Dresden 2026 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

BP: Fachverband Biologische Physik

BP 18: Focus session: Integrative Structural Modeling

BP 18.1: Invited Talk

Wednesday, March 11, 2026, 10:30–11:00, BAR/0106

Protein complex structure prediction, state-of-the-art and challenges — •Ezgi Karaca — Izmir Biomedicine and Genome Center — Dokuz Eylul University, Izmir, TR

The advent of artificial intelligence has reshaped structural biology, enabling unprecedented accuracy in protein modeling. As the assessor for CASP14 and CASP15, I witnessed remarkable progress facilitated by AlphaFold. While CASP14 participants struggled with intersubunit contacts, CASP15 saw prediction success rates rise from 31-percent to 90-percent by incorporating AlphaFold2 (AF2) into customized pipelines. This improvement was driven by curated multiple sequence alignments (MSAs), though complexes with shallow MSAs remained challenging. Addressing this, we developed MinnieFold, an efficient protocol achieving a 50-fold reduction in GPU usage while maintaining accuracy in antibody-antigen complexes. Further investigating AF’s performance, we analyzed 276 human-parasite interactions across 15 species. Comparing AF2 and AlphaFold3 (AF3), we observed striking differences in predicted structures. This presentation highlights these advances, challenges, and our contributions to AI-driven structural biology.