🤖 AI Summary
In molecular generative models, the SELFIES string representation often introduces artifacts—such as those related to sequence length, branching, and ring structures—that obscure genuine chemical signals in the latent space. This work addresses this challenge by employing linear probes within a frozen Transformer-VAE latent space to identify global directions governing key physicochemical properties. The authors propose a confounding-aware evaluation framework that combines residualization analysis with decoded molecular traversal to systematically disentangle representation artifacts from true chemical signals. Using this approach, they demonstrate for the first time that six properties—including cLogP and FractionCSP3—exhibit robust and monotonically controllable directions even within an entangled latent space, thereby validating the feasibility of chemically meaningful latent space editing.
📝 Abstract
Molecular generative models often assume meaningful latent geometry, but apparent property predictability can reflect sequence-level shortcuts rather than chemical organization. We study this issue in an unsupervised autoregressive Transformer-VAE trained on SELFIES. After training, we freeze the model, fit linear probes to RDKit descriptors, and use the probe weights as candidate global steering directions. To separate chemical signal from SELFIES artifacts, we introduce a confound-aware evaluation based on residualization, confound-direction alignment analysis, and decoded-molecule traversal. This is necessary because SELFIES length, branch tokens, ring tokens, and token entropy are strongly encoded in the latent space. Under this confound-aware evaluation, we find robust monotonic steering for cLogP, FractionCSP3, HeavyAtomCount, TPSA, BertzCT, and HBA. Nonlinear probes further show that some properties admit stable global directions, while others are better described by local latent gradients. Overall, our results show that chemically meaningful steering can emerge in entangled molecular latent spaces, but only when validated through decoded molecules and controlled for representation-level confounds.