Design of metalloproteins and novel protein folds using variational autoencoders

By Joe G. Greener, Lewis Moffat, David T. Jones. 2018

140 citations recorded by the source corpus.

The design of novel proteins has many applications but remains an attritional process with success in isolated cases. Meanwhile, deep learning technologies have exploded in popularity in recent years and are increasingly applicable to biology due to the rise in available data. We attempt to link protein design and deep learning by using variational autoencoders to generate protein sequences conditioned on desired properties. Potential copper and calcium binding sites are added to non-metal binding proteins without human intervention and compared to a hidden Markov model. In another use case, a grammar of protein structures is developed and used to produce sequences for a novel protein topology. One candidate structure is found to be stable by molecular dynamics simulation. The ability of our model to confine the vast search space of protein sequences and to scale easily has the potential to assist in a variety of protein design tasks.

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

Generating functional protein variants with variational autoencoders (2020, methodology): “Conditional 337 VAEs with feed-forward decoders have previously been used to generate sequences with specified 338 metal binding sites or structural topologies when trained on raw sequences spanning multiple fami339 lies [25].”
Accelerating Protein Design Using Autoregressive Generative Models (2019, background): “We also anticipate better strategies to explore diverse yet ﬁt regions of sequence space during generation, either by exploiting variance explained by latent variables (Greener et al., 2018) or diverse beam search strategies (Vijayakumar et al., 2016).”
Learning from Protein Structure with Geometric Vector Perceptrons (2020, background): “, 2017), or context-free grammars of protein topology (Greener et al., 2018).”
Protein sequence design with deep generative models (2021, background): “For example, a VAE decoder can be conditioned on the identity of the metal cofactors bound [34].”
Protein sequence design with a learned potential (2020, background): “Other studies that attempt to leverage machine learning for the structure-guided sequence design task include using neural networks to individually predict residues in a sequence [80, 81], building a variational autoencoder conditioned on protein topology to generate designs [82], and framing protei”
De Novo Protein Design for Novel Folds using Guided Conditional Wasserstein Generative Adversarial Networks (gcWGAN) (2019, methodology): “In contrast to the forward problem, the inverse problem of de novo protein design has witnessed limited impacts from deeply exploiting data with advanced artiﬁcial intelligence (especially deep learning) technologies (Greener et al. , 2018; Ingraham et al. , 2019).”
Using AlphaFold for Rapid and Accurate Fixed Backbone Protein Design (2021, methodology): “A variety of neural network architectures have been used including variational autoencoders (Greener et al., 2018; Hawkins-Hooker et al., 2021), deep exploration networks (Linder et al., 2020), graph neural networks (Strokach et al., 2020), recurrent neural networks (Alley et al., 2019) and…”
Recent developments in deep learning applied to protein structure prediction (2019, background): “learning is also starting to show a lot of promise in the area of protein design.(11,12) This article is not intended to be a detailed exposition of every”
Efficient generative modeling of protein sequences using simple autoregressive models (2021, methodology): “Several research groups explored the possibility of generating sequences with given three-dimensional structure [39–41], e.g. via a VAE [42] or a Graph Neural Network [43], or by inverting structural prediction models [44–47].”
Generative Models for Graph-Based Protein Design (2019, background): “[15] introduced a generative model for protein sequences conditioned on a 1D, contextfree grammar based specification of the fold topology.”
Function-guided protein design by deep manifold sampling (2021, methodology): “Additionally, a few generative models have been proposed for conditional design [17, 6, 8]: Greener et al. [17] use a VAE conditioned on structural features for generating sequences with metal binding sites, Madani et al. [6] use a conditional LM for sampling proteins, where each amino acid is…”