I like works that explore different protein/ligand representations. I think existing representations, e.g. embeddings per-protein-residue, are unintuitive; a longer protein has a longer protein though it may be far less complex than a shorter one. The density map representation for the ligand and the protein pocket is nice in this sense!
First, encoding the local structure of a protein by estimating various forces on a dense grid is a common technique, and largely used in various docking algorithms. These so-called density maps provide a natural representation for 3D structural information, encoding multiple physical properties (charge, hydrophobicity, hydrogen bonding, van der Waals interactions) in a unified grid-based format. This allows for modeling diverse sets of protein pockets with identically-sized inputs, which confers a number of computational efficiencies
This allows for essentially equivalent representations of pocket-conditioned ligand generation and generation seeded with a scaffold/fragment!