De Novo Generation of Odorant Molecules with Targeted Olfactory Receptor Activation Patterns

Kexin Zhang, Manhao Guan

Keywords: molecular generation, reinforcement learning, olfactory receptors, combinatorial code

TL;DR: We present a profile-guided de novo odorant generation framework that designs molecules with targeted olfactory receptor activation patterns.

Abstract: Odor perception arises from a combinatorial code in which each molecule activates a specific subset of olfactory receptors (ORs), and the resulting activation pattern determines the perceived smell. We present a framework for de novo generation of molecules predicted to match target multi-receptor activation profiles. We first construct a panoramic binding matrix over 20 ORs, then cluster it to discover representative activation patterns, and finally formulate a structured selectivity loss, combining pairwise ranking, distribution matching, and complexity constraints, to steer a molecular generator via reinforcement learning. Across nine target profiles of varying complexity, our method achieves high predicted selectivity within 18000 oracle calls and provides the largest gains on profiles that are rare under the prior, especially broad-spectrum and sparse-activator targets.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 24