Simulation-based Protein Engineering with Transferable Implicit Transfer Operators

Projektbeschreibung

Implizite Transferoperatoren unterstützen schnelle, allgemeine und skalierbare Proteinsimulationen

Eine einzige Aminosäure-Substitution oder eine geringfügige Konformationsänderung kann die Funktion eines Proteins erheblich verändern. Die Modellierung solch winziger Veränderungen und ihrer dynamischen funktionellen Auswirkungen würde die rationale Proteintechnik für Bereiche von der Energie bis zur Biomedizin revolutionieren. Das ERC-finanzierte SPETITO-Projekt zielt darauf ab, seine Modellierungsplattform – die Simulationen 100 000 Mal schneller als herkömmliche Molekulardynamiksimulationen durchführen kann – so zu optimieren, dass sie auf alle molekularen Systeme anwendbar und für größere Moleküle skalierbar ist. Die Arbeit umfasst die Entwicklung eines übertragbaren „impliziten Transferoperator“-Modells, das die universelle Proteinphysik im Mikrosekundenbereich erfasst, einen auf Proteinsequenzgradienten basierenden Optimierungsrahmen zur effizienten Erforschung des Sequenzraums und eine automatisierte thermodynamikintegrierte Design-Pipeline zur Entwicklung maßgeschneiderter Proteine.

Ziel

Minute changes in proteins—such as a single amino acid substitution—can profoundly impact cellular function. A computational model capable of systematically mapping the relationship between protein sequences, their dynamics, and their interactions with other biomolecules would revolutionize biology and enable rational, physics-informed bioengineering applications. SPETITO outlines the methodological foundations for such a framework.

Recent advances in deep generative AI (GenAI) have enabled surrogate methods for efficient molecular dynamics (MD) simulations, tackling the challenge of sampling rare events and conformational transitions. The PI co-developed groundbreaking methods—Boltzmann Generators (BG) and Implicit Transfer Operators (ITO)— that enable the study of conformational changes, protein folding, and prediction of experiments five orders of magnitude faster than MD.
However, these methods cannot generalize across molecular systems nor scale to larger molecules, severely
limiting their utility. To address this, we propose developing and disseminating transformative technologies
through widely accessible software:

• A transferable ITO (TransITO) model that captures universal protein physics, enabling rapid sampling
with microsecond-time-steps for any protein system without requiring system-specific training data.
• A ‘protein sequence gradient’-based optimization framework, enabling efficient exploration and optimization of protein sequence space for targeted functional outcomes.
• An automated protocol for protein design that integrates multiple thermodynamic constraints, facilitating the rational engineering of proteins with desired properties.

Beyond turbo-charged atomic-resolution insights into biophysics, TransITO will enable rational, physics-informed protein engineering. With experimental collaborators, we will apply TransITO to vaccine development, to accelerate therapeutic design for infectious diseases.