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.

Low-N protein engineering with data-efficient deep learning


Protein engineering has enormous academic and industrial potential. However, it is limited by the lack of experimental assays that are consistent with the design goal and sufficiently high throughput to find rare, enhanced variants. Here we introduce a machine learning-guided paradigm that can use as few as 24 functionally assayed mutant sequences to build an accurate virtual fitness landscape and screen ten million sequences via in silico directed evolution. As demonstrated in two dissimilar proteins, GFP from Aequorea victoria (avGFP) and E. coli strain TEM-1 β-lactamase, top candidates from a single round are diverse and as active as engineered mutants obtained from previous high-throughput efforts. By distilling information from natural protein sequence landscapes, our model learns a latent representation of ‘unnaturalness’, which helps to guide search away from nonfunctional sequence neighborhoods. Subsequent low-N supervision then identifies improvements to the activity of interest. In sum, our approach enables efficient use of resource-intensive high-fidelity assays without sacrificing throughput, and helps to accelerate engineered proteins into the fermenter, field and clinic.

This is a preview of subscription content

Access options

Buy article

Get time limited or full article access on ReadCube.


All prices are NET prices.

Fig. 1: UniRep-guided in silico directed evolution for low-N protein engineering.
Fig. 2: eUniRep enables low-N engineering of avGFP.
Fig. 3: eUniRep enables low-N engineering of the enzyme TEM-1 β-lactamase using only single mutants as training data.
Fig. 4: eUniRep designs are structurally non-trivial and require both unsupervised training and low-N supervised training to discover >WT variants.

Data availability

Data required to reproduce all analyses in this work are provided or can be found at All referenced PDB structures were obtained from The Sarkisyan dataset was obtained from

Code availability

Code for UniRep model training and inference with trained weights along with links to all necessary data is available at Code required to reproduce all analyses in this work is provided at


  1. Romero, P. A. & Arnold, F. H. Exploring protein fitness landscapes by directed evolution. Nat. Rev. Mol. Cell Biol. 10, 866–876 (2009).

    CAS  Article  Google Scholar 

  2. Packer, M. S. & Liu, D. R. Methods for the directed evolution of proteins. Nat. Rev. Genet. 16, 379–394 (2015).

    CAS  Article  Google Scholar 

  3. Lutz, S. & Patrick, W. M. Novel methods for directed evolution of enzymes: quality, not quantity. Curr. Opin. Biotechnol. 15, 291–297 (2004).

  4. Goldsmith, M. & Tawfik, D. S. Directed enzyme evolution: beyond the low-hanging fruit. Curr. Opin. Struct. Biol. 22, 406–412 (2012).

  5. Zhao, H. & Arnold, F. H. Combinatorial protein design: strategies for screening protein libraries. Curr. Opin. Struct. Biol. 7, 480–485 (1997).

    CAS  Article  Google Scholar 

  6. You, L. & Arnold, F. H. Directed evolution of subtilisin E in Bacillus subtilis to enhance total activity in aqueous dimethylformamide. Protein Eng. 9, 77–83 (1996).

    CAS  Article  Google Scholar 

  7. Lagassé, H. A. D. et al. Recent advances in (therapeutic protein) drug development. F1000Res. 6, 113 (2017).

    Article  Google Scholar 

  8. Marshall, S. A., Lazar, G. A., Chirino, A. J. & Desjarlais, J. R. Rational design and engineering of therapeutic proteins. Drug Discov. Today 8, 212–221 (2003).

    CAS  Article  Google Scholar 

  9. Rao, A. G. The outlook for protein engineering in crop improvement. Plant Physiol. 147, 6–12 (2008).

    CAS  Article  Google Scholar 

  10. Schmid, A. et al. Industrial biocatalysis today and tomorrow. Nature 409, 258–268 (2001).

    CAS  Article  Google Scholar 

  11. Sheldon, R. A. & Pereira, P. C. Biocatalysis engineering: the big picture. Chem. Soc. Rev. 46, 2678–2691 (2017).

    CAS  Article  Google Scholar 

  12. Mullard, A. Better screening and disease models needed. Nat. Rev. Drug Discov. 15, 751–769 (2016).

    Article  Google Scholar 

  13. Scannell, J. W. & Bosley, J. When quality beats quantity: decision theory, drug discovery, and the reproducibility crisis. PLoS ONE 11, e0147215 (2016).

    Article  Google Scholar 

  14. Hughes, J. P., Rees, S., Kalindjian, S. B. & Philpott, K. L. Principles of early drug discovery. Br. J. Pharmacol. 162, 1239–1249 (2011).

    CAS  Article  Google Scholar 

  15. Scannell, J. W., Blanckley, A., Boldon, H. & Warrington, B. Diagnosing the decline in pharmaceutical R&D efficiency. Nat. Rev. Drug Discov. 11, 191–200 (2012).

    CAS  Article  Google Scholar 

  16. Laverty, H. et al. How can we improve our understanding of cardiovascular safety liabilities to develop safer medicines? Br. J. Pharmacol. 163, 675–693 (2011).

    CAS  Article  Google Scholar 

  17. Silver, L. L. Challenges of antibacterial discovery. Clin. Microbiol. Rev. 24, 71–109 (2011).

    CAS  Article  Google Scholar 

  18. Wu, Z., Jennifer Kan, S. B., Lewis, R. D., Wittmann, B. J. & Arnold, F. H. Machine learning-assisted directed protein evolution with combinatorial libraries. Proc. Natl Acad. Sci. USA 116, 8852–8858 (2019).

    CAS  Article  Google Scholar 

  19. Lutz, S. Beyond directed evolution—semi-rational protein engineering and design. Curr. Opin. Biotechnol. 21, 734–743 (2010).

  20. Bedbrook, C. N., Yang, K. K., Rice, A. J., Gradinaru, V. & Arnold, F. H. Machine learning to design integral membrane channelrhodopsins for efficient eukaryotic expression and plasma membrane localization. PLoS Comput. Biol. 13, e1005786 (2017).

    Article  Google Scholar 

  21. Bedbrook, C. N. et al. Machine learning-guided channelrhodopsin engineering enables minimally invasive optogenetics. Nat. Methods 16, 1176–1184 (2019).

    CAS  Article  Google Scholar 

  22. Romney, D. K., Murciano-Calles, J., Wehrmüller, J. E. & Arnold, F. H. Unlocking reactivity of TrpB: a general biocatalytic platform for synthesis of tryptophan analogues. J. Am. Chem. Soc. 139, 10769–10776 (2017).

    CAS  Article  Google Scholar 

  23. Silva, D. A., Yu, S., Ulge, U. Y., Spangler, J. B. & Jude, K. M. De novo design of potent and selective mimics of IL-2 and IL-15. Nature 565, 186–191 (2019).

  24. Marcandalli, J., Fiala, B., Ols, S. & Perotti, M. Induction of potent neutralizing antibody responses by a designed protein nanoparticle vaccine for respiratory syncytial virus. Cell 176, 1420–1431 (2019).

  25. Alley, E. C., Khimulya, G., Biswas, S., AlQuraishi, M. & Church, G. M. Unified rational protein engineering with sequence-based deep representation learning. Nat. Methods 16, 1315–1322 (2019).

    CAS  Article  Google Scholar 

  26. Halevy, A., Norvig, P. & Pereira, F. The unreasonable effectiveness of data. In IEEE Intelligent Systems (IEEE, 2009).

  27. Hénaff, O. J. et al. Data-efficient image recognition with contrastive predictive coding. In Proc. 37th Int. Conf. Machine Learning 119, 4182–4192 (2020).

  28. Ogden, P. J., Kelsic, E. D., Sinai, S. & Church, G. M. Comprehensive AAV capsid fitness landscape reveals a viral gene and enables machine-guided design. Science 336, 1139–1143 (2019).

  29. Biswas, S. et al. Toward machine-guided design of proteins. Preprint at bioRxiv (2018).

  30. Brookes, D. H., Park, H. & Listgarten, J. Conditioning by adaptive sampling for robust design. Preprint at (2019).

  31. Gupta, A. & Zou, J. Feedback GAN for DNA optimizes protein functions. Nat. Mach. Intell. 1, 105–111 (2019).

    Article  Google Scholar 

  32. Cadet, F., Fontaine, N., Li, G., Sanchis, J. & Chong, M. N. F. A machine learning approach for reliable prediction of amino acid interactions and its application in the directed evolution of enantioselective enzymes. Sci. Rep. 8, 16757 (2018).

  33. Saito, Y., Oikawa, M., Nakazawa, H. & Niide, T. Machine-learning-guided mutagenesis for directed evolution of fluorescent proteins. ACS Synth. Biol. 7, 2014–2022 (2018).

  34. Musdal, Y., Govindarajan, S. & Mannervik, B. Exploring sequence–function space of a poplar glutathione transferase using designed information-rich gene variants. Protein Eng. Des. Sel. 30, 543–549 (2017).

    CAS  Article  Google Scholar 

  35. Romero, P. A., Krause, A. & Arnold, F. H. Navigating the protein fitness landscape with Gaussian processes. Proc. Natl Acad. Sci. USA 110, E193–E201 (2013).

    CAS  Article  Google Scholar 

  36. Liao, J. et al. Engineering proteinase K using machine learning and synthetic genes. BMC Biotechnol. 7, 16 (2007).

  37. Fox, R. J. et al. Improving catalytic function by ProSAR-driven enzyme evolution. Nat. Biotechnol. 25, 338–344 (2007).

    CAS  Article  Google Scholar 

  38. Riesselman, A. J., Ingraham, J. B. & Marks, D. S. Deep generative models of genetic variation capture the effects of mutations. Nat. Methods 15, 816–822 (2018).

    CAS  Article  Google Scholar 

  39. Hopf, T. A., Ingraham, J. B., Poelwijk, F. J. & Schärfe, C. P. I. Mutation effects predicted from sequence co-variation. Nature 35, 128–135 (2017).

  40. Sinai, S., Kelsic, E., Church, G. M. & Nowak, M. A. Variational auto-encoding of protein sequences. Preprint at (2017).

  41. Shin, J.-E. et al. Protein design and variant prediction using autoregressive generative models. Preprint at bioRxiv (2019).

  42. Sarkisyan, K. S. et al. Local fitness landscape of the green fluorescent protein. Nature 533, 397–401 (2016).

    CAS  Article  Google Scholar 

  43. Ashkenazy, H. & Penn, O. FastML: a web server for probabilistic reconstruction of ancestral sequences. Nucleic Acids Res. 40, W580–W584 (2012).

  44. Gumulya, Y. & Gillam, E. M. J. Exploring the past and the future of protein evolution with ancestral sequence reconstruction: the ‘retro’ approach to protein engineering. Biochem. J. 474, 1–19 (2017).

  45. Sternke, M., Tripp, K. W. & Barrick, D. Consensus sequence design as a general strategy to create hyperstable, biologically active proteins. Proc. Natl Acad. Sci. USA 116, 11275–11284 (2019).

    CAS  Article  Google Scholar 

  46. Porebski, B. T. & Buckle, A. M. Consensus protein design. Protein Eng. Des. Sel. 29, 245–251 (2016).

  47. Russ, W. P. et al. An evolution-based model for designing chorismate mutase enzymes. Science 369, 440–445 (2020).

    CAS  PubMed  Google Scholar 

  48. Firnberg, E., Labonte, J. W. & Gray, J. J. A comprehensive, high-resolution map of a gene’s fitness landscape. Mol. Biol. Evol. 31, 1581–1592 (2014).

  49. Breen, M. S., Kemena, C., Vlasov, P. K., Notredame, C. & Kondrashov, F. A. Epistasis as the primary factor in molecular evolution. Nature 490, 535–538 (2012).

    CAS  Article  Google Scholar 

  50. Povolotskaya, I. S. & Kondrashov, F. A. Sequence space and the ongoing expansion of the protein universe. Nature 465, 922–926 (2010).

    CAS  Article  Google Scholar 

  51. Schenk, M. F., Szendro, I. G., Salverda, M. L. M., Krug, J. & de Visser, J. A. G. M. Patterns of epistasis between beneficial mutations in an antibiotic resistance gene. Mol. Biol. Evol. 30, 1779–1787 (2013).

    CAS  Article  Google Scholar 

  52. Repecka, D. et al. Expanding functional protein sequence spaces using generative adversarial networks. Nat. Mach. Intell. (2021).

  53. Manheim, D. & Garrabrant, S. Categorizing variants of Goodhart’s Law. Preprint at (2018).

  54. Dou, J. et al. De novo design of a fluorescence-activating β barrel. Nature 561, 485–491 (2018).

  55. Lu, P., Min, D., DiMaio, F., Wei, K. Y. & Vahey, M. D. Accurate computational design of multipass transmembrane proteins. Science 359, 1042–1046 (2018).

  56. Bick, M. J. et al. Computational design of environmental sensors for the potent opioid fentanyl. eLife 6, e28909 (2017).

    Article  Google Scholar 

  57. Zhang, R. K., Chen, K., Huang, X. & Wohlschlager, L. Enzymatic assembly of carbon–carbon bonds via iron-catalysed sp3 C–H functionalization. Nature 565, 67–72 (2019).

  58. Bornscheuer, U. T. & Pohl, M. Improved biocatalysts by directed evolution and rational protein design. Curr. Opin. Chem. Biol. 5, 137–134 (2001).

  59. Huang, P.-S., Boyken, S. E. & Baker, D. The coming of age of de novo protein design. Nature 537, 320–327 (2016).

    CAS  Article  Google Scholar 

  60. Chen, R. Enzyme engineering: rational redesign versus directed evolution. Trends Biotechnol. 19, 13–14 (2001).

  61. Alford, R. F. et al. The Rosetta all-atom energy function for macromolecular modeling and design. J. Chem. Theory Comput. 13, 3031–3048 (2017).

    CAS  Article  Google Scholar 

  62. Pédelacq, J.-D., Cabantous, S., Tran, T., Terwilliger, T. C. & Waldo, G. S. Engineering and characterization of a superfolder green fluorescent protein. Nat. Biotechnol. 24, 79–88 (2006).

    Article  Google Scholar 

  63. Dror, A., Shemesh, E. & Dayan, N. Protein engineering by random mutagenesis and structure-guided consensus of Geobacillus stearothermophilus lipase T6 for enhanced stability in methanol. Appl. Environ. Microbiol. 80, 1515–1527 (2014).

  64. Rocklin, G. J., Chidyausiku, T. M., Goreshnik, I. & Ford, A. Global analysis of protein folding using massively parallel design, synthesis, and testing. Science 357, 168–175 (2017).

  65. Wannier, T. M. et al. Monomerization of far-red fluorescent proteins. Proc. Natl Acad. Sci. USA 115, E11294–E11301 (2018).

    CAS  Article  Google Scholar 

  66. Xie, Q., Dai, Z., Hovy, E., Luong, M.-T. & Le, Q. V. Unsupervised data augmentation for consistency training. Preprint at (2019).

  67. Berthelot, D. et al. MixMatch: a holistic approach to semi-supervised learning. Preprint at (2019).

  68. Radford, A., Jozefowicz, R. & Sutskever, I. Learning to generate reviews and discovering sentiment. Preprint at (2017).

  69. Devlin, J., Chang, M.-W., Lee, K. & Toutanova, K. BERT: pre-training of deep bidirectional transformers for language understanding. Preprint at (2018).

  70. Potter, S. C., Luciani, A., Eddy, S. R. & Park, Y. HMMER web server: 2018 update. Nucleic Acids Res. 46, W200–W204 (2018).

  71. Caruana, R., Lawrence, S. & Giles, C. L. Overfitting in neural nets: backpropagation, conjugate gradient, and early stopping. In Advances in Neural Information Processing Systems (NIPS, 2001).

  72. Maclaurin, D., Duvenaud, D. & Adams, R. P. Early stopping is nonparametric variational inference. Preprint at (2015).

  73. Yang, K. K., Wu, Z., Bedbrook, C. N. & Arnold, F. H. Learned protein embeddings for machine learning. Bioinformatics 34, 2642–2648 (2018).

  74. Lambert, T. J. FPbase: a community-editable fluorescent protein database. Nat. Methods 16, 277–278 (2019).

  75. Arnold, F. H. & Georgiou, G. (eds) Directed Evolution Library Creation: Methods and Protocols. (Humana Press, 2010).

  76. van Dijk, D. et al. Recovering gene interactions from single-cell data using data diffusion. Cell 174, 716–729 (2018).

    Article  Google Scholar 

  77. Le, Q. & Mikolov, T. Distributed representations of sentences and documents. In Proc. 31st Int. Conf. Machine Learning 32, 1188–1196 (PMLR, 2014).

  78. Efron, B., Hastie, T., Johnstone, I. & Tibshirani, R. Least angle regression. Ann. Stat. 32, 407–499 (2004).

  79. Sohka, T. et al. An externally tunable bacterial band-pass filter. Proc. Natl Acad. Sci. USA 106, 10135–10140 (2009).

  80. Oberacker, P. et al. Bio-On-Magnetic-Beads (BOMB): open platform for high-throughput nucleic acid extraction and manipulation. PLoS Biol. 17, e3000107 (2019).

    Article  Google Scholar 

  81. Kosuri, S. et al. Composability of regulatory sequences controlling transcription and translation in Escherichia coli. Proc. Natl Acad. Sci. USA 110, 14024–14029 (2013).

    CAS  Article  Google Scholar 

  82. Magoč, T. & Salzberg, S. L. FLASH: fast length adjustment of short reads to improve genome assemblies. Bioinformatics 27, 2957–2963 (2011).

  83. Stiffler, M. A., Hekstra, D. R. & Ranganathan, R. Evolvability as a function of purifying selection in TEM-1 β-lactamase. Cell 160, 882–892 (2015).

  84. AlQuraishi, M. ProteinNet: a standardized data set for machine learning of protein structure. BMC Bioinformatics 20, 311 (2019).

  85. Kabsch, W. & Sander, C. Dictionary of protein secondary structure: pattern recognition of hydrogen‐bonded and geometrical features. Biopolymers 22, 2577–2637 (1983).

  86. Chen, H. & Zhou, H. X. Prediction of solvent accessibility and sites of deleterious mutations from protein sequence. Nucleic Acids Res. 33, 3193–3199 (2005).

Download references


We thank M. AlQuraishi, C. Bakerlee, A. Chiappino-Pepe, A. Eremina, K. Fish, S. Gosai, X. Guo, E. Kelsic, S. Kosuri, P. Ogden, S. Sinai, M. Schubert, A. Taylor-Weiner, D. Thompson and A. Tucker for feedback on earlier drafts of this manuscript. We thank members of the Esvelt and Church laboratories for valuable discussion. S.B. was supported by an NSF GRFP Fellowship under grant number DGE1745303. G.K. was supported by a grant from the Center for Effective Altruism. E.C.A. was supported by a scholarship from the Open Philanthropy Project. This material is based upon work supported by the US Department of Energy, Office of Science under award number DE‐FG02‐02ER63445. Computational resources were, in part, generously provided by the AWS Cloud Credits for Research Program and Lambda Labs, Inc.

Author information

Authors and Affiliations



S.B., G.K. and E.C.A. conceived the study. S.B. performed wet-lab experiments and managed data. S.B., G.K. and E.C.A. performed machine learning modeling and data analyses. K.M.E. and G.M.C. supervised the project. S.B., G.K. and E.C.A. wrote the manuscript with help from all authors.

Corresponding author

Correspondence to George M. Church.

Ethics declarations

Competing interests

A full list of G.M.C.’s technology transfer, advisory roles and funding sources can be found on the laboratory’s website at S.B. is employed by and holds equity in Nabla Bio, Inc. G.K. is employed by and holds equity in Telis Bioscience Inc. E.C.A. and K.M.E. declare no competing interests.

Additional information

Peer review information Nature Methods thanks Gabriel Rocklin, Guillaume Lamoureux, and the other, anonymous reviewer, for their contribution to the peer review of this work. Arunima Singh was the primary editor on this article and managed its editorial process and peer review in collaboration with the rest of the editorial team.

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

Supplementary information

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Biswas, S., Khimulya, G., Alley, E.C. et al. Low-N protein engineering with data-efficient deep learning. Nat Methods 18, 389–396 (2021).

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI:

Further reading


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