Comprehensive in silico analyses of keratin heterodimerisation

Keratins are the largest and most diverse group of intermediate filament proteins, providing structural integrity and mechanical strength to epithelial cells. Although their assembly as heterodimers is well established, the specific pairing preferences and molecular basis of keratin dimerisation remain largely unknown. Here, we employ a high-throughput computational pipeline that integrates AlphaFold Multimer (AFM) modelling, VoroIF-GNN interaction interface quality assessment, interaction energy calculations and structural comparisons with experimentally solved structures to systematically investigate keratin heterodimerisation and to provide a guideline for further analysis of intermediate filament assembly. To validate our in silico approach, we include the well-studied vimentin homodimer as a reference. The predicted vimentin homodimer shows strong agreement with available experimental data, supporting the accuracy of our modelling pipeline. Our results show that keratin heterodimers generally have lower interaction energies, indicating more favourable interactions, than their homodimer counterparts, and exhibit structural configurations that closely resemble known intermediate filament structures. Comparative analyses of different keratin pairs also reveal the importance of the coil 1 region for dimer stability. Furthermore, co-expression of keratin pairings is demonstrated by analysis of spatial transcriptomics data in skin under physiological and pathological conditions. Collectively, these findings highlight structural principles underlying canonical keratin heterodimerisation and establish a robust computational workflow for elucidating alternative keratin dimerisations.