Machine-learning-guided directed evolution for protein engineering

Abstract

Machine learning (ML)-guided directed evolution is a new paradigm for biological design that enables optimization of complex functions. ML methods use data to predict how sequence maps to function without requiring a detailed model of the underlying physics or biological pathways. To demonstrate ML-guided directed evolution, we introduce the steps required to build ML sequence-function models and use them to guide engineering, making recommendations at each stage. This review covers basic concepts relevant to using ML for protein engineering as well as the current literature and applications of this new engineering paradigm. ML methods accelerate directed evolution by learning from information contained in all measured variants and using that information to select sequences that are likely to be improved. We then provide two case studies that demonstrate the ML-guided directed evolution process. We also look to future opportunities where ML will enable discovery of new protein functions and uncover the relationship between protein sequence and function.

Expanding functional protein sequence space using generative adversarial networks

“…This imposes a great burden on experimental approaches aiming to design novel protein sequences, such as random mutagenesis 4 and recombination of naturally occurring homologous proteins 8,9 , as up to 70% of random amino acid substitutions typically result in a decline of protein activity and 50% are deleterious to protein function 4,[10][11][12][13][14][15][16] . On the other hand, Artificial intelligence (AI) is not limited by the amount of sequence variations it can process [17][18][19] and, instead of depending on a blind search process, is based on an inference-based one -it infers protein properties 18,20 and function 19,21 directly from training examples.…”

“…Recent AI approaches have also demonstrated great potential in capturing both the structural and evolutionary information found in natural protein sequences 17,22 . Nevertheless, the majority of existing machine learning models in biology are discriminative 17,18,21 , i.e., the model is trained, using readily available data, to predict the properties of a given protein sequence.…”

“…Recent AI approaches have also demonstrated great potential in capturing both the structural and evolutionary information found in natural protein sequences 17,22 . Nevertheless, the majority of existing machine learning models in biology are discriminative 17,18,21 , i.e., the model is trained, using readily available data, to predict the properties of a given protein sequence. A generative modeling approach, in contrast, could generate new sequence samples from the learned portion of functional protein sequence space, giving direct access to unexplored sequence diversity within functional protein structures, without the need to test a large number of non-functional protein sequence variants.…”

Label free fluorescence quantification of hydrolytic enzyme activity on native substrates reveal how lipase function depends on membrane curvature

“…To support this plethora of functionalities lipases have been extensively engineered 12,13 , to meet the demand for increasingly specific chemical reactions 14 . Our capacity to optimize or alter protein function has radically improved through site directed mutagenesis of the active site 15,16 , directed evolution 17 , generation of novel functions using molecular dynamics (MD) simulations 18 or machine learning 17 . The advancement of high throughput methodologies on the other hand may provide large numbers of mutants that have to be screened for the desired properties.…”

On generative models of T-cell receptor sequences

“…Deep learning methods are proving a very useful approach in many areas of physics and the natural sciences [1,2]. These algorithms are successful in identifying hidden patterns in large amounts of data, often helping make progress in situations where traditional analyses reach their limits [3][4][5]. Despite the black box aspect of how the algorithm works and the lack of interpretability of the model features, machine learning is undoubtedly useful, especially in cases where the natural system of interest escapes our intuition or knowledge.…”