Navigating the Pre‐ and Post‐ <scp>AlphaFold</scp> Divide: <scp>CAPRI</scp> 8th Evaluation Meeting, February 12–14, Grenoble, <scp>FR</scp>

Biomolecular interactions govern a broad spectrum of cellular and intercellular processes. Characterizing these interactions at atomic resolution is fundamental for elucidating molecular mechanisms in health and disease and for guiding the development of novel therapeutics. In addition to classical structural biology methods—including X-ray crystallography, NMR, cryo-electron microscopy, and tomography—as well as a wide array of complementary experimental techniques, computational approaches for predicting the structures of protein complexes have long played an essential and synergistic role. Over time, these computational methods have advanced to model increasingly large and intricate assemblies, often integrating diverse experimental and bioinformatics data within the framework of integrative structural biology. More recently, the advent of artificial intelligence (AI) has introduced powerful new predictive strategies, further expanding the toolkit of computational structural biology. Progress in this field has been closely monitored and continues to be catalyzed by the Critical Assessment of Predicted Interactions (CAPRI) initiative (capri-docking.org; https://www.ebi.ac.uk/pdbe/complex-pred/capri/), established in 2001. CAPRI provides a community-wide platform for blind assessment of modeling algorithms on experimentally determined protein complexes (called “targets”), made available to the organizers prior to publication. Prediction Rounds are launched as new targets arise and typically last 3–6 weeks. Registered participants are invited to predict the three-dimensional structure of target assemblies, using either sequence information alone or, when available, the structures of unbound components. CAPRI Rounds also include a Scoring challenge, in which participants must discriminate near-native solutions from ensembles of models gathered from all submissions. Groups may participate as Predictors, Scorers, or both. Comprehensive details on the organization of CAPRI Rounds, evaluation procedures, and group performance rankings are available at the CAPRI websites. To date, 59 CAPRI Rounds have been completed, comprising more than 300 targets. Results from Rounds 1–45 were presented at seven assessment meetings between 2002 and 2018. Initially centered on homo- and hetero-protein complexes, CAPRI has progressively expanded its scope to encompass protein–peptide, protein–nucleic acid, and protein–carbohydrate assemblies. Several Rounds have further explored the prediction of binding affinities and the explicit modeling of interfacial water molecules. Since 2014, CAPRI has also conducted joint experiments with the Critical Assessment of Structure Prediction (CASP), fostering deeper integration across modeling methodologies and research communities. While CASP originally focused on the prediction of individual protein structures (http://predictioncenter.org/), it has since broadened its scope to include protein assemblies, protein–nucleic acid, and protein–ligand complexes. Outcomes from joint CAPRI–CASP challenges have been published in dedicated Special Issues of Proteins [1-7]. As highlighted in the overview of Lensink et al. and in the 11 reports of the invited groups contributing to this issue, this 8th CAPRI evaluation captures a watershed moment in structural biology: the dawn of AlphaFold-driven predictions of biomolecular interactions showcasing both the transformative power and persistent limits of artificial intelligence in structure prediction. The broad assessment makes clear that AlphaFold has changed the baseline of expectations. In retrospective analyses, AlphaFold performed strongly even on many of the difficult targets offered in CAPRI, validating its near-experimental accuracy for rigid or moderately flexible complexes. Yet, human predictors continue to outperform purely AI-based methods when it comes to the hardest problems—especially those involving antibodies and nucleic acids, which is in line with the reports from the latest CASP meeting in December 2024. This underscores that deep learning methods in structural biology are of unprecedented value, but do not provide yet universal solutions. Across the participating groups, hybrid strategies emerged as the dominant theme. Most teams reported substantial gains once AlphaFold predictions were integrated into their pipelines. For some, this meant adopting AlphaFold's multimer models directly; for others, it involved combining AlphaFold outputs with template-based docking or massive sampling techniques. This is reflected in the quality of the predictions, with only rare pre-AlphaFold acceptable predictions, while post-AlphaFold integration often yielded multiple medium- or high-quality interfaces. Yet the challenges of scoring and sampling remain unresolved, emphasizing the importance of robust evaluation methods, as AlphaFold confidence metrics alone cannot reliably rank complex models. MassiveFold's brute-force sampling, producing thousands of models per target, highlighted both the promise and inefficiency of current strategies—great models exist in the pool, but finding them requires better selection metrics. Meanwhile, simple, physics-based scoring stood out in its ability to identify acceptable models in particularly challenging antibody–antigen targets, suggesting that classical physics-based insight still has a role alongside AI. Another striking trend is the push for efficiency and sustainability: The MinnieFold protocol, with its reduced sampling footprint, demonstrated that “green” AlphaFold pipelines can still produce competitive results. Given the rising computational costs of deep learning approaches, this line of work may become increasingly essential for the field's future scalability. CAPRI Rounds 47–55 also reaffirm the importance of human insight. The LZerD team's use of literature-derived interface information, combined with rank aggregation and expert inspection, delivered top-quality models where no other group succeeded. Such results remind us that human-guided integrative modeling still matters, especially when dealing with ambiguous or flexible targets beyond the reach of algorithms. Taken together, these rounds have marked a transition toward an increased use of AI-based prediction tools. Still, those have not solved all challenges, and their integration with more classical approaches and human insights will remain critical to address prediction challenges such as those involving antibodies and nucleic acids, large conformational changes, and multimeric assemblies. The field now faces the task of harnessing AI not as a replacement for classical docking and scoring, but as a partner—complemented by physics, human intuition, and efficient sampling strategies. After nearly 25 years, CAPRI continues to reflect the state of the art in biomolecular modeling of interactions. The current assessment tells a story of a discipline standing on the threshold of a new era: one where AI's promise is undeniable, but where progress will hinge on integration, innovation, and collaboration. Alexandre M. J. J. Bonvin: writing – original draft, writing – review and editing. Marc F. Lensink: writing – review and editing. CAPRI thrives on the enthusiasm and dedication of its vibrant community, and on the continuous support and guidance of the CAPRI Management Committee (www.capri-docking.org). We gratefully acknowledge PDB Europe (http://pdbe.org) for hosting the CAPRI management infrastructure under the leadership of Dr. Sameer Velankar and the Lensink team for valuable assistance in conducting the CAPRI assessments. We extend our sincere thanks to all structural biologists who generously provided the targets for CAPRI Rounds 47–55 discussed in this issue: Han Remaut and Antonella Fioravanti (VIB-VUB Center for Structural Biology, Brussels, Belgium; T160); Remy Loris and Gabriela Garcia-Rodriguez (VIB-VUB, Brussels, Belgium; T161 and T162); Owen Davies (Newcastle University, UK; T163); Rouslan Efremov (VIB-VUB, Brussels, Belgium; T186–T188); Matti Pronker and Bert Janssen (Utrecht University, The Netherlands; T231); Wei Huang (Case Western Reserve University, USA; T232); and José Marquez (EMBL Grenoble, France; T233 and T234). Finally, we express our deepest gratitude to Dr. Joel Janin and Dr. Shoshana Wodak, the pioneers of CAPRI, whose leadership and vision have sustained and advanced this community effort for nearly 25 years, ensuring its continued success and impact.