Abstract 1018 Translation efficiency covariation across cell types is a conserved organizing principle of mammalian transcriptomes

Understanding how mutations affect the stability of folded proteins is crucial.In medicine, it helps reveal mechanisms of diseases linked to protein misfolding, while in protein engineering, it aids in identifying stable catalytically active sequences.Additionally, it can help evolutionary studies shed light on the effects of specific mutation pathways on fitness.Predicting free energy and its changes remains challenging with AI due to the elusive nature of entropy, which makes pattern recognition hard and balanced databases difficult to build.Physics-based methods like Molecular Dynamics (MD) offer an alternative principled approach to calculating these changes.Most MD-based approaches rely on defining -a priori or on the flya physical or unphysical reaction coordinate (RC) that connects the states we are interested in comparing.But, defining the unfolded state in terms of RC is challenging, thus limiting the applicability of most MD-based free energy calculation approaches to the problem of protein stability.MELDxMD (Modelling Employing Limited Data accelerated MD) does not rely on the definition of a reaction coordinate.Instead, it uses external information to limit the system's conformational space.At the end of the simulation, the states' relative free energy is computed from their relative populations.In this work, we use MELDxMD to compute the relative stability of proteins whose sequences differ only by one amino acid mutation.To do so, we set up a horse race between two proteins.Using MELD, we force one and only one of the two proteins to be folded at any given time, allowing the other protein to explore its unfolded state.This horse race approach gets around the need to define the unfolded state.Our predictions agree with experimentally determined folding stabilities values for a wide set of small proteins and mutations from the "Mega-scale Rocklin's database.