Controllable protein design via autoregressive direct coupling analysis conditioned on principal components

We present FeatureDCA, a statistical framework for protein sequence modeling and generation that extends Direct Coupling Analysis (DCA) with biologically meaningful conditioning. The method can leverage different kinds of information, such as phylogeny, optimal growth temperature, enzymatic activity or, as in the case presented here, principal components derived from multiple sequence alignments, and use it to improve the learning process and consequently efficiently condition the generative process. FeatureDCA allows sampling to be guided toward specific regions of sequence space while maintaining the efficiency and interpretability of Potts-based inference. Across multiple protein families, our autoregressive implementation of FeatureDCA matches or surpasses the generative accuracy of established models in reproducing higher-order sequence statistics while preserving substantial sequence diversity. Structural validation with AlphaFold and ESMFold confirms that generated sequences adopt folds consistent with their intended wild-type targets. In a detailed case study of the Response Regulator family (PF00072), which comprises distinct structural subclasses linked to different DNA-binding domains, FeatureDCA accurately reproduces class-specific architectures when conditioned on subtype-specific principal components, highlighting its potential for fine-grained structural control. Predictions of experimental deep mutational scanning data show accuracy comparable to that of unconditioned autoregressive Potts models, indicating that FeatureDCA also captures local functional constraints. These results position FeatureDCA as a flexible and transparent approach for targeted sequence generation, bridging statistical fidelity, structural realism, and interpretability in protein design.Author summary: Designing new proteins with desired functions is a major challenge in biology and biotechnology. Current statistical approaches can generate protein-like sequences, but they rarely allow users to guide designs toward specific structures or functions in a clear, interpretable way. We introduce FeatureDCA, a statistical model that learns from sets of evolutionarily related protein sequences how structural and functional properties encoded in the sequences relate to their positions in a low-dimensional projection of protein space. This makes it possible to guide the generation of new protein sequences towards biologically meaningful features. We show that FeatureDCA can generate novel sequences that resemble natural proteins in specific user-defined positions in protein space. In a case study, the method generated bacterial signaling proteins that, when folded with AlphaFold, adopted different dimerization modes highly consistent with experimental structures. FeatureDCA thus provides a transparent and efficient framework for guiding protein design, with potential applications in biotechnology, synthetic biology, and evolutionary research.