Shetti-AI: Precise prediction of protein sequence mutational landscapes to accelerate functional assays

Functional protein motifs dictate molecular interactions, yet experimentally screening large sequence-variant spaces is slow and expensive. Here, we describe Shetti-AI, a computational framework that starts from curated motifs to generate motif-like variants, aiming to increase the success rate of downstream functional and structural assays. The framework integrates physicochemical property and genetic distance matrices to constrain and guide candidate generation. While Lambda and Bayesian models evaluate closely related mutations, GAN and WGAN-GP models generate candidate variants, which are subsequently evaluated and ranked using Bayesian scores. By predicting prospective protein sequence mutational landscapes rather than relying solely on historical evolutionary conservation, Shetti-AI provides a high-throughput, cost-effective pipeline to reduce experimental search spaces for synthetic biology, emerging viral proteins, and functional assays.