BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:From Data to Dynamics: Machine Learning in Statistical Mechanics a
 nd Molecular Simulations
DTSTART;VALUE=DATE:20261014
DTSTAMP:20260430T234431Z
UID:6447cf6381d6ec8af18e672bc85589d11951af6df863648b534b6d03
CATEGORIES:Conferences - Seminars
DESCRIPTION:You can apply to participate and find all the relevant informa
 tion (speakers\, abstracts\, program\,...) on the event website: https://
 www.cecam.org/workshop-details/from-data-to-dynamics-machine-learning-in-s
 tatistical-mechanics-and-molecular-simulations-1487.\n\nRegistration is re
 quired to attend the full event\, take part in the social activities and p
 resent a poster at the poster session (if any).  However\, the EPFL comm
 unity is welcome to attend specific lectures without registration if t
 he topic is of interest to their research. Do not hesitate to contact the
  CECAM Event Manager if you have any question.\n\nDescription\nSince its
  introduction in the 1970s\, molecular dynamics (MD) has become an indispe
 nsable computational microscope for studying complex biological systems at
  atomic resolution. It has enabled detailed investigations into protein fo
 lding\, conformational dynamics\, and ligand binding and unbinding. Over t
 he past decade\, increasing computational power has made microsecond-scale
  simulations routine\, producing massive datasets that demand sophisticate
 d analysis strategies [1]. Despite these advances\, conventional MD simula
 tions still face a fundamental limitation: many biologically relevant even
 ts occur over milliseconds to seconds—timescales largely inaccessible to
  standard MD.\nTo bridge this gap\, researchers increasingly turn to enhan
 ced sampling techniques—such as metadynamics and umbrella sampling [2\,3
 ]—and coarse-grained (CG) modeling approaches [4]. These methods enable 
 more comprehensive exploration of the system’s free energy landscape\, y
 et their success critically depends on the selection of appropriate reacti
 on coordinates or collective variables (CVs). CVs must capture the slowest
 \, most functionally relevant motions to accurately reflect thermodynamic 
 and kinetic behavior. However\, identifying suitable CVs remains one of th
 e field’s most challenging tasks\, typically requiring domain expertise 
 and iterative refinement [5\, 6].\nThis complexity has fueled growing inte
 rest in machine learning (ML) techniques\, which are now transforming how 
 MD simulations are analyzed\, interpreted\, and even conducted. ML methods
  have been applied to automate CV discovery\, perform dimensionality reduc
 tion\, build thermodynamic and kinetic models\, and enhance sampling effic
 iency [7]. These models often employ artificial neural networks or graph n
 eural networks to map high-dimensional molecular configurations—such as 
 Cartesian coordinates or molecular descriptors—into low-dimensional repr
 esentations suitable for analysis [8].\nDepending on the structure and typ
 e of data\, ML algorithms can be broadly categorized into supervised\, uns
 upervised\, and reinforcement learning paradigms [9]. Supervised learning 
 uses labeled input-output pairs to predict properties such as molecular en
 ergies or binding affinities [10]\, while unsupervised learning enables th
 e identification of latent features\, such as CVs\, directly from data [11
 ].\nA cornerstone of modern ML-driven simulation is the development of sym
 metry-aware molecular representations. The predictive power of ML models h
 inges on encoding physical symmetries—like rotation and translation—di
 rectly into the model. E(3)-equivariant neural networks have emerged as po
 werful tools for this purpose\, significantly improving data efficiency an
 d generalization in learning potential energy surfaces [12]. Ongoing resea
 rch continues to explore the optimal balance between enforcing strict symm
 etry and retaining model flexibility.\nMeanwhile\, breakthroughs in struct
 ural prediction—most notably the advent of AlphaFold 3—have revolution
 ized how researchers obtain initial molecular configurations. AlphaFold no
 w provides remarkably accurate models of not only proteins but also their 
 complexes with nucleic acids\, ions\, and small-molecule ligands [13]. How
 ever\, these are static snapshots. They cannot capture dynamic behaviors\,
  allosteric transitions\, or binding kinetics—areas where physics-based 
 simulations remain indispensable. Initial benchmarks suggest that even sta
 te-of-the-art predictors still fall short in modeling protein dynamics and
  ranking ligand binding affinities\, further emphasizing the role of MD [1
 4].\nTo address the dimensionality and sampling bottlenecks\, unsupervised
  ML approaches such as time-lagged autoencoders have reframed CV identific
 ation as a data-driven task. More recently\, generative models—including
  diffusion models and variational autoencoders—have emerged as a new fro
 ntier. These models can learn the full conformational landscape of biomole
 cules and enable enhanced sampling\, in some cases eliminating the need fo
 r predefined CVs altogether [15].\nOnce accurate structural models and CVs
  are established\, ML can significantly improve the estimation of thermody
 namic and kinetic properties. In drug discovery\, for instance\, predictin
 g protein–ligand binding affinity remains a central challenge. ML potent
 ials trained on quantum mechanical data can be combined with enhanced samp
 ling to yield highly accurate free energy landscapes and binding kinetics
 —results previously unattainable due to computational limitations [16]. 
 However\, challenges in data quality\, model interpretability\, and transf
 erability remain critical areas of ongoing investigation [17].\nFinally\, 
 ML is driving a renaissance in CG modeling. Deep neural networks can now l
 earn many-body CG potentials directly from all-atom simulations\, capturin
 g emergent properties and enhancing transferability [18]. These models ope
 n the door to longer\, larger-scale simulations with greater physical accu
 racy.\nIn this rapidly evolving context\, it becomes imperative to critica
 lly assess both the promise and limitations of ML in biomolecular simulati
 on. The excitement surrounding these developments must be tempered by care
 ful validation and benchmarking. This workshop thus serves as a timely opp
 ortunity—especially for early-career researchers—to explore these cutt
 ing-edge methods\, engage in constructive dialogue\, and chart new directi
 ons in the application of machine learning to molecular dynamics and drug 
 discovery.\n \nReferences\n\n[1] J. Behler\, M. Parrinello\, Phys. Rev. L
 ett.\, 98\, 146401 (2007)\n[2] P. Sahrmann\, G. Voth\, Current Opinion in
  Structural Biology\, 90\, 102972 (2025)\n[3] K. Kříž\, L. Schmidt\, A
 . Andersson\, M. Walz\, D. van der Spoel\, J. Chem. Inf. Model.\, 63\, 41
 2-431 (2023)\n[4] K. Ahmad\, A. Rizzi\, R. Capelli\, D. Mandelli\, W. Lyu\
 , P. Carloni\, Front. Mol. Biosci.\, 9\, (2022)\n[5] S. Mehdi\, Z. Smith\
 , L. Herron\, Z. Zou\, P. Tiwary\, Annual Review of Physical Chemistry\, 
 75\, 347-370 (2024)\n[6] H. Zheng\, H. Lin\, A. Alade\, J. Chen\, E. Monro
 y\, M. Zhang\, J. Wang\, AlphaFold3 in Drug Discovery: A Comprehensive Ass
 essment of Capabilities\, Limitations\, and Applications\, 2025\n[7] J. Ab
 ramson\, J. Adler\, J. Dunger\, R. Evans\, T. Green\, A. Pritzel\, O. Ronn
 eberger\, L. Willmore\, A. Ballard\, J. Bambrick\, S. Bodenstein\, D. Evan
 s\, C. Hung\, M. O’Neill\, D. Reiman\, K. Tunyasuvunakool\, Z. Wu\, A. 
 Žemgulytė\, E. Arvaniti\, C. Beattie\, O. Bertolli\, A. Bridgland\, A. C
 herepanov\, M. Congreve\, A. Cowen-Rivers\, A. Cowie\, M. Figurnov\, F. Fu
 chs\, H. Gladman\, R. Jain\, Y. Khan\, C. Low\, K. Perlin\, A. Potapenko\,
  P. Savy\, S. Singh\, A. Stecula\, A. Thillaisundaram\, C. Tong\, S. Yakne
 en\, E. Zhong\, M. Zielinski\, A. Žídek\, V. Bapst\, P. Kohli\, M. Jader
 berg\, D. Hassabis\, J. Jumper\, Nature\, 630\, 493-500 (2024)\n[8] Fabia
 n B. Fuchs\, Daniel E. Worrall\, Volker Fischer\, Max Welling\, NIPS'20: P
 roceedings of the 34th International Conference on Neural Information Proc
 essing Systems\, Article No.: 166\, Pages 1970 - 1981 (2020)\n[9] H. Sidky
 \, W. Chen\, A. Ferguson\, Molecular Physics\, 118\, (2020)\n[10] Y. Wang
 \, J. Lamim Ribeiro\, P. Tiwary\, Current Opinion in Structural Biology\,
  61\, 139-145 (2020)\n[11] K. Butler\, D. Davies\, H. Cartwright\, O. Isa
 yev\, A. Walsh\, Nature\, 559\, 547-555 (2018)\n[12] F. Noé\, A. Tkatche
 nko\, K. Müller\, C. Clementi\, Annu. Rev. Phys. Chem.\, 71\, 361-390 (2
 020)\n[13] S. Kaptan\, I. Vattulainen\, Advances in Physics: X\, 7\, (202
 2)\n[14] V. Limongelli\, WIREs. Comput. Mol. Sci.\, 10\, (2020)\n[15] A. 
 Glielmo\, B. Husic\, A. Rodriguez\, C. Clementi\, F. Noé\, A. Laio\, Chem
 . Rev.\, 121\, 9722-9758 (2021)\n[16] A. Pak\, G. Voth\, Current Opinion 
 in Structural Biology\, 52\, 119-126 (2018)\n[17] M. Lelimousin\, V. Limo
 ngelli\, M. Sansom\, J. Am. Chem. Soc.\, 138\, 10611-10622 (2016)\n[18] C
 . Abrams\, G. Bussi\, Entropy\, 16\, 163-199 (2013)\n 
LOCATION:Aula Magna\, USI Lugano https://www.desk.usi.ch/en/lugano-campus-
 map-access-facilities
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR