The functions of proteins and RNAs are defined by the collective interactions of many residues, and yet most statistical models of biological sequences consider sites nearly independently. Recent approaches have demonstrated benefits of including interactions to capture pairwise covariation, but leave higher-order dependencies out of reach. Here we show how it is possible to capture higher-order, context-dependent constraints in biological sequences via latent variable models with nonlinear dependencies. We found that DeepSequence ( https://github.com/debbiemarkslab/DeepSequence ), a probabilistic model for sequence families, predicted the effects of mutations across a variety of deep mutational scanning experiments substantially better than existing methods based on the same evolutionary data. The model, learned in an unsupervised manner solely on the basis of sequence information, is grounded with biologically motivated priors, reveals the latent organization of sequence families, and can be used to explore new parts of sequence space.

Quantifying the pathogenicity of protein variants in human disease-related genes would have a profound impact on clinical decisions, yet the overwhelming majority (over 98%) of these variants still have unknown consequences [1][2][3] . In principle, computational methods could support the largescale interpretation of genetic variants. However, prior methods 4-7 have relied on training machine learning models on available clinical labels. Since these labels are sparse, biased, and of variable quality, the resulting models have been considered insufficiently reliable 8 . By contrast, our approach leverages deep generative models to predict the clinical significance of protein variants without relying on labels. The natural distribution of protein sequences we observe across organisms is the result of billions of evolutionary experiments 9,10 . By modeling that distribution, we implicitly capture constraints on the protein sequences that maintain fitness. Our model EVE (Evolutionary model of Variant Effect) not only outperforms computational approaches that rely on labelled data, but also performs on par, if not better than, high-throughput assays which are increasingly used as strong evidence for variant classification [11][12][13][14][15][16][17][18][19][20][21][22][23] . After thorough validation on clinical labels, we predict the pathogenicity of 11 million variants across 1,081 disease genes, and assign high-confidence reclassification for 72k Variants of Unknown Significance 8 . Our work suggests that models of evolutionary information can provide a strong source of independent evidence for variant interpretation and that the approach will be widely useful in research and clinical settings.