Proteins are complex molecules responsible for different functions in nature. Enhancing the functionality of proteins and cellular fitness can significantly impact various industries. However, protein optimization using computational methods remains challenging, especially when starting from low-fitness sequences. We propose LatProtRL, an optimization method to efficiently traverse a latent space learned by an encoder-decoder leveraging a large protein language model. To escape local optima, our optimization is modeled as a Markov decision process using reinforcement learning acting directly in latent space. We evaluate our approach on two important fitness optimization tasks, demonstrating its ability to achieve comparable or superior fitness over baseline methods. Our findings and in vitro evaluation show that the generated sequences can reach high-fitness regions, suggesting a substantial potential of LatProtRL in lab-in-the-loop scenarios.

MOTIVATION: Structural characterization of protein interactions is necessary for understanding and modulating biological processes. On one hand, X-ray crystallography or NMR spectroscopy provide atomic resolution structures but the data collection process is typically long and the success rate is low. On the other hand, computational methods for modeling assembly structures from individual components frequently suffer from high false-positive rate, rarely resulting in a unique solution. RESULTS: Here, we present a combined approach that computationally integrates data from a variety of fast and accessible experimental techniques for rapid and accurate structure determination of protein-protein complexes. The integrative method uses atomistic models of two interacting proteins and one or more datasets from five accessible experimental techniques: a small-angle X-ray scattering (SAXS) profile, 2D class average images from negative-stain electron microscopy micrographs (EM), a 3D density map from single-particle negative-stain EM, residue type content of the protein-protein interface from NMR spectroscopy and chemical cross-linking detected by mass spectrometry. The method is tested on a docking benchmark consisting of 176 known complex structures and simulated experimental data. The near-native model is the top scoring one for up to 61% of benchmark cases depending on the included experimental datasets; in comparison to 10% for standard computational docking. We also collected SAXS, 2D class average images and 3D density map from negative-stain EM to model the PCSK9 antigen-J16 Fab antibody complex, followed by validation of the model by a subsequently available X-ray crystallographic structure.