勃林格殷格翰制药

Human epidermal growth factor receptor 2 (HER2) mutations in non-small cell lung cancer (NSCLC) are associated with aggressive disease and poor prognosis. Improved understanding of patient characteristics is vital to advance personalized treatment and improve outcomes. In this structured protocol-based targeted literature review, we assessed the epidemiology and ‘real-world’ outcomes in patients with HER2-mutant NSCLC by region, and by mutation category (tyrosine kinase domain [TKD] and non-TKD). Across 64 studies, the frequency of HER2 mutations ranged from 1

Automated methods for discovering mechanistic simulator models from observational data offer a promising path toward accelerating scientific progress. Such methods often take the form of agentic-style iterative workflows that repeatedly propose and revise candidate models by imitating human discovery processes. However, existing LLM-based approaches typically implement such workflows via hand-crafted heuristic procedures, without an explicit probabilistic formulation. We recast model discovery as probabilistic inference, i.e., as sampling from an unknown distribution over mechanistic models capable of explaining the data. This perspective provides a unified way to reason about model proposal, refinement, and selection within a single inference framework. As a concrete instantiation of this view, we introduce ModelSMC, an algorithm based on Sequential Monte Carlo sampling that represents candidate models as particles which are iteratively proposed and refined by an LLM, and weighted using likelihood-based criteria. Experiments on real-world scientific systems illustrate that this formulation discovers models with interpretable mechanisms and improves posterior predictive checks. More broadly, this perspective provides a probabilistic lens for understanding and developing LLM-based approaches to model discovery.

This study examines the application of Bayesian approach in the context of clinical trials, emphasizing their increasing importance in contemporary biomedical research. While conventional frequentist approach provides a foundational basis for analysis, it often lacks the flexibility to integrate prior knowledge, which can constrain its effectiveness in adaptive settings. In contrast, Bayesian methods enable continual refinement of statistical inferences through the assimilation of accumulating evidence, thereby supporting more informed decision-making and improving the reliability of trial findings. This paper also considers persistent challenges in clinical investigations, including replication difficulties and the misinterpretation of statistical results, suggesting that Bayesian strategies may offer a path toward enhanced analytical robustness. Moreover, discrete probability models, specifically the Binomial, Poisson, and Negative Binomial distributions are explored for their suitability in modeling clinical endpoints, particularly in trials involving binary responses or data with overdispersion. The discussion further incorporates Bayesian networks and Bayesian estimation techniques, with a comparative evaluation against maximum likelihood estimation to elucidate differences in inferential behavior and practical implementation.

Accurately computing the free energies of biological processes is a cornerstone of computer-aided drug design, but it is a daunting task. The need to sample vast conformational spaces and account for entropic contributions makes the estimation of binding free energies very expensive. While classical methods, such as thermodynamic integration and alchemical free energy calculations, have significantly contributed to reducing computational costs, they still face limitations in terms of efficiency and scalability. We tackle this through a quantum algorithm for the estimation of free energy differences by adapting the existing Liouvillian approach and introducing several key algorithmic improvements. We directly implement the Liouvillian operator and provide an efficient description of electronic forces acting on both nuclear and electronic particles on the quantum ground state potential energy surface. This leads to super-polynomial runtime scaling improvements in the precision of our Liouvillian simulation approach and quadratic improvements in the scaling with the number of particles relative to prior quantum algorithms. Second, our algorithm calculates free energy differences via a fully quantum implementation of thermodynamic integration and alchemy, thereby foregoing expensive entropy estimation subroutines used in prior works. Our results open new avenues towards the application of quantum computers in drug discovery.

Despite recent advances in molecular foundation models, several limitations remain, such as chemically invalid augmentations, modality collapse, and incomplete representation of biochemical environments. To address these challenges, we present Mol-JEPA, a scalable framework for learning molecular world models. Rather than relying on suboptimal molecular perturbations, our model uses modality masking to exploit information from molecular structures, cellular phenotypes, binding affinities, ADMET profiles, quantum chemistry simulations and other drug discovery data. Across various benchmarks, we show that the representations learned by Mol-JEPA deliver strong performance, demonstrating the value of incorporating biochemical context through latent space prediction.