Joost Schymkowitz

Protein structure predictors can recover protein folds with remarkable accuracy, and their models are increasingly used as if they were experimental structures. Yet even for a confidently predicted fold, it remains difficult to judge how faithfully the model reproduces protein geometry. We introduce WhatRes, which assesses a structure from its coordinates on two complementary axes. Apparent resolution estimates the experimental resolution its local geometry most closely resembles. Realism measures how closely that geometry matches experimental structures at the same apparent resolution. Predicted models can score well on apparent resolution while showing systematically less variation in bond lengths and angles than is observed in high-resolution crystal structures. WhatRes reveals this over-idealisation even in high-confidence models.

Medin, a 50-residue fragment of lactadherin (MFG-E8), forms aortic medial amyloid, the most common localized amyloid in humans, and co-deposits with vascular amyloid-β (Aβ) in Alzheimer’s disease. In the full-length medin fibril the N-terminal half forms the entire protofilament interface, yet the strongest predicted amyloidogenic determinants lie in the C-terminal region. Here we show that this region aggregates on its own. The isolated C-terminal aggregation-prone region (medin C-APR, medin 31–53) begins to aggregate with the same lag time as full-length medin and assembles into two coexisting polymorphs, determined here by cryo-EM at 2.71 and 2.91 Å. Both are built from the same two-stranded β-arch and a pseudo-2₁ protofilament pair closed by an aromatic–aliphatic zipper, and differ only in how many accessory chains the exposed hydrophobic face recruits. The medin C-terminus therefore carries an amyloid core of its own, unrelated to that of the full-length fibril: medin has more than one aggregation-prone region, and the structural contribution of each region depends on sequence and assembly context.

This Perspective explores recent methods and prospective ideas for developing hybrid AI‐physics‐based pipelines for protein and antibody de novo design. We argue that the highest‐confidence candidates emerge where deep learning and first‐principles models agree, a “sweet spot” that balances generative flexibility with thermodynamic realism. For example, although interface confidence scores such as ipTM, pDockQ2, or ipSAE are widely used to rank generated designs, we show that they are not well suited to rank similar sequences, which suggests the need to combine them with physics‐based methods to improve design filtering and ranking. Furthermore, we describe a generalizable framework for implementing antibody design pipelines that combine AI with physics‐based modeling and scoring methods and also showcase MadraX, a differentiable and AI‐compatible implementation of the FoldX force field. In addition, we classify three tiers of AI‐physics integration, from post hoc filtering to full embedding of differentiable physics inside deep learning models. Finally, we discuss the future of the protein design community and underline the need to support current initiatives for community wide blind assessments of the growing number of de novo design pipelines.

Co-translational protein folding is shaped by the vectorial nature of translation, which causes residues to emerge sequentially from the ribosome. As a result, residues whose native interaction partners lie downstream in sequence cannot immediately form their native contacts and remain transiently unsatisfied until those partners are synthesized. These unsatisfied residues are vulnerable to non-native interactions and often require the engagement of co-translational chaperones. We previously developed the Native Fold Delay (NFD) metric to quantify the time lag between the synthesis of a residue and the point at which it can form all its native contacts. Here, we present the FoldDelay web server, a freely accessible platform that extends the NFD concept into a more comprehensive framework for analyzing native residue–residue contact formation during translation. Starting from user-submitted AlphaFold or PDB structures, the site identifies all N- to C-terminal residue–residue contacts, estimates their earliest possible formation times, and integrates domain annotations to distinguish between intra- and inter-domain contacts. The server provides a suite of linked interactive visualizations that allows users to explore native contact formation dynamics and detect transiently unsatisfied regions. The FoldDelay web server is freely accessible at https://folddelay.switchlab.org.

Amyloid assembly in vivo occurs in complex environments where multiple aggregation-prone species coexist. Aβ and medin are prevalent amyloids in ageing humans that co-localize in cerebral amyloid angiopathy (CAA), yet their structural interactions remain poorly understood. Here, using cryo-electron microscopy, we determine high-resolution fibril structures from in vitro mixtures of Aβ40 and medin. From the same reaction, we resolve three distinct fibril populations: (i) a previously characterized Aβ40 polymorph that also forms in isolation, (ii) a Aβ40 polymorph with Aβ42-like features, including ordered N- and C-terminal regions, and (iii) the atomic structure of full-length medin fibrils. Biochemical and immunogold analyses demonstrate Aβ–medin association within mixed assemblies, though medin is not resolved within the ordered Aβ core. These findings support two non-exclusive mechanisms: transient heterotypic interactions redirecting Aβ folding, or partial medin incorporation into fibril architecture. Our data reveal how coexisting amyloids remodel each other’s polymorphic landscapes.

Amyloidosis is a disorder caused by the extracellular deposition of misfolded protein fibrils, leading to organ dysfunction. Diagnosis remains challenging due to non-specific clinical presentations and the diversity of amyloid subtypes. Accurate identification of the amyloid precursor protein is key for prognostication and treatment strategy. This document aims to provide practical recommendations for the tissue diagnosis of amyloidosis within the Belgian healthcare context. It targets clinicians managing amyloidosis patients and pathologists evaluating biopsies with suspected amyloid deposits. A structured PubMed search ('amyloidosis AND biopsy AND stain*'; 'amyloidosis AND mass spectro*') was conducted in August 2025. After exclusion of case reports, preclinical studies, and Alzheimer-related articles, 298 publications were reviewed. Recommendations were formulated based on available evidence and discussed among Belgian clinical and pathology experts. Key recommendations emphasize that tissue biopsies remain essential for amyloidosis diagnosis and typing. Congo red staining with birefringence and fluorescence confirmation is required. Immunohistochemistry and immunofluorescence are first-line subtyping tools, while mass spectrometry serves as a reference method when results remain inconclusive. Centralization of complex analyses in experienced centres is encouraged. These recommendations promote standardized, early and accurate tissue diagnosis of amyloidosis in Belgium, supporting optimal patient management and harmonization of diagnostic practices across institutions.

The prevailing view frames microglia and macrophages as guardians against amyloid beta (Aβ) accumulation in Alzheimer’s disease (AD). Here, we overturn this paradigm by demonstrating that human phagocytic cells, including differentiated THP-1 macrophages and hESC-derived microglia, are not merely passive responders but active producers of extracellular, seeding-competent Aβ42 fibrils, the amyloid species most strongly linked to parenchymal plaque formation and neurodegeneration. These cell-generated aggregates differ structurally and functionally from synthetic fibrils, displaying enhanced seeding and tau cross-seeding activity in biosensor models. Notably, Aβ42 fibril formation in this system requires active cellular processes and is exacerbated by loss of Triggering Receptor Expressed on Myeloid Cells 2 (TREM2), a major AD risk gene. Transcriptomic profiling reveals an early inflammatory response resembling microglial states observed in human AD models. Together, these findings support emerging evidence from in vivo studies that macrophages and microglia can influence amyloid seeding and introduce a human-relevant in vitro platform to explore how Aβ aggregation intersects with innate immune function and genetic risk. Our results reinforce the concept that microglia may play a dual role in AD, acting both as responders and inadvertent facilitators of amyloid assembly, with implications for early therapeutic intervention.