Antibody-SGM, a Score-Based Generative Model for Antibody Heavy-Chain Design.

Traditional computational methods for antibody design involved random mutagenesis followed by energy function assessment for candidate selection. Recently, diffusion models have garnered considerable attention as cutting-edge generative models, lauded for their remarkable performance. However, these methods often focus solely on the backbone or sequence, resulting in the incomplete depiction of the overall structure and necessitating additional techniques to predict the missing component. This study presents Antibody-SGM, an innovative joint structure-sequence diffusion model that addresses the limitations of existing protein backbone generation models. Unlike previous models, Antibody-SGM successfully integrates sequence-specific attributes and functional properties into the generation process. Our methodology generates full-atom native-like antibody heavy chains by refining the generation to create valid pairs of sequences and structures, starting with random sequences and structural properties. The versatility of our method is demonstrated through various applications, including the design of full-atom antibodies, antigen-specific CDR design, antibody heavy chains optimization, validation with Alphafold3, and the identification of crucial antibody sequences and structural features. Antibody-SGM also optimizes protein function through active inpainting learning, allowing simultaneous sequence and structure optimization. These improvements demonstrate the promise of our strategy for protein engineering and significantly increase the power of protein design models.

HelixDiff, a Score-Based Diffusion Model for Generating All-Atom α-Helical Structures.

Here, we present HelixDiff, a score-based diffusion model for generating all-atom helical structures. We developed a hot spot-specific generation algorithm for the conditional design of α-helices targeting critical hotspot residues in bioactive peptides. HelixDiff generates α-helices with near-native geometries for most test scenarios with root-mean-square deviations (RMSDs) less than 1 Å. Significantly, HelixDiff outperformed our prior GAN-based model with regard to sequence recovery and Rosetta scores for unconditional and conditional generations. As a proof of principle, we employed HelixDiff to design an acetylated GLP-1 D-peptide agonist that activated the glucagon-like peptide-1 receptor (GLP-1R) cAMP accumulation without stimulating the glucagon-like peptide-2 receptor (GLP-2R). We predicted that this D-peptide agonist has a similar orientation to GLP-1 and is substantially more stable in MD simulations than our earlier D-GLP-1 retro-inverse design. This D-peptide analogue is highly resistant to protease degradation and induces similar levels of AKT phosphorylation in HEK293 cells expressing GLP-1R compared to the native GLP-1. We then discovered that matching crucial hotspots for the GLP-1 function is more important than the sequence orientation of the generated D-peptides when constructing D-GLP-1 agonists.

Engineering a Rigid Nucleic Acid Structure to Improve the Limit of Detection for Genetic Assays.

Detecting nucleic acids at ultralow concentrations is critical for research and clinical applications. Particle-based assays are commonly used to detect nucleic acids. However, DNA hybridization on particle surfaces is inefficient due to the instability of tethered sequences, which negatively influences the assay's detection sensitivity. Here, we report a method to stabilize sequences on particle surfaces using a double-stranded linker at the 5' end of the tethered sequence. We termed this method Rigid Double Stranded Genomic Linkers for Improved DNA Analysis (RIGID-DNA). Our method led to a 3- and 100-fold improvement of the assays' clinical and analytical sensitivity, respectively. Our approach can enhance the hybridization efficiency of particle-based assays without altering existing assay workflows. This approach can be adapted to other platforms and surfaces to enhance the detection sensitivity.

Computational Design of Potent and Selective d-Peptide Agonists of the Glucagon-like Peptide-2 Receptor.

Here, we designed three d-GLP-2 agonists that activated the glucagon-like peptide-2 receptor (GLP-2R) cyclic adenosine monophosphate (cAMP) accumulation without stimulating the glucagon-like peptide-1 receptor (GLP-1R). All the d-GLP-2 agonists increased the protein kinase B phosphorylated (p-AKT) expression levels in a time- and concentration-dependent manner in vitro. The most effective d-GLP-2 analogue boosted the AKT phosphorylation 2.28 times more effectively compared to the native l-GLP-2. The enhancement in the p-AKT levels induced by the d-GLP-2 analogues could be explained by GLP-2R's more prolonged activation, given that the d-GLP-2 analogues induce a lower ß-arrestin recruitment. The higher stability to protease degradation of our d-GLP-2 agonists helps us envision their potential applications in enhancing intestinal absorption and treating inflammatory bowel illness while lowering the high dosage required by the current treatments.

GDockScore: a graph-based protein-protein docking scoring function.

Summary: Protein complexes play vital roles in a variety of biological processes, such as mediating biochemical reactions, the immune response and cell signalling, with 3D structure specifying function. Computational docking methods provide a means to determine the interface between two complexed polypeptide chains without using time-consuming experimental techniques. The docking process requires the optimal solution to be selected with a scoring function. Here, we propose a novel graph-based deep learning model that utilizes mathematical graph representations of proteins to learn a scoring function (GDockScore). GDockScore was pre-trained on docking outputs generated with the Protein Data Bank biounits and the RosettaDock protocol, and then fine-tuned on HADDOCK decoys generated on the ZDOCK Protein Docking Benchmark. GDockScore performs similarly to the Rosetta scoring function on docking decoys generated using the RosettaDock protocol. Furthermore, state-of-the-art is achieved on the CAPRI score set, a challenging dataset for developing docking scoring functions. Availability and implementation: The model implementation is available at https://gitlab.com/mcfeemat/gdockscore. Supplementary information: Supplementary data are available at Bioinformatics Advances online.

Disrupting the α-synuclein-ESCRT interaction with a peptide inhibitor mitigates neurodegeneration in preclinical models of Parkinson's disease.

Accumulation of α-synuclein into toxic oligomers or fibrils is implicated in dopaminergic neurodegeneration in Parkinson's disease. Here we performed a high-throughput, proteome-wide peptide screen to identify protein-protein interaction inhibitors that reduce α-synuclein oligomer levels and their associated cytotoxicity. We find that the most potent peptide inhibitor disrupts the direct interaction between the C-terminal region of α-synuclein and CHarged Multivesicular body Protein 2B (CHMP2B), a component of the Endosomal Sorting Complex Required for Transport-III (ESCRT-III). We show that α-synuclein impedes endolysosomal activity via this interaction, thereby inhibiting its own degradation. Conversely, the peptide inhibitor restores endolysosomal function and thereby decreases α-synuclein levels in multiple models, including female and male human cells harboring disease-causing α-synuclein mutations. Furthermore, the peptide inhibitor protects dopaminergic neurons from α-synuclein-mediated degeneration in hermaphroditic C. elegans and preclinical Parkinson's disease models using female rats. Thus, the α-synuclein-CHMP2B interaction is a potential therapeutic target for neurodegenerative disorders.

Gate-based quantum computing for protein design.

Protein design is a technique to engineer proteins by permuting amino acids in the sequence to obtain novel functionalities. However, exploring all possible combinations of amino acids is generally impossible due to the exponential growth of possibilities with the number of designable sites. The present work introduces circuits implementing a pure quantum approach, Grover's algorithm, to solve protein design problems. Our algorithms can adjust to implement any custom pair-wise energy tables and protein structure models. Moreover, the algorithm's oracle is designed to consist of only adder functions. Quantum computer simulators validate the practicality of our circuits, containing up to 234 qubits. However, a smaller circuit is implemented on real quantum devices. Our results show that using [Formula: see text] iterations, the circuits find the correct results among all N possibilities, providing the expected quadratic speed up of Grover's algorithm over classical methods (i.e., [Formula: see text]).

HelixGAN a deep-learning methodology for conditional de novo design of α-helix structures.

MOTIVATION: Protein and peptide engineering has become an essential field in biomedicine with therapeutics, diagnostics and synthetic biology applications. Helices are both abundant structural feature in proteins and comprise a major portion of bioactive peptides. Precise design of helices for binding or biological activity is still a challenging problem. RESULTS: Here, we present HelixGAN, the first generative adversarial network method to generate de novo left-handed and right-handed alpha-helix structures from scratch at an atomic level. We developed a gradient-based search approach in latent space to optimize the generation of novel α-helical structures by matching the exact conformations of selected hotspot residues. The designed α-helical structures can bind specific targets or activate cellular receptors. There is a significant agreement between the helix structures generated with HelixGAN and PEP-FOLD, a well-known de novo approach for predicting peptide structures from amino acid sequences. HelixGAN outperformed RosettaDesign, and our previously developed structural similarity method to generate D-peptides matching a set of given hotspots in a known L-peptide. As proof of concept, we designed a novel D-GLP1_1 analog that matches the conformations of critical hotspots for the GLP1 function. MD simulations revealed a stable binding mode of the D-GLP1_1 analog coupled to the GLP1 receptor. This novel D-peptide analog is more stable than our previous D-GLP1 design along the MD simulations. We envision HelixGAN as a critical tool for designing novel bioactive peptides with specific properties in the early stages of drug discovery. AVAILABILITY AND IMPLEMENTATION: https://github.com/xxiexuezhi/helix_gan. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Computational and artificial intelligence-based methods for antibody development.

Due to their high target specificity and binding affinity, therapeutic antibodies are currently the largest class of biotherapeutics. The traditional largely empirical antibody development process is, while mature and robust, cumbersome and has significant limitations. Substantial recent advances in computational and artificial intelligence (AI) technologies are now starting to overcome many of these limitations and are increasingly integrated into development pipelines. Here, we provide an overview of AI methods relevant for antibody development, including databases, computational predictors of antibody properties and structure, and computational antibody design methods with an emphasis on machine learning (ML) models, and the design of complementarity-determining region (CDR) loops, antibody structural components critical for binding.

A universal deep-learning model for zinc finger design enables transcription factor reprogramming.

Cys2His2 zinc finger (ZF) domains engineered to bind specific target sequences in the genome provide an effective strategy for programmable regulation of gene expression, with many potential therapeutic applications. However, the structurally intricate engagement of ZF domains with DNA has made their design challenging. Here we describe the screening of 49 billion protein-DNA interactions and the development of a deep-learning model, ZFDesign, that solves ZF design for any genomic target. ZFDesign is a modern machine learning method that models global and target-specific differences induced by a range of library environments and specifically takes into account compatibility of neighboring fingers using a novel hierarchical transformer architecture. We demonstrate the versatility of designed ZFs as nucleases as well as activators and repressors by seamless reprogramming of human transcription factors. These factors could be used to upregulate an allele of haploinsufficiency, downregulate a gain-of-function mutation or test the consequence of regulation of a single gene as opposed to the many genes that a transcription factor would normally influence.

Antibodies as drugs-a Keystone Symposia report.

Therapeutic antibodies have broad indications across diverse disease states, such as oncology, autoimmune diseases, and infectious diseases. New research continues to identify antibodies with therapeutic potential as well as methods to improve upon endogenous antibodies and to design antibodies de novo. On April 27-30, 2022, experts in antibody research across academia and industry met for the Keystone symposium "Antibodies as Drugs" to present the state-of-the-art in antibody therapeutics, repertoires and deep learning, bispecific antibodies, and engineering.

Score-based generative modeling for de novo protein design.

The generation of de novo protein structures with predefined functions and properties remains a challenging problem in protein design. Diffusion models, also known as score-based generative models (SGMs), have recently exhibited astounding empirical performance in image synthesis. Here we use image-based representations of protein structure to develop ProteinSGM, a score-based generative model that produces realistic de novo proteins. Through unconditional generation, we show that ProteinSGM can generate native-like protein structures, surpassing the performance of previously reported generative models. We experimentally validate some de novo designs and observe secondary structure compositions consistent with generated backbones. Finally, we apply conditional generation to de novo protein design by formulating it as an image inpainting problem, allowing precise and modular design of protein structure.

Targeting the Receptor-Binding Motif of SARS-CoV-2 with D-Peptides Mimicking the ACE2 Binding Helix: Lessons for Inhibiting Omicron and Future Variants of Concern.

The COVID-19 pandemic continues to spread around the world, with several new variants emerging, particularly those of concern (VOCs). Omicron (B.1.1.529), a recent VOC with many mutations in the spike protein's receptor-binding domain (RBD), has attracted a great deal of scientific and public interest. We previously developed two D-peptide inhibitors for the infection of the original SARS-CoV-2 and its VOCs, alpha and beta, in vitro. Here, we demonstrated that Covid3 and Covid_extended_1 maintained their high-affinity binding (29.4-31.3 nM) to the omicron RBD. Both D-peptides blocked the omicron variant in vitro infection with IC50s of 3.13 and 5.56 µM, respectively. We predicted that Covid3 shares a larger overlapping binding region with the ACE2 binding motif than different classes of neutralizing monoclonal antibodies. We envisioned the design of D-peptide inhibitors targeting the receptor-binding motif as the most promising approach for inhibiting current and future VOCs of SARS-CoV-2, given that the ACE2 binding interface is more limited to tolerate mutations than most of the RBD's surface.

PepNN: a deep attention model for the identification of peptide binding sites.

Protein-peptide interactions play a fundamental role in many cellular processes, but remain underexplored experimentally and difficult to model computationally. Here, we present PepNN-Struct and PepNN-Seq, structure and sequence-based approaches for the prediction of peptide binding sites on a protein. A main difficulty for the prediction of peptide-protein interactions is the flexibility of peptides and their tendency to undergo conformational changes upon binding. Motivated by this, we developed reciprocal attention to simultaneously update the encodings of peptide and protein residues while enforcing symmetry, allowing for information flow between the two inputs. PepNN integrates this module with modern graph neural network layers and a series of transfer learning steps are used during training to compensate for the scarcity of peptide-protein complex information. We show that PepNN-Struct achieves consistently high performance across different benchmark datasets. We also show that PepNN makes reasonable peptide-agnostic predictions, allowing for the identification of novel peptide binding proteins.

Phage display identification of nanomolar ligands for human NEDD4-WW3: Energetic and dynamic implications for the development of broad-spectrum antivirals.

The recognition of PPxY viral Late domains by the third WW domain of the human HECT-E3 ubiquitin ligase NEDD4 (NEDD4-WW3) is essential for the budding of many viruses. Blocking these interactions is a promising strategy to develop broad-spectrum antivirals. As all WW domains, NEDD4-WW3 is a challenging therapeutic target due to the low binding affinity of its natural interactions, its high conformational plasticity, and its complex thermodynamic behavior. In this work, we set out to investigate whether high affinity can be achieved for monovalent ligands binding to the isolated NEDD4-WW3 domain. We show that a competitive phage-display set-up allows for the identification of high-affinity peptides showing inhibitory activity of viral budding. A detailed biophysical study combining calorimetry, nuclear magnetic resonance, and molecular dynamic simulations reveals that the improvement in binding affinity does not arise from the establishment of new interactions with the domain, but is associated to conformational restrictions imposed by a novel C-terminal -LFP motif in the ligand, unprecedented in the PPxY interactome. These results, which highlight the complexity of WW domain interactions, provide valuable insight into the key elements for high binding affinity, of interest to guide virtual screening campaigns for the identification of novel therapeutics targeting NEDD4-WW3 interactions.

Deep generative modeling for protein design.

Deep learning approaches have produced substantial breakthroughs in fields such as image classification and natural language processing and are making rapid inroads in the area of protein design. Many generative models of proteins have been developed that encompass all known protein sequences, model specific protein families, or extrapolate the dynamics of individual proteins. Those generative models can learn protein representations that are often more informative of protein structure and function than hand-engineered features. Furthermore, they can be used to quickly propose millions of novel proteins that resemble the native counterparts in terms of expression level, stability, or other attributes. The protein design process can further be guided by discriminative oracles to select candidates with the highest probability of having the desired properties. In this review, we discuss five classes of generative models that have been most successful at modeling proteins and provide a framework for model guided protein design.

Computational Design of Potent D-Peptide Inhibitors of SARS-CoV-2.

Blocking the association between the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spike protein receptor-binding domain (RBD) and the human angiotensin-converting enzyme 2 (ACE2) is an attractive therapeutic approach to prevent the virus from entering human cells. While antibodies and other modalities have been developed to this end, d-amino acid peptides offer unique advantages, including serum stability, low immunogenicity, and low cost of production. Here, we designed potent novel D-peptide inhibitors that mimic the ACE2 α1-binding helix by searching a mirror-image version of the PDB. The two best designs bound the RBD with affinities of 29 and 31 nM and blocked the infection of Vero cells by SARS-CoV-2 with IC50 values of 5.76 and 6.56 µM, respectively. Notably, both D-peptides neutralized with a similar potency the infection of two variants of concern: B.1.1.7 and B.1.351 in vitro. These potent D-peptide inhibitors are promising lead candidates for developing SARS-CoV-2 prophylactic or therapeutic treatments.

Computational generation of proteins with predetermined three-dimensional shapes using ProteinSolver.

Computational generation of new proteins with a predetermined three-dimensional shape and computational optimization of existing proteins while maintaining their shape are challenging problems in structural biology. Here, we present a protocol that uses ProteinSolver, a pre-trained graph convolutional neural network, to quickly generate thousands of sequences matching a specific protein topology. We describe computational approaches that can be used to evaluate the generated sequences, and we show how select sequences can be validated experimentally. For complete details on the use and execution of this protocol, please refer to Strokach et al. (2020).

ELASPIC2 (EL2): Combining Contextualized Language Models and Graph Neural Networks to Predict Effects of Mutations.

The ELASPIC web server allows users to evaluate the effect of mutations on protein folding and protein-protein interaction on a proteome-wide scale. It uses homology models of proteins and protein-protein interactions, which have been precalculated for several proteomes, and machine learning models, which integrate structural information with sequence conservation scores, in order to make its predictions. Since the original publication of the ELASPIC web server, several advances have motivated a revisiting of the problem of mutation effect prediction. First, progress in neural network architectures and self-supervised pre-trained has resulted in models which provide more informative embeddings of protein sequence and structure than those used by the original version of ELASPIC. Second, the amount of training data has increased several-fold, largely driven by advances in deep mutation scanning and other multiplexed assays of variant effect. Here, we describe two machine learning models which leverage the recent advances in order to achieve superior accuracy in predicting the effect of mutation on protein folding and protein-protein interaction. The models incorporate features generated using pre-trained transformer- and graph convolution-based neural networks, and are trained to optimize a ranking objective function, which permits the use of heterogeneous training data. The outputs from the new models have been incorporated into the ELASPIC web server, available at http://elaspic.kimlab.org.