Skip to main content

Thank you for visiting You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Deep learning study of tyrosine reveals that roaming can lead to photodamage


Amino acids are among the building blocks of life, forming peptides and proteins, and have been carefully ‘selected’ to prevent harmful reactions caused by light. To prevent photodamage, molecules relax from electronic excited states to the ground state faster than the harmful reactions can occur; however, such photochemistry is not fully understood, in part because theoretical simulations of such systems are extremely expensive—with only smaller chromophores accessible. Here, we study the excited-state dynamics of tyrosine using a method based on deep neural networks that leverages the physics underlying quantum chemical data and combines different levels of theory. We reveal unconventional and dynamically controlled ‘roaming’ dynamics in excited tyrosine that are beyond chemical intuition and compete with other ultrafast deactivation mechanisms. Our findings suggest that the roaming atoms are radicals that can lead to photodamage, offering a new perspective on the photostability and photodamage of biological systems.

This is a preview of subscription content, access via your institution

Access options

Rent or buy this article

Prices vary by article type



Prices may be subject to local taxes which are calculated during checkout

Fig. 1: Roaming hydrogen atoms in tyrosine discovered during a photodynamics simulation.
Fig. 2: Photoproducts of tyrosine observed in ‘non-roaming’ and ‘roaming’ trajectories.
Fig. 3: Kinetic analysis of photodissociation in tyrosine.

Data availability

The molecular coordinates of the conformers used in this study that are shown in Supplementary Fig. 1 and Supplementary Fig. 9 are available in the Supplementary Information. Additionally, the dataset is made available at (ref. 66) in the Atomic Simulation Environment (ase)67 database format including the initial conditions (geometries and velocities) to set up the dynamics. A detailed description of the ML models, training set generation, training procedure, dynamics simulations and analysis is provided in the Supplementary Information. For reproduction of this work, replace the example dataset provided in the tutorial with the provided dataset and set the necessary parameters indicated in the Supplementary Information and listed in the tutorial instructions. For dynamics, set up the trajectories with SHARC68 using the initial conditions file available at (ref. 66) and follow the tutorial on how to run dynamics with SchNarc. Source data are provided with this paper.

Code availability

The ML code used in this work is available at (ref. 25) and included as a supplementary code. The development branch includes a tutorial for training SchNarc models and running dynamics with it.


  1. Collini, E. et al. Coherently wired light-harvesting in photosynthetic marine algae at ambient temperature. Nature 463, 644–647 (2010).

    Article  CAS  PubMed  Google Scholar 

  2. Cerullo, G. et al. Photosynthetic light harvesting by carotenoids: detection of an intermediate excited state. Science 298, 2395–2398 (2002).

    Article  CAS  PubMed  Google Scholar 

  3. Domcke, W. & Sobolewski, A. L. Peptide deactivation: spectroscopy meets theory. Nat. Chem. 5, 257–258 (2013).

    Article  CAS  PubMed  Google Scholar 

  4. Ashfold, M. N. R., Cronin, B., Devine, A. L., Dixon, R. N. & Nix, M. G. D. The role of πσ* excited states in the photodissociation of heteroaromatic molecules. Science 312, 1637–1640 (2006).

    Article  CAS  PubMed  Google Scholar 

  5. Schreier, W. J. et al. Thymine dimerization in DNA is an ultrafast photoreaction. Science 315, 625–629 (2007).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  6. Rauer, C., Nogueira, J. J., Marquetand, P. & González, L. Cyclobutane thymine photodimerization mechanism revealed by nonadiabatic molecular dynamics. J. Am. Chem. Soc. 138, 15911–15916 (2016).

    Article  CAS  PubMed  Google Scholar 

  7. Wang, Y. et al. Intravenous treatment of choroidal neovascularization by photo-targeted nanoparticles. Nat. Commun. 10, 804 (2019).

  8. Marder, S. R., Kippelen, B., Jen, N., Alex, K.-Y. & Peyghambarian, N. Design and synthesis of chromophores and polymers for electro-optic and photorefractive applications. Nature 388, 845–851 (1997).

    Article  CAS  Google Scholar 

  9. Sanchez-Lengeling, B. & Aspuru-Guzik, A. Inverse molecular design using machine learning: generative models for matter engineering. Science 361, 360–365 (2018).

    Article  CAS  PubMed  Google Scholar 

  10. Chen, J. et al. Optical nano-imaging of gate-tunable graphene plasmons. Nature 487, 77–81 (2012).

    Article  CAS  PubMed  Google Scholar 

  11. Zewail, A. H. in Femtochemistry 3–22 (World Scientific, 1994).

  12. Wörner, H. J., Bertrand, J. B., Kartashov, D. V., Corkum, P. B. & Villeneuve, D. M. Following a chemical reaction using high-harmonic interferometry. Nature 466, 604–607 (2010).

    Article  PubMed  CAS  Google Scholar 

  13. Mai, S. & González, L. Molecular photochemistry: recent developments in theory. Angew. Chem. Int. Ed. 59, 16832–16846 (2020).

  14. Tseng, C.-M. et al. Photostability of amino acids: photodissociation dynamics of phenylalanine chromophores. Phys. Chem. Chem. Phys. 12, 4989–4995 (2010).

    Article  CAS  PubMed  Google Scholar 

  15. Roberts, G. M. & Stavros, V. G. The role of πσ* states in the photochemistry of heteroaromatic biomolecules and their subunits: insights from gas-phase femtosecond spectroscopy. Chem. Sci. 5, 1698–1722 (2014).

    Article  CAS  Google Scholar 

  16. Iqbal, A. & Stavros, V. G. Active participation of 1πσ* states in the photodissociation of tyrosine and its subunits. J. Phys. Chem. Lett. 1, 2274–2278 (2010).

    Article  CAS  Google Scholar 

  17. Tseng, C.-M., Lee, Y. T., Ni, C.-K. & Chang, J.-L. Photodissociation dynamics of the chromophores of the amino acid tyrosine: p-methylphenol, p-ethylphenol, and p-(2-aminoethyl)phenol. J. Phys. Chem. A 111, 6674–6678 (2007).

    Article  CAS  PubMed  Google Scholar 

  18. Sobolewski, A. L. & Domcke, W. Ab initio investigations on the photophysics of indole. Chem. Phys. Lett. 315, 293–298 (1999).

    Article  CAS  Google Scholar 

  19. Oliver, T. A. A., Zhang, Y., Roy, A., Ashfold, M. N. R. & Bradforth, S. E. Exploring autoionization and photoinduced proton-coupled electron transfer pathways of phenol in aqueous solution. J. Phys. Chem. Lett. 6, 4159–4164 (2015).

    Article  CAS  PubMed  Google Scholar 

  20. Xie, C. et al. Nonadiabatic tunneling in photodissociation of phenol. J. Am. Chem. Soc. 138, 7828–7831 (2016).

    Article  CAS  PubMed  Google Scholar 

  21. Iqbal, A. Towards Understanding the Photochemistry of Tyrosine. PhD thesis, Univ. of Warwick (2010).

  22. Tomasello, G., Wohlgemuth, M., Petersen, J. & Mitrić, R. Photodynamics of free and solvated tyrosine. J. Phys. Chem. B 116, 8762–8770 (2012).

    Article  CAS  PubMed  Google Scholar 

  23. Sobolewski, A. L., Shemesh, D. & Domcke, W. Computational studies of the photophysics of neutral and zwitterionic amino acids in an aqueous environment: tyrosine-(H2O)2 and tryptophan-(H2O)2 clusters. J. Phys. Chem. A 113, 542–550 (2009).

    Article  CAS  PubMed  Google Scholar 

  24. Westermayr, J. et al. Machine learning enables long time scale molecular photodynamics simulations. Chem. Sci. 10, 8100–8107 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  25. Westermayr, J., Gastegger, M. & Marquetand, P. Combining SchNet and SHARC: the SchNarc machine learning approach for excited-state dynamics. J. Phys. Chem. Lett. 11, 3828–3834 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  26. Bowman, J. M. & Suits, A. G. Roaming reactions: the third way. Phys. Today 64, 33 (2011).

    Article  CAS  Google Scholar 

  27. Bowman, J. M. & Shepler, B. C. Roaming radicals. Ann. Rev. Phys. Chem. 62, 531–553 (2011).

    Article  CAS  Google Scholar 

  28. Herath, N. & Suits, A. G. Roaming radical reactions. J. Phys. Chem. Lett. 2, 642–647 (2011).

    Article  CAS  Google Scholar 

  29. Townsend, D. et al. The roaming atom: straying from the reaction path in formaldehyde decomposition. Science 306, 1158–1161 (2004).

    Article  CAS  PubMed  Google Scholar 

  30. Ekanayake, N. et al. Mechanisms and time-resolved dynamics for trihydrogen cation (H3+) formation from organic molecules in strong laser fields. Sci. Rep. 7, 4703 (2017).

  31. Lu, Z., Chang, Y. C., Yin, Q.-Z., Ng, C. Y. & Jackson, W. M. Evidence for direct molecular oxygen production in CO2 photodissociation. Science 346, 61–64 (2014).

    Article  CAS  PubMed  Google Scholar 

  32. Mereshchenko, A. S., Butaeva, E. V., Borin, V. A., Eyzips, A. & Tarnovsky, A. N. Roaming-mediated ultrafast isomerization of geminal tri-bromides in the gas and liquid phases. Nat. Chem. 7, 562–568 (2015).

    Article  CAS  PubMed  Google Scholar 

  33. Tso, C.-J., Kasai, T. & Lin, K.-C. Roaming dynamics and conformational memory in photolysis of formic acid at 193 nm using time-resolved Fourier-transform infrared emission spectroscopy. Sci. Rep. 10, 4769 (2020).

  34. Endo, T. et al. Capturing roaming molecular fragments in real time. Science 370, 1072–1077 (2020).

    Article  CAS  PubMed  Google Scholar 

  35. Nandi, A., Zhang, P., Chen, J., Guo, H. & Bowman, J. M. Quasiclassical simulations based on cluster models reveal vibration-facilitated roaming in the isomerization of CO adsorbed on NaCl. Nat. Chem. 13, 249–254 (2021).

    Article  CAS  PubMed  Google Scholar 

  36. Suits, A. G. Roaming reactions and dynamics in the van der Waals region. Annu. Rev. Phys. Chem. 71, 77–100 (2020).

    Article  CAS  PubMed  Google Scholar 

  37. Dreuw, A. & Wormit, M. The algebraic diagrammatic construction scheme for the polarization propagator for the calculation of excited states. Wiley Interdiscip. Rev. Comput. Mol. Sci. 5, 82–95 (2015).

    Article  CAS  Google Scholar 

  38. Roos, B. O., Taylor, P. R. & Siegbahn, P. E. A complete active space SCF method (CASSCF) using a density matrix formulated super-CI approach. Chem. Phys. 48, 157–173 (1980).

    Article  CAS  Google Scholar 

  39. Finley, J., Malmqvist, P.-A., Roos, B. O. & Serrano-Andrés, L. The multi-state CASPT2 method. Chem. Phys. Lett. 288, 299–306 (1998).

    Article  CAS  Google Scholar 

  40. Westermayr, J. & Marquetand, P. Deep learning for UV absorption spectra with SchNarc: first steps toward transferability in chemical compound space. J. Chem. Phys. 153, 154112 (2020).

    Article  CAS  PubMed  Google Scholar 

  41. Gastegger, M., Behler, J. & Marquetand, P. Machine learning molecular dynamics for the simulation of infrared spectra. Chem. Sci. 8, 6924–6935 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  42. Schütt, K. T. et. al. Machine Learning Meets Quantum Physics (Springer International Publishing, Cham, 2020).

  43. Pedregosa, F. et al. Scikit-learn: machine learning in Python. J. Mach. Learn. Res. 12, 2825–2830 (2011).

    Google Scholar 

  44. Richter, M., Marquetand, P., González-Vázquez, J., Sola, I. & González, L. Femtosecond intersystem crossing in the DNA nucleobase cytosine. J. Phys. Chem. Lett. 3, 3090–3095 (2012).

    Article  CAS  PubMed  Google Scholar 

  45. Richter, M., Mai, S., Marquetand, P. & González, L. Ultrafast intersystem crossing dynamics in uracil unravelled by ab initio molecular dynamics. Phys. Chem. Chem. Phys. 16, 24423–24436 (2014).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  46. Marazzi, M., Sancho, U., Castano, O., Domcke, W. & Frutos, L. M. Photoinduced proton transfer as a possible mechanism for highly efficient excited-state deactivation in proteins. J. Phys. Chem. Lett. 1, 425–428 (2009).

    Article  CAS  Google Scholar 

  47. Shemesh, D., Sobolewski, A. L. & Domcke, W. Efficient excited-state deactivation of the Gly-Phe-Ala tripeptide via an electron-driven proton-transfer process. J. Am. Chem. Soc. 131, 1374–1375 (2009).

    Article  CAS  PubMed  Google Scholar 

  48. Behler, J. Four generations of high-dimensional neural network potentials. Chem. Rev. 121, 10037–10072 (2021).

    Article  CAS  PubMed  Google Scholar 

  49. Westermayr, J. & Marquetand, P. Machine learning for electronically excited states of molecules. Chem. Rev. 121, 9873–9926 (2021).

    Article  CAS  PubMed  Google Scholar 

  50. Crespo-Otero, R. & Barbatti, M. Recent advances and perspectives on nonadiabatic mixed quantum-classical dynamics. Chem. Rev. 118, 7026–7068 (2018).

    Article  CAS  PubMed  Google Scholar 

  51. Giesbertz, K. & Baerends, E. Failure of time-dependent density functional theory for excited state surfaces in case of homolytic bond dissociation. Chem. Phys. Lett. 461, 338–342 (2008).

    Article  CAS  Google Scholar 

  52. Kidwell, N., Li, H., Wang, X., Bowman, J. M. & Lester, M. I. Unimolecular dissociation dynamics of vibrationally activated CH3CHOO Criegee intermediates to OH radical products. Nat. Chem. 8, 509–514 (2016).

    Article  CAS  PubMed  Google Scholar 

  53. Truong, T., Behrsohn, R., Brumer, P., Luk, C. K. & Tao, T. Effect of pH on the phosphorescence of tryptophan, tyrosine, and proteins. J. Biol. Chem. 242, 2979–2985 (1967).

    Article  CAS  PubMed  Google Scholar 

  54. Schütt, K. T., Gastegger, M., Tkatchenko, A., Müller, K.-R. & Maurer, R. J. Unifying machine learning and quantum chemistry with a deep neural network for molecular wavefunctions. Nat. Commun. 10, 5024 (2019).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  55. Westermayr, J. & Maurer, R. J. Physically inspired deep learning of molecular excitations and photoemission spectra. Chem. Sci. 12, 10755–10764 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  56. Tully, J. C. Molecular dynamics with electronic transitions. J. Chem. Phys. 93, 1061–1071 (1990).

    Article  CAS  Google Scholar 

  57. Tully, J. C. Nonadiabatic molecular dynamics. Int. J. Quantum Chem. 40, 299–309 (1991).

    Article  Google Scholar 

  58. Richter, M., Marquetand, P., González-Vázquez, J., Sola, I. & González, L. SHARC: ab initio molecular dynamics with surface hopping in the adiabatic representation including arbitrary couplings. J. Chem. Theory Comput. 7, 1253–1258 (2011).

    Article  CAS  PubMed  Google Scholar 

  59. Mai, S., Marquetand, P. & González, L. Nonadiabatic dynamics: the SHARC approach. Wiley Interdiscip. Res. Comput. Mol. Sci. 8, e1370 (2018).

    Google Scholar 

  60. Westermayr, J., Faber, F. A., Christensen, A. S., von Lilienfeld, O. A. & Marquetand, P. Neural networks and kernel ridge regression for excited states dynamics of CH2NH2+: from single-state to multi-state representations and multi-property machine learning models. Mach. Learn. Sci. Technol. 1, 025009 (2020).

    Article  Google Scholar 

  61. Westermayr, J. M. Machine Learning for Excited-State Molecular Dynamics Simulations. PhD thesis, Univ. of Vienna (2020).

  62. Schütt, K. T., Sauceda, H. E., Kindermans, P.-J., Tkatchenko, A. & Müller, K.-R. SchNet – a deep learning architecture for molecules and materials. J. Chem. Phys. 148, 241722 (2018).

    Article  PubMed  CAS  Google Scholar 

  63. Schütt, K. T. et al. SchNetPack: a deep learning toolbox for atomistic systems. J. Chem. Theory Comput. 15, 448–455 (2019).

    Article  PubMed  CAS  Google Scholar 

  64. Hirshfeld, F. Bonded-atom fragments for describing molecular charge densities. Theoret. Chim. Acta 44, 129–138 (1977).

    Article  CAS  Google Scholar 

  65. Mulliken, R. S. Electronic population analysis on LCAO–MO molecular wave functions. I. J. Chem. Phys. 23, 1833–1840 (1955).

    Article  CAS  Google Scholar 

  66. Westermayr, J. Tyrosine_ExcitedStates. figshare (2021).

  67. Larsen, A. H. et al. The atomic simulation environment—a Python library for working with atoms. J. Phys. Condens. Matter 29, 273002 (2017).

    Article  Google Scholar 

  68. Mai, S. et al. SHARC2.0: Surface Hopping Including ARbitrary Couplings – Program Package for Non-Adiabatic Dynamics (, 2018).

Download references


This work was financially supported by the Austrian Science Fund (W 1232 (MolTag doctoral program “Ion channels and transporters as molecular drug targets”) and J 4522-N (J.W.)) and the uni:docs programme of the University of Vienna (J.W.). The computational results presented have been achieved in part using the Vienna Scientific Cluster. P.M. and L.G. thank the University of Vienna for continuous support, and also support in the frame of the research platform ViRAPID (Vienna Research Platform on Accelerating Photoreaction Discovery). J.W. and P.M. are grateful for an Nvidia Hardware Grant. M.G. works at the BASLEARN Joint Lab for Machine Learning, cofinanced by the Technische Universität Berlin and BASF SE.

Author information

Authors and Affiliations



P.M. and L.G. proposed the project and supervised it. J.W. and P.M. implemented and designed the methods. M.G. and D.V. helped with fitting the neural networks. D.V., L.P. and F.J. contributed to the reference calculations and the training set generation, that is, the generation of adjusted data points. J.W. performed the model training, data acquisition and model analysis. J.W. and P.M. interpreted the data, designed the analysis and wrote the initial manuscript. L.G., P.M., J.W. and M.G. revised the manuscript. All authors proofread the final manuscript and the Supplementary Information.

Corresponding author

Correspondence to Philipp Marquetand.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Chemistry thanks Joel Bowman and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended Data Fig. 1 Validation of our method to combine ADC(2) and CASPT2, the generation of ad hoc data, and their influence on neural network training.

(a) Potential energy curves along the PhO-H bond using CASSCF(12,11) (left panel, solid lines), CASPT2(12,11) (left panel, dotted lines), and ADC(2) (right panel). (b) Absorption spectra of tyrosine computed with ADC(2), TD-DFT/PBE0/SV(P), CASSCF(12,11), and CASPT(12,11) spectra using 1000 Wigner-sampled conformations. For CASPT2(12,11) and CASSCF(12,11) 758 calculations converged. The full width at half maximum for the Gaussian convolution was 0.2 eV in ADC(2) and TD-DFT spectra and 0.5 eV in CASPT2(12,11) and CASSCF(12,11) spectra and absorption peaks are scaled such that the lowestenergy peak has the same height as the energetically lowest-lying experimental absorption peak that was extracted from ref. 46. (c) Spectrum obtained from ADC(2) redshifted by 0.3 eV with contributions from different excited states. (d) Tyrosine shown with circled hydrogen atoms that were detached to generate ad hoc data (Supplementary Table 3 and 4). (e) The Ph-OH and (f) the N-H bond as example reaction coordinates to show the differences of ADC(2) (solid and dashed lines), CASPT2(12,11) (circles), and MP2 calculations (crosses) that manifest the degeneracy of singlet and triplet states. Second plots illustrates the data used for training and the third and fourth plots show neural network predictions using a simple multi-layer feed-forward neural network (Supplementary section 7.A) faster trainable than SchNarc models that were trained without ad hoc data points and with ad hoc data points, respectively. S refers to singlet and T to triplet states.

Source data

Supplementary information

Supplementary Information

Supplementary Figs. 1–17, Tables 1–6 and Discussion.

Supplementary Video 1

Roaming atoms in tyrosine.

Supplementary Video 2

Recoiling atoms in tyrosine.

Supplementary Software 1

Code for SchNarc dynamics as available on github at

Supplementary Data 1

The xyz structures of tyrosine conformers, roaming geometries and dissociation data points.

Supplementary Data 2

Source data for the Supplementary Information.

Source data

Source Data Fig. 1

Structures of the first trajectory, in which roaming atoms were observed.

Source Data Fig. 3

Analysis of dynamics simulations.

Source Data Extended Data Fig. 1

Potential energy curves and UV/visible absorption spectra to support the method validation and the generation of ad hoc data points.

Rights and permissions

Reprints and Permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Westermayr, J., Gastegger, M., Vörös, D. et al. Deep learning study of tyrosine reveals that roaming can lead to photodamage. Nat. Chem. 14, 914–919 (2022).

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI:


Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing