Design of efficient molecular organic light-emitting diodes by a high-throughput virtual screening and experimental approach

Journal name:
Nature Materials
Year published:
Published online


Virtual screening is becoming a ground-breaking tool for molecular discovery due to the exponential growth of available computer time and constant improvement of simulation and machine learning techniques. We report an integrated organic functional material design process that incorporates theoretical insight, quantum chemistry, cheminformatics, machine learning, industrial expertise, organic synthesis, molecular characterization, device fabrication and optoelectronic testing. After exploring a search space of 1.6 million molecules and screening over 400,000 of them using time-dependent density functional theory, we identified thousands of promising novel organic light-emitting diode molecules across the visible spectrum. Our team collaboratively selected the best candidates from this set. The experimentally determined external quantum efficiencies for these synthesized candidates were as large as 22%.

At a glance


  1. Discovery pipeline.
    Figure 1: Discovery pipeline.

    a, Diagram of the collaborative discovery approach: the search space decreases by over five orders of magnitude as the screening progresses. The cubes represent the size of the chemical space considered at any given stage of the process. The distinct screening stages, from left to right, involve different theoretical and computational approaches as well as experimental input and testing. b, Dependency tree for the quantum chemistry calculations employed in this study. The calculations labelled as backbone were performed for all analysed molecules, leading compounds were also characterized using the methods labelled emission, and the benchmarking calculations were used to assess predictive power.

  2. Experiment-theory calibration.
    Figure 2: Experiment–theory calibration.

    a, TD-DFT/B3LYP/6-31G(d) vertical absorption against photoluminescent emission maximum in toluene solution. Data in red represent measurements reported in the literature, while data in blue correspond to compounds synthesized in this screening work. The lines indicate a linear fit against the literature data (solid) with 95% confidence bounds (dashed). b, Experiment–theory comparison of TD-DFT/B3LYP/6-31G(d) singlet–triplet gap against experimental values, determined via the thermal activation energy method, measured in frozen toluene solution. The lines indicate a linear fit against the literature data (solid) with 95% confidence bounds (dashed).

  3. Effectiveness of machine learning.
    Figure 3: Effectiveness of machine learning.

    a, Fraction of molecules in the test set correctly ranked in the top 5%, as a function of the amount of training data. b, Root mean square error (RMSE) in log(kTADF) as a function of the amount of training data. c, Linear model predictions against the TD-DFT-derived data with the largest training set. R2 = 0.80. d, Neural network predictions against the TD-DFT-derived data with the largest training set. R2 = 0.94.

  4. Candidate statistics and voting tool.
    Figure 4: Candidate statistics and voting tool.

    a, Number of screened molecules as a function of singlet–triplet splitting (ΔEST) and oscillator strength (f). Contour lines represent estimated kTADF (μs−1) assuming S1 at 3.0eV. b, Number of screened molecules as a function of kTADF and S1 energy. Vertical dashed line corresponds to kTADF = 1μs−1. c, ΔEST and f for decision batches A–N, colour-coded by each of the generations in the candidate library. The box-and-whisker plots represent the statistical distribution of predictions for all the candidates in each batch. The bottom and top of the bar are the first and third quartiles, and the band inside the box is the median. The lines extending vertically from the boxes indicate the maximum and minimum of the range. d, Screenshot from the interactive web tool for molecular voting.

  5. Lead candidates and optoelectronic characterization.
    Figure 5: Lead candidates and optoelectronic characterization.

    a, Promising molecules that were synthesized and tested. Compound abbreviations are composed by the first letter of the batch of origin and a running index. b, Device structure and energy band diagram of lead candidates. Energies in eV. Thickness in nm. c, Electroluminescence spectra. d, External quantum efficiency as a function of current density. e, Current density and luminance as a function of applied voltage.


  1. Polishchuk, P. G., Madzhidov, T. I. & Varnek, A. Estimation of the size of drug-like chemical space based on GDB-17 data. J. Comput. Aided Mol. Des. 27, 675679 (2013).
  2. Shoichet, B. K. Virtual screening of chemical libraries. Nature 432, 862865 (2004).
  3. Curtarolo, S. et al. The high-throughput highway to computational materials design. Nature Mater. 12, 191201 (2013).
  4. Pyzer-Knapp, E. O., Suh, C., Gómez-Bombarelli, R., Aguilera-Iparraguirre, J. & Aspuru-Guzik, A. What is high-throughput virtual screening? A perspective from organic materials discovery. Annu. Rev. Mater. Res. 45, 195216 (2015).
  5. Yang, K., Setyawan, W., Wang, S., Buongiorno Nardelli, M. & Curtarolo, S. A search model for topological insulators with high-throughput robustness descriptors. Nature Mater. 11, 614619 (2012).
  6. Carrete, J., Li, W., Mingo, N., Wang, S. & Curtarolo, S. Finding unprecedentedly low-thermal-conductivity half-Heusler semiconductors via high-throughput materials modeling. Phys. Rev. X 4, 011019 (2014).
  7. Huskinson, B. et al. A metal-free organic–inorganic aqueous flow battery. Nature 505, 195198 (2014).
  8. Er, S., Suh, C., Marshak, M. P. & Aspuru-Guzik, A. Computational design of molecules for an all-quinone redox flow battery. Chem. Sci. 6, 885893 (2015).
  9. Hachmann, J. et al. The Harvard Clean Energy Project: large-scale computational screening and design of organic photovoltaics on the world community grid. J. Phys. Chem. Lett. 2, 22412251 (2011).
  10. Shin, Y., Liu, J., Quigley, J. J., Luo, H. & Lin, X. Combinatorial design of copolymer donor materials for bulk heterojunction solar cells. ACS Nano 8, 60896096 (2014).
  11. Hachmann, J. et al. Lead candidates for high-performance organic photovoltaics from high-throughput quantum chemistry—the Harvard Clean Energy Project. Energy Environ. Sci. 7, 698704 (2014).
  12. Baldo, M. A. et al. Highly efficient phosphorescent emission from organic electroluminescent devices. Nature 395, 151154 (1998).
  13. Yersin, H. Transition Metal and Rare Earth Compounds 126 (Springer, 2004).
  14. Jou, J.-H., Kumar, S., Agrawal, A., Li, T.-H. & Sahoo, S. Approaches for fabricating high efficiency organic light emitting diodes. J. Mater. Chem. C 3, 29743002 (2015).
  15. Tao, Y. et al. Thermally activated delayed fluorescence materials towards the breakthrough of organoelectronics. Adv. Mater. 26, 79317958 (2014).
  16. Endo, A. et al. Efficient up-conversion of triplet excitons into a singlet state and its application for organic light emitting diodes. Appl. Phys. Lett. 98, 083302 (2011).
  17. Zhang, Q. et al. Design of efficient thermally activated delayed fluorescence materials for pure blue organic light emitting diodes. J. Am. Chem. Soc. 134, 1470614709 (2012).
  18. Zhang, Q. et al. Efficient blue organic light-emitting diodes employing thermally activated delayed fluorescence. Nature Photon. 8, 326332 (2014).
  19. Yersin, H., Rausch, A. F., Czerwieniec, R., Hofbeck, T. & Fischer, T. The triplet state of organo-transition metal compounds. Triplet harvesting and singlet harvesting for efficient OLEDs. Coord. Chem. Rev. 255, 26222652 (2011).
  20. Dias, F. B. et al. Triplet harvesting with 100% efficiency by way of thermally activated delayed fluorescence in charge transfer OLED emitters. Adv. Mater. 25, 37073714 (2013).
  21. Jankus, V. et al. Highly efficient TADF OLEDs: How the emitter–host interaction controls both the excited state species and electrical properties of the devices to achieve near 100% triplet harvesting and high efficiency. Adv. Funct. Mater. 24, 61786186 (2014).
  22. Zhang, Q. et al. Nearly 100% internal quantum efficiency in undoped electroluminescent devices employing pure organic emitters. Adv. Mater. 27, 20962100 (2015).
  23. Tanaka, H., Shizu, K., Lee, J. & Adachi, C. Effect of atom substitution in chalcogenodiazole-containing thermally activated delayed fluorescence emitters on radiationless transition. J. Phys. Chem. C 119, 29482955 (2015).
  24. Lee, J. et al. Controlled emission colors and singlet–triplet energy gaps of dihydrophenazine-based thermally activated delayed fluorescence emitters. J. Mater. Chem. C 3, 21752181 (2015).
  25. Wang, H. et al. Novel thermally activated delayed fluorescence materials–thioxanthone derivatives and their applications for highly efficient OLEDs. Adv. Mater. 26, 51985204 (2014).
  26. Lee, D. R., Hwang, S.-H., Jeon, S. K., Lee, C. W. & Lee, J. Y. Benzofurocarbazole and benzothienocarbazole as donors for improved quantum efficiency in blue thermally activated delayed fluorescent devices. Chem. Commun. 51, 81058107 (2015).
  27. Shizu, K. et al. Strategy for designing electron donors for thermally activated delayed fluorescence emitters. J. Phys. Chem. C 119, 12911297 (2015).
  28. Sagara, Y. et al. Highly efficient thermally activated delayed fluorescence emitters with a small singlet–triplet energy gap and large oscillator strength. Chem. Lett. 44, 360362 (2015).
  29. Shu, Y. & Levine, B. G. Simulated evolution of fluorophores for light emitting diodes. J. Chem. Phys. 142, 104104 (2015).
  30. Zhang, X. et al. Theoretical investigation of dihydroacridine and diphenylsulphone derivatives as thermally activated delayed fluorescence emitters for organic light-emitting diodes. RSC Adv. 5, 5158651591 (2015).
  31. RDKit: open source cheminformatics software (accessed 23 June 2015).
  32. Ertl, P. & Schuffenhauer, A. Estimation of synthetic accessibility score of drug-like molecules based on molecular complexity and fragment contributions. J. Cheminform. 1, 8 (2009).
  33. Baleizão, C. & Berberan-Santos, M. N. Thermally activated delayed fluorescence as a cycling process between excited singlet and triplet states: application to the fullerenes. J. Chem. Phys. 126, 204510 (2007).
  34. Hilborn, R. C. Einstein coefficients, cross sections, f values, dipole moments, and all that. Am. J. Phys. 50, 982986 (1982).
  35. Dreuw, A. & Head-Gordon, M. Single-reference ab initio methods for the calculation of excited states of large molecules. Chem. Rev. 105, 40094037 (2005).
  36. Gritsenko, O. & Baerends, E. J. Asymptotic correction of the exchange–correlation kernel of time-dependent density functional theory for long-range charge-transfer excitations. J. Chem. Phys. 121, 655660 (2004).
  37. Huang, S. et al. Computational prediction for singlet- and triplet-transition energies of charge-transfer compounds. J. Chem. Theory Comput. 9, 38723877 (2013).
  38. Penfold, T. J. On predicting the excited-state properties of thermally activated delayed fluorescence emitters. J. Phys. Chem. C 119, 1353513544 (2015).
  39. Moral, M., Muccioli, L., Son, W-J., Olivier, Y. & Sancho-García, J. C. Theoretical rationalization of the singlet–triplet gap in OLEDs materials: impact of charge-transfer character. J. Chem. Theory Comput. 11, 168177 (2015).
  40. Bergeron, C., Krein, M., Moore, G., Breneman, C. M. & Bennett, K. P. Modeling choices for virtual screening hit identification. Mol. Inform. 30, 765777 (2011).
  41. Eckert, H. & Bajorath, J. Molecular similarity analysis in virtual screening: foundations, limitations and novel approaches. Drug Discov. Today 12, 225233 (2007).
  42. Dahl, G. E., Jaitly, N. & Salakhutdinov, R. Multi-task neural networks for QSAR predictions. Preprint at (2014).
  43. Rogers, D. & Hahn, M. Extended-connectivity fingerprints. J. Chem. Inf. Model. 50, 742754 (2010).
  44. Nair, V. & Hinton, G. E. Rectified linear units improve restricted Boltzmann machines. Proc. 27th International Conference on Machine Learning (ICML-10) 807814 (2010).
  45. Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I. & Salakhutdinov, R. Dropout: a simple way to prevent neural networks from overfitting. J. Mach. Learn. Res. 15, 19291958 (2014).
  46. Snoek, J., Larochelle, H. & Adams, R. P. Practical Bayesian optimization of machine learning algorithms. Adv. Neural Inf. Process. Syst. 25, 29512959 (2012).
  47. Maclaurin, D., Duvenaud, D. & Johnson, M. J. HIPS/autograd GitHub (accessed 29 October 2015).
  48. Hirata, S. et al. Highly efficient blue electroluminescence based on thermally activated delayed fluorescence. Nature Mater. 14, 330336 (2015).
  49. Tosco, P., Stiefl, N. & Landrum, G. Bringing the MMFF force field to the RDKit: implementation and validation. J. Cheminformat. 6, 37 (2014).
  50. Shao, Y. et al. Advances in molecular quantum chemistry contained in the Q-Chem 4 program package. Mol. Phys. 113, 184215 (2015).
  51. Ufimtsev, I. S. & Martinez, T. J. Quantum chemistry on graphical processing units. 3. Analytical energy gradients, geometry optimization, and first principles molecular dynamics. J. Chem. Theory Comput. 5, 26192628 (2009).

Download references

Author information


  1. Department of Chemistry and Chemical Biology, 12 Oxford Street, Harvard University, Cambridge, Massachusetts 02138, USA

    • Rafael Gómez-Bombarelli,
    • Jorge Aguilera-Iparraguirre,
    • Timothy D. Hirzel,
    • Martin A. Blood-Forsythe &
    • Alán Aspuru-Guzik
  2. John A. Paulson School of Engineering and Applied Sciences, 33 Oxford Street, Harvard University, Cambridge, Massachusetts 02138, USA

    • David Duvenaud,
    • Dougal Maclaurin &
    • Ryan P. Adams
  3. Samsung Research America, 255 Main Street, Suite 702, Cambridge, Massachusetts 02142, USA

    • Hyun Sik Chae &
    • Seong Ik Hong
  4. Department of Electrical Engineering and Computer Science, 77 Massachusetts Avenue, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA

    • Markus Einzinger,
    • Tony Wu &
    • Marc Baldo
  5. Department of Materials Science and Engineering, 77 Massachusetts Avenue, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA

    • Dong-Gwang Ha
  6. Department of Chemistry, 77 Massachusetts Avenue, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA

    • Georgios Markopoulos &
    • Wenliang Huang
  7. Samsung Advanced Institute of Technology, Samsung Electronics Co., Ltd., 130 Samsung-ro, Yeongtong-gu, Suwon-si, Gyeonggi-do 16678, Korea

    • Soonok Jeon,
    • Hosuk Kang,
    • Hiroshi Miyazaki,
    • Masaki Numata &
    • Sunghan Kim


A.A.-G., M.B. and R.P.A. conceived the project. T.D.H. designed and wrote the custom computer code for molecular screening, with contributions from R.G.-B. and J.A.-I. R.G.-B. and J.A.-I. designed the molecules, with contributions from A.A.-G., H.M., M.N. and H.S.C. R.G.-B. and J.A.-I. performed calculations and analysed theoretical predictions. M.A.B.-F. carried out the experimental calibration of the theoretical methods. D.M., D.D. and R.P.A. applied machine learning to the computational predictions. H.S.C. and G.M. assessed synthetic feasibility of molecular candidates, with contributions from W.H., S.J., H.M., M.N. and S.K. R.G.-B., J.A.-I., T.D.H., H.S.C., M.A.B.-F., G.M., D.M., D.D., S.H., S.J., H.M., M.N., S.K., R.P.A., M.B. and A.A.-G. selected the molecules for characterization. S.J. synthesized J1-2 and L1. S.J., H.S.C., T.W., D.-G.H. and M.E. collected and analysed spectroscopic data. D.-G.H., M.E. and T.W. manufactured and tested devices for F1, J1, J2 and L1, with contributions from H.K. R.G.-B., J.A.-I. and T.D.H. wrote the first version of the manuscript. All authors contributed to the discussion, writing and editing of the manuscript. A.A.-G. and R.P.A. supervised the computational chemistry study. R.P.A. and A.A.-G. supervised the machine learning approach. M.B. supervised the device fabrication.

Competing financial interests

The authors declare no competing financial interests.

Corresponding author

Correspondence to:

Author details

Supplementary information

PDF files

  1. Supplementary Information (7,705 KB)

    Supplementary Information

Additional data