Deep generative models of genetic variation capture the effects of mutations

By Adam J. Riesselman, John Ingraham, Debora Marks. 2018

627 citations recorded by the source corpus.

Subfields: Molecular Biology, Generative Models, Computational Biology & Bioinformatics

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Molecular Biology

In Generative Models

In Computational Biology & Bioinformatics

Citations in the corpus

Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences (2019, background): “The generative picture can be extended to include latent variables parameterized by neural networks that capture higher order interactions than pairwise (Riesselman et al., 2018).”
MSA Transformer (2021, methodology): “Models ﬁt to MSAs are widely used in computational biology including in applications such as ﬁtness landscape prediction (Riesselman et al., 2018), pathogenicity prediction (Sundaram et al., 2018; Frazer et al., 2020), remote homology detection (Hou et al., 2018), and protein design (Russ et al.,…”
Generating functional protein variants with variational autoencoders (2020, methodology): “Unlike [22] we 321 focus on the generative capabilities of VAE models, and, when working with aligned sequence 322 input, filter the alignment in such a way as to allow the generation of full variant sequences of 323 a single target protein.”
Evolutionary context-integrated deep sequence modeling for protein engineering (2020, methodology): “Several existing computational methods predict the mutation effects by leveraging the evolutionary information of homologous sequences [24,25] .”
Accelerating Protein Design Using Autoregressive Generative Models (2019, methodology): “Mutation effects, sequence families, and previous effect predictions for validation were curated from published work (Riesselman et al., 2018).”
Evaluating Protein Transfer Learning with TAPE (2019, positive): “[32] trained Variational Auto Encoders on aligned families of proteins to predict the functional impact of mutations.”
Transformer protein language models are unsupervised structure learners (2020, background): “Recently deep learning methods have also been applied to ﬁtting MSAs, and Riesselman et al. (2018) found evidence that factors learned by a VAE model may correlate with protein structure.”
Language models enable zero-shot prediction of the effects of mutations on protein function (2021, methodology): “Compared with EVMutation [4] and DeepSequence [20], MSA Transformer and ESM-1v require no task-specific model training for inference.”
Generative Models for Graph-Based Protein Design (2019, background): “This may be seen as a kind of evolutionary unsupervised anomaly detection [18].”
Artificial intelligence in clinical and genomic diagnostics (2019, background): “In addition, deep generative models have shown promise for predicting the effects of genetic variants [61], and are especially intriguing given their ability to evaluate the joint influence of multiple genetic variants and/or complex indels on protein function, a capability that is largely absent…”
How to Hallucinate Functional Proteins (2019, methodology): “(18; 19) used their models for downstream classification tasks and did not use them explicitly for design.”
Evolutionary velocity with protein language models (2021, positive): “In contrast to previousmethods for predicting evolution, which assume protein-specific evolutionary models (for example, training a language model specific to a protein family) (Hie et al., 2021; Riesselman et al., 2018), we use a single language model to make all of our predictions across a diversi”
High-performance medicine: the convergence of human and artificial intelligence (2019, background): “There has been much written about the black box of algorithms, and much controversy surrounding this topic 191–193 ; especially in the case of DNNs, it may not be possible to understand the determination of output.”
Mega-scale experimental analysis of protein folding stability in biology and design (2023, result): “Notably, these offsets are similar to findings from an independent analysis of global discrepancies between variant effect data and sequence-likelihood modelling 41 .”
Efficient generative modeling of protein sequences using simple autoregressive models (2021, neutral): “The DeepSequence implementation [43] was originally designed and tested for predicting the effects of mutations around a given wild type.”