Invariant Point Attention (IPA) is a key algorithm for geometry-aware modeling in structural biology, central to many protein and RNA models. However, its quadratic complexity limits the input sequence length. We introduce FlashIPA, a factorized reformulation of IPA that leverages hardware-efficient FlashAttention to achieve linear scaling in GPU memory and wall-clock time with sequence length. FlashIPA matches or exceeds standard IPA performance while substantially reducing computational costs. FlashIPA extends training to previously unattainable lengths, and we demonstrate this by re-training generative models without length restrictions and generating structures of thousands of residues. FlashIPA is available at https://github.com/flagshippioneering/flash_ipa.

Protein-protein interactions, which form the basis for most cellular processes, result in the formation of protein interfaces. Believing that the local shape of proteins is crucial, we take a geometric approach and present a definition of an interface surface formed by two or more proteins as a subset of their Voronoi diagram. The definition deals with the difficult and important problem of specifying interface boundaries by invoking methods used in the alpha shape representation of molecules, the discrete flow on Delaunay simplices to define pockets and reconstruct surfaces, and the assessment of the importance of topological features. We present an algorithm to construct the surface and define a hierarchy that distinguishes core and peripheral regions. This hierarchy is shown to have correlation with hot-spots in protein-protein interactions. Finally, we study the geometric and topological properties of interface surfaces and show their high degree of contortion.

Protein-protein interactions, which form the basis for most cellular processes, result in the formation of protein interfaces. Believing that the local shape of proteins is crucial, we take a geometric approach and present a definition of an interface surface formed by two or more proteins. We also present an algorithm and study the geometric and topological properties of these surfaces, thus paving the way for future biochemical studies of protein-protein interactions.