https://3demmethods.i2pc.es/index.php?action=history&feed=atom&title=2024Klindt_Disentanglement 2026-05-24T19:33:19Z Revision history for this page on the wiki MediaWiki 1.44.2 https://3demmethods.i2pc.es/index.php?title=2024Klindt_Disentanglement&diff=4713&oldid=prev 2024-08-21T05:54:31Z

<p>Created page with &quot;== Citation == Klindt, David A. / Hyvärinen, Aapo / Levy, Axel / Miolane, Nina / Poitevin, Frédéric. Towards interpretable Cryo-EM: disentangling latent spaces of molecular conformations. 2024. Frontiers in Molecular Biosciences, Vol. 11, p. 1393564 == Abstract == Molecules are essential building blocks of life and their different conformations (i.e., shapes) crucially determine the functional role that they play in living organisms. Cryogenic Electron Microscopy...&quot;</p> <p><b>New page</b></p><div>== Citation ==<br /> <br /> Klindt, David A. / Hyvärinen, Aapo / Levy, Axel / Miolane, Nina / Poitevin, Frédéric. Towards interpretable Cryo-EM: disentangling latent spaces of molecular conformations. 2024. <br /> Frontiers in Molecular Biosciences, Vol. 11, p. 1393564<br /> <br /> == Abstract ==<br /> <br /> Molecules are essential building blocks of life and their different conformations<br /> (i.e., shapes) crucially determine the functional role that they play in living<br /> organisms. Cryogenic Electron Microscopy (cryo-EM) allows for acquisition of<br /> large image datasets of individual molecules. Recent advances in computational<br /> cryo-EM have made it possible to learn latent variable models of conformation<br /> landscapes. However, interpreting these latent spaces remains a challenge as<br /> their individual dimensions are often arbitrary. The key message of our work is<br /> that this interpretation challenge can be viewed as an Independent Component<br /> Analysis (ICA) problem where we seek models that have the property of<br /> identifiability. That means, they have an essentially unique solution,<br /> representing a conformational latent space that separates the different<br /> degrees of freedom a molecule is equipped with in nature. Thus, we aim to<br /> advance the computational field of cryo-EM beyond visualizations as we connect<br /> it with the theoretical framework of (nonlinear) ICA and discuss the need for<br /> identifiable models, improved metrics, and benchmarks. Moving forward, we<br /> propose future directions for enhancing the disentanglement of latent spaces in<br /> cryo-EM, refining evaluation metrics and exploring techniques that leverage<br /> physics-based decoders of biomolecular systems. Moreover, we discuss how<br /> future technological developments in time-resolved single particle imaging may<br /> enable the application of nonlinear ICA models that can discover the true<br /> conformation changes of molecules in nature. The pursuit of interpretable<br /> conformational latent spaces will empower researchers to unravel complex<br /> biological processes and facilitate targeted interventions. This has significant<br /> implications for drug discovery and structural biology more broadly. More<br /> generally, latent variable models are deployed widely across many scientific<br /> disciplines. Thus, the argument we present in this work has much broader<br /> applications in AI for science if we want to move from impressive nonlinear<br /> neural network models to mathematically grounded methods that can help us<br /> learn something new about nature.<br /> <br /> == Keywords ==<br /> <br /> == Links ==<br /> <br /> https://www.frontiersin.org/journals/molecular-biosciences/articles/10.3389/fmolb.2024.1393564/full<br /> <br /> == Related software ==<br /> <br /> == Related methods ==<br /> <br /> == Comments ==</div>

WikiSysop