full text linkClinical reasoning is a core competence for pharmacists and forms the foundation for effective clinical decision-making (CDM)–a complex interplay of cognitive processes and skills that enable pharmacists to make patient-centered clinical decisions in pharmacy practice. Yet, how pharmacists reason and make clinical decisions is less understood than in other health professions, posing teaching and learning challenges. This dissertation aims to enhance the understanding of clinical reasoning by pharmacists, identify the cognitive processes underlying CDM, examine factors influencing decision-making, and evaluate educational interventions that foster both CDM and interprofessional collaboration (IPC). Findings show that pharmacists use a combination of analytical and intuitive reasoning to interpret clinical data in context. Interviews identified 21 cognitive processes organized into eight CDM steps, highlighting challenges in contextual reasoning and limited attention to outcome evaluation and reflection. These insights informed the development of a pharmacy-specific CDM model and learning guide to support educators and learners across settings. Surveys indicated that both students and pharmacists found the model practical and supportive. Finally, an interprofessional pharmacotherapy program showed promise in enhancing IPC competencies. Collectively, this dissertation provides a comprehensive framework for understanding and teaching CDM, offering tools to foster more reflective, collaborative, and patient-centered pharmacy practice.This dissertation received an unconditional grant from the Royal Dutch Pharmacists Association (‘Koninklijke Nederlandse Maatschappij ter bevordering der Pharmacie’ (KNMP)).LUMC / Geneeskund
New dimensions of the cellular response to DNA damage
Full text linkThirty years ago, mutations in the BRCA1 gene were first linked to hereditary breast and ovarian cancer, establishing a genetic basis for cancer development. BRCA1 was soon recognized as essential for repairing DNA double-strand breaks through homologous recombination (HR), a process critical for maintaining genome stability. This thesis investigates the mechanistic underpinnings of HR and evaluates therapeutic strategies for HR-deficient cancers, including platinum-based chemotherapy and PARP inhibitors (PARPi), while proposing approaches to overcome therapy resistance.We explore the poorly understood late stages of HR, identifying FIRRM as a crucial factor in resolving HR intermediates. Loss of FIRRM causes sensitivity to DNA-damaging agents, chromosomal instability, and accelerated tumorigenesis, suggesting its role as a predictor of therapy response. Furthermore we examine biomarkers of HR deficiency in triple-negative breast cancer, showing that shallow genome sequencing effectively predicts platinum sensitivity and identifying XRCC3 and ORC1 mutations as novel predictors. We review PARPi therapy, resistance mechanisms, and strategies to enhance efficacy. Finally, we revealed that PARPi disrupt chromatin by evicting histones, a process mediated by NASP and PARP1, and demonstrates that targeting histone supply pathways increases PARPi sensitivity. The thesis concludes by highlighting future directions to advance synthetic lethality strategies against HR-deficient tumors.The research described in this thesis was performed at the division of Molecular Pathology at the Netherlands Cancer Institute – Antoni van Leeuwenhoek Hospital (NKI-AVL) (Amsterdam, the Netherlands). The research was financially supported by a Boehringer Ingelheim Fonds PhD fellowship, the Dutch Cancer Society and Oncode Institute.Drug Delivery Technolog
Dodo through the looking-glass: a mirror for modern and contemporary culture
Full text linkInaugural Lecture by Prof.dr. Nidesh Lawtoo on the acceptance of his position as Professor Modern/Contemporary European Literature and Culture at Leiden University on Friday, 14 November 2025Modern and Contemporary Studie
The meaning and demeaning of political ideals: on Quentin Skinner's Liberty as Independence
No full textFGW – Publications not associated with a particular research are
Judicial transformation: the case of the Polish constitutional tribunal
No full textSecurity and Global Affair
Vrijheid van meningsuiting en de bestrijding van discriminatie in Nederland: een constitutioneel ontwikkelingsperspectief
Full text linkThis thesis concerns the use of criminal law to restrict freedom of expression in public debate, a topic that has been the subject of both societal and academic interest since the conviction of politician Geert Wilders in 2020. A notable aspect of this case was the application of a consideration by the Dutch Supreme Court in which repressive action against 'hate speech' – criminalised in the Netherlands as group defamation (Article 137c of the Criminal Code) and incitement to hatred or discrimination (Article 137d of the Criminal Code) – is considered a matter of liberal-democratic self-defence. More specifically, this consideration concerns the question whether statements made by politicians “incite intolerance”. By applying the Articles 137c and 137d CC in this way, these provisions have become more than just offences against public order; they have also acquired a function in a constitutional doctrine that aims to protect the liberal democratic constitution as such. This thesis explores how this doctrine has become part of the relevant European and Dutch case law and attempts to explain its origins from various theoretical perspectives. This reveals that weighing up the importance of combating hate speech against the protection of freedom of expression as a matter of liberal democratic self-defence or the paradox of tolerance is, on the one hand, a logical development but, on the other hand, also entails major risks that could undermine liberal democracy as well. To substantiate this thesis, various theories are employed, including the theory of constitutional development of Wim Couwenberg (1926-2019). This theory in particular offers a number of useful tools for explaining and assessing the turbulent development the legal framework has undergone and will undoubtedly continue to undergo.The Legitimacy and Effectiveness of Law & Governance in a World of Multilevel Jurisdiction
Exploring the synergies between transfer in reinforcement learning and procedural content generation
Full text linkIn this dissertation (titled: Exploring the Synergies between Transfer in Reinforcement Learning and Procedural Content Generation) we explore how the two research fields named in the title, namely Transfer in Reinforcement Learning (TRL) and Procedural Content Generation (PCG) can synergize together. Our journey began with a comparison of different AI algorithms on the board game Tetris Link. We compared a heuristic, Monte Carlo Tree Search (MCTS), and Reinforcement Learning (RL). For both MCTS & RL, the number of possible actions per turn indicates how difficult a game is. In Tetris Link, the number of possible actions per turn is between chess and go. This combined with difficult game mechanics where most actions result in negative points that are hard to recover from make it an interesting game for research. We expected MCTS to perform best here, especially due to our implementation of the game in Rust with a focus on speed, and the fact that our MCTS implementation was heavily parallelized and got a reasonable thinking time. Especially given the thinking time we expected different results, because in principle MCTS will always find the optimal solution, provided that it has simulated every possible outcome. Nevertheless, the heuristic outperformed both RL and MCTS, and even RL outperformed MCTS. Still, we were disappointed by the performance of RL and we were determined to find out how to make it work better. One promising avenue is using transfer in RL to increase the difficulty of tasks slowly. Instead of learning a problem from scratch, one starts with an easier task. After understanding the simple version, one continues training on a more difficult version. Every change in difficulty is then a transfer in RL. To be sure about the field and what to do we set out to create an overview of the field. To do so we scraped all available papers that contained the three keywords transfer reinforcement, and learning in the Microsoft Academic Graph. After combing out duplicates and irrelevant papers we still ended up with almost 300 contributions that we analyzed according to Taylor & Stones TRL categories. The main takeaway from our overview is that TRL mainly focuses on a static amount of tasks that only slightly differ and are all within the same domain. A true cross-domain transfer is an unsolved challenge and the few papers that actually did it had domains that were fairly similar. However, the small incremental task changes gave us the idea that this could be improved by leveraging PCG so that the learned policy generalizes better to unseen task varieties that have not been included in the static task set. This can also be extended to small difficulty increases which then is akin to automated curriculum learning. Having an idea of how to sensefully apply TRL we attempted our first experiments but quickly hit the problem of reproducibility. Training RL policies based on neural networks is hardly reproducible, a problem that plagues the whole machine-learning field. However, reproducibility is a strongly desirable trait because in TRL every task switch creates a chain of non-reproducibility. While we did not come up with a solution to the problem we did find an intermediary patch: Replay traces. Most RL environments are deterministic and hence reproducible if the same actions are played out with the same initial seed. If we save both we can re-simulate the rest of the data such as observations and rewards while saving hard-disk storage as we do not have to save these. Moreover, sharing replay traces enables reviewers of scientific publications to verify the validity of portrayed figures and tables. With a solid foundation to do verifiable experiments, we set out to actually apply TRL and PCG together. For that, we transformed the creative Linerider game from 2D into 3D space. The game itself is purely creative and does not have a set goal, which has sparked players to come up with wonderful creations such as tracks that play out synchronously to music. This is challenging as after a track is drawn one presses a play button and then a rider on a sled starts to drive down the track without any interaction from the user. Physics and gravity are the only guiding powers here. For our RL-environment, we had to of course define a goal, which in our case is to ensure that the rider arrives at a target position. This goal can be below, above, or at the same height as the rider’s starting position. The interesting thing about this environment is that the task to solve itself is an application of PCG. Initial experiments showed that the distance between the starting position and the goal has to be small for RL to learn the task. However, through small incremental increases in distance, hence applying TRL to the problem, we were able to train policies that were able to build tracks for larger distances as well. Lastly, in the vein of PCG, we also investigated Large Language Models (LLMs) to generate content for games. Related research focuses on having dynamic dialogues where the user can input anything they want to. However, user input to LLMs can easily be used to jailbreak its safety systems and derail them to spew toxic content. Our idea is Chatter. Only generate small utterances of NPCs based on pre-defined prompts of developers. We empirically show that this works quite well. Moreover, in this work, we show that the majority of consumer gaming hardware (≈ 70%) would be powerful enough to run a local LLM next to a AAA-Game such as Cyberpunk. In conclusion, we have shown that PCG and TRL synergize well. PCG can be applied to create many different tasks that help train more general RL policies. And TRL can help to achieve better results when applying RL to a PCG problem.Algorithms and the Foundations of Software technolog
Van cowboypak tot witte avondjurk: filmkostuums in een hiëroglyfisch universum
No full textModern and Contemporary Studie
