Lysine-specific demethylase 1 (LSD1/KDM1A) is a key epigenetic regulator that modulates chromatin dynamics and transcriptional programs through both catalytic demethylation and scaffold-mediated interactions, rendering it an attractive therapeutic target. Although numerous reversible LSD1 inhibitors have been developed, their structural diversity and binding complexity remain insufficiently characterized due to limited structural data. Here, we combined cheminformatics with cutting-edge cofolding-based structural prediction to systematically map the chemical and conformational landscape of reversible LSD1 inhibitors. Our analysis delineated distinct chemical clusters with uneven scaffold distributions, revealing substantial, computationally unexplored regions within the current inhibitor repertoire whose druggability awaits experimental confirmation. Consistent predictions across three state-of-the-art cofolding models uncovered several novel scaffolds with nanomolar potency and previously unrecognized binding modes. In our data set, intermodel prediction consistency was positively associated with both structurally similarity to training data ligands and experimental inhibitory activity. Notably, AlphaFold3 achieved superior accuracy under sparse-data conditions. Besides, cofolding models effectively captured ligand-induced conformational changes (induced-fit effect) but failed to account for allosteric regulation. Together, these findings establish a high-resolution structural atlas of reversible LSD1 inhibitors and provide practical guidance for rational scaffold optimization. Beyond LSD1, this work offers methodological insights into leveraging cofolding-driven prediction for structure-based drug design.
Yang et al. (Thu,) studied this question.