Generative machine learning models are increasingly being used to design novel proteins. However, their major limitation is the inability to account for protein flexibility, a property crucial for protein function. Learning to engineer flexibility is difficult because the relevant data is scarce, heterogeneous, and costly to obtain using computational and experimental methods. Our contributions are three-fold. First, we perform a comprehensive comparison of methods for evaluating protein flexibility and identify relevant data for learning. Second, we overcome the data scarcity issue by leveraging a pre-trained protein language model. We design and train flexibility predictors utilizing either only sequential or both sequential and structural information on the input. Third, we introduce a method for fine-tuning a protein inverse folding model to make it steerable toward desired flexibility at specified regions. We demonstrate that our method Flexpert enables guidance of inverse folding models toward increased flexibility. This opens up a transformative possibility of engineering protein flexibility.

Dear colleagues, We thank you for your comments on our manuscript [1] and for raising the important question of automation [2]. The publications by Sepp et al. (2019) [3] and Liu et al. (2024) [4] focus on one type of PBPK models, denominated ‘for biologics’. In these models, each compartment (tissue/organ) is divided into vascular, endothelial endosomal and interstitial (sub-)compartments, and lymphatic flows are also represented [3, 4]. In these examples, the pharmacokinetics of proteins are represented. Notably, the protein distribution within tissues includes processes of passive transport across two types of pores via diffusion or fluid convection, processes of pinocytosis, binding, recycling and degradation [4]. This type of PBPK models is complex and their authors successfully used specific tools for automated code generation. In particular, Liu et al. (2024) used ‘mathematical sets’ available in the Ubiquity package, which facilitates assembling model components in R-language [4]. Sepp et al. (2019) used the MATLAB script PBPKassembler.m for automated code generation, where files in Simbiology (MATLAB) and Excel formats were combined [3]. Briefly, automation can help build models, which is highly valuable notably for complex models. Having a robust, thoroughly validated platform for automated code generation will indeed contribute to accessibility and reproducibility. Also, automation of model building seems to be a natural way to address the case of large models, in which manual scripting will be prone to introducing errors. Smaller models or those with fewer equations do not necessarily require automation. From our experience, writing the script manually provides a deep understanding of the underlying model mechanics. Manually written scripts also allow to give a clear view of how equations and parameters are applied, which we consider especially valuable for teaching and knowledge transfer. To conclude, as observed in other niches of systems biology, automation brings significant benefits to the modelling field, e.g. there is a plethora of tools that reconstruct and benchmark genome-scale metabolic models. As the correspondence authors argue, there is more scope for automation in the current practices of PBPK modelling. To facilitate this, we have opted to begin the journey of standardization via open collaboration through ELIXIR. We are looking forward to continuing to engage the research community towards automatic construction, validation and deposition of PBPK models. Thank you. Best regards, Elena Domínguez-Romero, Stanislav Mazurenko, Martin Scheringer, Vítor Martins dos Santos, Chris Evelo, Mihail Anton, John M. Hancock, Anže Županič, and Maria Suarez-Diez. None declared.

Discovering mutations enhancing protein-protein interactions (PPIs) is critical for advancing biomedical research and developing improved therapeutics. While machine learning approaches have substantially advanced the field, they often struggle to generalize beyond training data in practical scenarios. The contributions of this work are three-fold. First, we construct PPIRef, the largest and non-redundant dataset of 3D protein-protein interactions, enabling effective large-scale learning. Second, we leverage the PPIRef dataset to pre-train PPIformer, a new SE(3)-equivariant model generalizing across diverse protein-binder variants. We fine-tune PPIformer to predict effects of mutations on protein-protein interactions via a thermodynamically motivated adjustment of the pre-training loss function. Finally, we demonstrate the enhanced generalization of our new PPIformer approach by outperforming other state-of-the-art methods on new, non-leaking splits of standard labeled PPI mutational data and independent case studies optimizing a human antibody against SARS-CoV-2 and increasing the thrombolytic activity of staphylokinase.

Systems biology aims to understand living organisms through mathematically modeling their behaviors at different organizational levels, ranging from molecules to populations. Modeling involves several steps, from determining the model purpose to developing the mathematical model, implementing it computationally, simulating the model's behavior, evaluating, and refining the model. Importantly, model simulation results must be reproducible, ensuring that other researchers can obtain the same results after writing the code de novo and/or using different software tools. Guidelines to increase model reproducibility have been published. However, reproducibility remains a major challenge in this field. In this paper, we tackle this challenge for physiologically-based pharmacokinetic (PBPK) models, which represent the pharmacokinetics of chemicals following exposure in humans or animals. We summarize recommendations for PBPK model reporting that should apply during model development and implementation, in order to ensure model reproducibility and comprehensibility. We make a proposal aiming to harmonize abbreviations used in PBPK models. To illustrate these recommendations, we present an original and reproducible PBPK model code in MATLAB, alongside an example of MATLAB code converted to Systems Biology Markup Language format using MOCCASIN. As directions for future improvement, more tools to convert computational PBPK models from different software platforms into standard formats would increase the interoperability of these models. The application of other systems biology standards to PBPK models is encouraged. This work is the result of an interdisciplinary collaboration involving the ELIXIR systems biology community. More interdisciplinary collaborations like this would facilitate further harmonization and application of good modeling practices in different systems biology fields.

PredictONCO 1.0 is a unique web server that analyzes effects of mutations on proteins frequently altered in various cancer types. The server can assess the impact of mutations on the protein sequential and structural properties and apply a virtual screening to identify potential inhibitors that could be used as a highly individualized therapeutic approach, possibly based on the drug repurposing. PredictONCO integrates predictive algorithms and state-of-the-art computational tools combined with information from established databases. The user interface was carefully designed for the target specialists in precision oncology, molecular pathology, clinical genetics and clinical sciences. The tool summarizes the effect of the mutation on protein stability and function and currently covers 44 common oncological targets. The binding affinities of Food and Drug Administration/ European Medicines Agency -approved drugs with the wild-type and mutant proteins are calculated to facilitate treatment decisions. The reliability of predictions was confirmed against 108 clinically validated mutations. The server provides a fast and compact output, ideal for the often time-sensitive decision-making process in oncology. Three use cases of missense mutations, (i) K22A in cyclin-dependent kinase 4 identified in melanoma, (ii) E1197K mutation in anaplastic lymphoma kinase 4 identified in lung carcinoma and (iii) V765A mutation in epidermal growth factor receptor in a patient with congenital mismatch repair deficiency highlight how the tool can increase levels of confidence regarding the pathogenicity of the variants and identify the most effective inhibitors. The server is available at https://loschmidt.chemi.muni.cz/predictonco.

SUMMARY: Protein design requires information about how mutations affect protein stability. Many web-based predictors are available for this purpose, yet comparing them or using them en masse is difficult. Here, we present BenchStab, a console tool/Python package for easy and quick execution of 19 predictors and result collection on a list of mutants. Moreover, the tool is easily extensible with additional predictors. We created an independent dataset derived from the FireProtDB and evaluated 24 different prediction methods. AVAILABILITY AND IMPLEMENTATION: BenchStab is an open-source Python package available at https://github.com/loschmidt/BenchStab with a detailed README and example usage at https://loschmidt.chemi.muni.cz/benchstab. The BenchStab dataset is available on Zenodo: https://zenodo.org/records/10637728.