Neural networks to learn protein sequence–function relationships from deep mutational scanning data
Нейронные сети для изучения взаимосвязей между последовательностью и функцией белков по данным глубокого мутационного сканирования
2021-11-23
SCID: 54.1/ww63sezy
Discuss with AI
deep mutational scanninggraph convolutional networkprotein G B1 domainprotein sequence designsequence-function mapping
Figures from the paper
Abstract (AI)
Significance Understanding the relationship between protein sequence and function is necessary to design new and useful proteins with applications in bioenergy, medicine, and agriculture. The mapping from sequence to function is tremendously complex because it involves thousands of molecular interactions that are coupled over multiple lengths and timescales. We show that neural networks can learn the sequence–function mapping from large protein datasets. Neural networks are appealing for this task because they can learn complicated relationships from data, make few assumptions about the nature of the sequence–function relationship, and can learn general rules that apply across the length of the protein sequence. We demonstrate that learned models can be applied to design new proteins with properties that exceed natural sequences.
Key Findings
1
A supervised deep learning framework learns protein sequence–function mappings from deep mutational scanning data and predicts uncharacterized variants.
2
Nonlinear interactions and parameter sharing across sequence positions are important architectural features for learning sequence–function mappings.
3
Supervised neural networks outperform physics-based and unsupervised methods for predicting protein sequence–function relationships.
4
The models can navigate sequence space to design proteins beyond the training set; a designed GB1 variant binds immunoglobulin G with substantially higher affinity than wild-type GB1.
5
Trained models learn biologically meaningful information about protein structure and molecular mechanisms, including when using structure-aware graph convolutional networks.
Research Object
protein sequence variants, including the protein G B1 domain (GB1)
Research Subject
sequence–function relationships and the ability to predict and design protein variants with altered function, including immunoglobulin G binding affinity
Publication Details
Publication Date
2021-11-23
Journal
Publisher
ISSN
Cited by
183
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest
References available in scid.ai4
A Comprehensive Survey on Graph Neural Networks2020
Inferring the shape of global epistasis2018
Epistatic Net allows the sparse spectral regularization of deep neural networks for inferring fitness functions2021
Investigating the determinants of performance in machine learning for protein fitness prediction2025
Cited by4
Global epistasis on fitness landscapes2023
MoCHI: neural networks to fit interpretable models and quantify energies, energetic couplings, epistasis, and allostery from deep mutational scanning data2024
Data‐Driven Protein Engineering for Improving Catalytic Activity and Selectivity2023
Investigating the determinants of performance in machine learning for protein fitness prediction2025