Project for MASH Pipeline targeting THRB

This project leverages InVirtuoLabs' integrated computational drug discovery platform to identify novel small molecule modulators of the Thyroid Hormone Receptor Beta (THRB) for the treatment of Metabolic dysfunction-Associated Steatohepatitis (MASH). The core of drug discovery lies in identifying molecular entities capable of selectively interacting with crucial target proteins implicated in disease pathogenesis. In recent years, computational methods have revolutionized this field, offering unprecedented speed, accuracy, and cost-effectiveness. Our approach synergizes three powerful modules: InVirtuoMOL (machine learning for molecular property and activity prediction), InVirtuoGEN (generative chemistry based on a chemical language model for exploring the deep chemical universe), and InVirtuoSIM (physics-based biomolecular simulations generating synthetic data). InVirtuoMOL will screen ultra-large commercially available chemical libraries (>100 billion molecules, e.g., Enamine REAL Space) to rapidly identify candidate THRB binders. InVirtuoGEN, powered by a 200-million-parameter chemical language model (scaling to 1 billion) and fragment-based generation, will design novel THRB-selective chemotypes with optimized ADMET properties. InVirtuoSIM will generate atomic-level synthetic data from molecular dynamics simulations of THRB-ligand complexes, including binding poses, interaction energies, conformational changes and residence times, to refine ML predictions through active learning closed loops. By combining the speed and pattern recognition of AI with the mechanistic accuracy of physics-based simulations, this pipeline aims to accelerate hit identification and lead optimization for MASH, increase selectivity over THRA to reduce cardiac side effects, and deliver a validated set of lead candidates within 12 months. The project will exploit EuroHPC AI Factory resources to train large chemical foundation models, perform ultra-large virtual screening campaigns, and run enhanced-sampling MD simulations at unprecedented scale.