23 November 2021 in Proceedings of the National Academy of Sciences
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.
15 April 2015 in Protein Engineering, Design and Selection
Systemic injection of granulocyte colony-stimulating factor (G-CSF) has yielded encouraging results in treating Alzheimer's Disease (AD) and other central nervous system (CNS) disorders. Making G-CSF a viable AD therapeutic will, however, require increasing G-CSF's ability to stimulate neurons within the brain. This objective could be realized by increasing transcytosis of G-CSF across the blood brain barrier (BBB). An established correlation between G-CSF receptor (G-CSFR) binding pH responsiveness and increased recycling of G-CSF to the cell exterior after endocytosis motivated development of G-CSF variants with highly pH responsive G-CSFR binding affinities. These variants will be used in future validation of our hypothesis that increased BBB transcytosis can enhance G-CSF therapeutic efficacy. Flow cytometric screening of a yeast-displayed library in which G-CSF/G-CSFR interface residues were mutated to histidine yielded a G-CSF triple His mutant (L109H/D110H/Q120H) with highly pH responsive binding affinity. This variant's KD, measured by surface plasmon resonance (SPR), increases ∼20-fold as pH decreases from 7.4 to below histidine's pKa of ∼6.0; an increase 2-fold greater than for previously reported G-CSF His mutants. Cell-free protein synthesis (CFPS) enabled expression and purification of soluble, bioactive G-CSF triple His variant protein, an outcome inaccessible via Escherichia coli inclusion body refolding. This purification and bioactivity validation will enable future identification of correlations between pH responsiveness and transcytosis in BBB cell culture model and animal experiments. Furthermore, the library screening and CFPS methods employed here could be applied to developing other pH responsive hematopoietic or neurotrophic factors for treating CNS disorders.
25 October 2020 in bioRxiv
The mapping from protein sequence to function is highly complex, making it challenging to predict how sequence changes will affect a protein’s behavior and properties. We present a supervised deep learning framework to learn the sequence-function mapping from deep mutational scanning data and make predictions for new, uncharacterized sequence variants. We test multiple neural network architectures, including a graph convolutional network that incorporates protein structure, to explore how a network’s internal representation affects its ability to learn the sequence-function mapping. Our supervised learning approach displays superior performance over physics-based and unsupervised prediction methods. We find networks that capture nonlinear interactions and share parameters across sequence positions are important for learning the relationship between sequence and function. Further analysis of the trained models reveals the networks’ ability to learn biologically meaningful information about protein structure and mechanism. Finally, we demonstrate the models’ ability to navigate sequence space and design new proteins beyond the training set. We applied the GB1 models to design a sequence that binds to IgG with substantially higher affinity than wild-type GB1. Our software is available from https://github.com/gitter-lab/nn4dms . 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. In this work, we show 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 sequencefunction relationship, and can learn general rules that apply across the length of the protein sequence. We demonstrate the learned models can be applied to design new proteins with properties that exceed natural sequences.
15 May 2015 in Journal of Biological Engineering
Engineered antibodies with pH responsive cell surface target antigen-binding affinities that decrease at the acidic pH (5.5-5.8) within the endosomes have been found to have reduced susceptibility to degradation within the lysosomes and increased serum half-life. Such pH responsive recombinant antibodies have been developed for the treatment of cancer and cardiovascular disease. Engineered tenth type III human fibronectin (Fn3) domains are emerging as a class of target antigen-binding biopharmaceuticals that could complement or be superior to recombinant antibodies in a number of biomedical contexts. As such, there is strong motivation for demonstrating the feasibility of engineering Fn3s with pH responsive antigen binding behavior that could lead to improved Fn3 pharmacokinetics. A yeast surface-displayed Fn3 histidine (His) mutant library screening approach yielded epidermal growth factor receptor (EGFR)-binding Fn3 domains with EGFR binding affinities that markedly decrease at endosomal pH; the first reported case of engineering Fn3s with pH responsive antigen binding. Yeast surface-displayed His mutant Fn3s, which contain either one or two His mutations, have equilibrium binding dissociation constants (KDs) that increase up to four-fold relative to wild type when pH is decreased from 7.4 to 5.5. Assays in which Fn3-displaying yeast were incubated with soluble EGFR after ligand-free incubation in respective neutral and acidic buffers showed that His mutant Fn3 pH responsiveness is due to reversible changes in Fn3 conformation and/or EGFR binding interface properties rather than irreversible unfolding. We have established a generalizable method for efficiently constructing and screening Fn3 His mutant libraries that could enable both our laboratory and others to develop pH responsive Fn3s for use in a wide range of biomedical applications.
9 November 2023 in bioRxiv
Machine learning (ML) has transformed protein engineering by constructing models of the underlying sequence-function landscape to accelerate the discovery of new biomolecules. ML-guided protein design requires models, trained on local sequence-function information, to accurately predict distant fitness peaks. In this work, we evaluate neural networks’ capacity to extrapolate beyond their training data. We perform model-guided design using a panel of neural network architectures trained on protein G (GB1)-Immunoglobulin G (IgG) binding data and experimentally test thousands of GB1 designs to systematically evaluate the models’ extrapolation. We find each model architecture infers markedly different landscapes from the same data, which give rise to unique design preferences. We find simpler models excel in local extrapolation to design high fitness proteins, while more sophisticated convolutional models can venture deep into sequence space to design proteins that fold but are no longer functional. Our findings highlight how each architecture’s inductive biases prime them to learn different aspects of the protein fitness landscape.