Unseen Proteins Research Articles

From protein structure to an optimized chromatographic capture step using multiscale modeling.

Optimizing a biopharmaceutical chromatographic purification process is currently the greatest challenge during process development. A lack of process understanding calls for extensive experimental efforts in pursuit of an optimal process. In silico techniques, such as mechanistic or data driven modeling, enhance the understanding, allowing more cost-effective and time efficient process optimization. This work presents a modeling strategy integrating quantitative structure property relationship (QSPR) models and chromatographic mechanistic models (MM) to optimize a cation exchange (CEX) capture step, limiting experiments. In QSPR, structural characteristics obtained from the protein structure are used to describe physicochemical behavior. This QSPR information can be applied in MM to predict the chromatogram and optimize the entire process. To validate this approach, retention profiles of six proteins were determined experimentally from mixtures, at different pH (3.5, 4.3, 5.0, and 7.0). Four proteins at different pH's were used to train QSPR models predicting the retention volumes and characteristic charge, subsequently the equilibrium constant was determined. For an unseen protein knowing only the protein structure, the retention peak difference between the modeled and experimental peaks was 0.2% relative to the gradient length (60 column volume). Next, the CEX capture step was optimized, demonstrating a consistent result in both the experimental and QSPR-based methods. The impact of model parameter confidence on the final optimization revealed two viable process conditions, one of which is similar to the optimization achieved using experimentally obtained parameters. The multiscale modeling approach reduces the required experimental effort by identification of initial process conditions, which can be optimized.

Effectiveness and Efficiency: Label-Aware Hierarchical Subgraph Learning for Protein–Protein Interaction

The study of protein–protein interactions (PPIs) holds immense significance in understanding various biological activities, as well as in drug discovery and disease diagnosis. Existing deep learning methods for PPI prediction, including graph neural networks (GNNs), have been widely employed as the solutions, while they often experience a decline in performance in the real world. We claim that the topological shortcut is one of the key problems contributing negatively to the performance, according to our analysis. By modeling the PPIs as a graph with protein as nodes and interactions as edge types, the prevailing models tend to learn the pattern of nodes’ degrees rather than intrinsic sequence-structure profiles, leading to the problem termed topological shortcut. The huge data growth of PPI leads to intensive computational costs and challenges computing devices, causing infeasibility in practice. To address the discussed problems, we propose a label-aware hierarchical subgraph learning method (laruGL-PPI) that can effectively infer PPIs while being interpretable. Specifically, we introduced edge-based subgraph sampling to effectively alleviate the problems of topological shortcuts and high computing costs. Besides, the inner-outer connections of PPIs are modeled as a hierarchical graph, together with the dependencies between interaction types constructed by a label graph. Extensive experiments conducted across various scales of PPI datasets have conclusively demonstrated that the laruGL-PPI method surpasses the most advanced PPI prediction techniques currently available, particularly in the testing of unseen proteins. Also, our model can recognize crucial sites of proteins, such as surface sites for binding and active sites for catalysis.

Prediction of protein subcellular localization in single cells.

The subcellular localization of a protein is important for its function and interaction with other molecules, and its mislocalization is linked to numerous diseases. While atlas-scale efforts have been made to profile protein localization across various cell lines, existing datasets only contain limited pairs of proteins and cell lines which do not cover all human proteins. We present a method that uses both protein sequences and cellular landmark images to perform Predictions of Unseen Proteins' Subcellular localization (PUPS), which can generalize to both proteins and cell lines not used for model training. PUPS combines a protein language model and an image inpainting model to utilize both protein sequence and cellular images for protein localization prediction. The protein sequence input enables generalization to unseen proteins and the cellular image input enables cell type specific prediction that captures single-cell variability. PUPS' ability to generalize to unseen proteins and cell lines enables us to assess the variability in protein localization across cell lines as well as across single cells within a cell line and to identify the biological processes associated with the proteins that have variable localization. Experimental validation shows that PUPS can be used to predict protein localization in newly performed experiments outside of the Human Protein Atlas used for training. Collectively, PUPS utilizes both protein sequences and cellular images to predict protein localization in unseen proteins and cell lines with the ability to capture single-cell variability.

DynamicBind: predicting ligand-specific protein-ligand complex structure with a deep equivariant generative model

While significant advances have been made in predicting static protein structures, the inherent dynamics of proteins, modulated by ligands, are crucial for understanding protein function and facilitating drug discovery. Traditional docking methods, frequently used in studying protein-ligand interactions, typically treat proteins as rigid. While molecular dynamics simulations can propose appropriate protein conformations, they’re computationally demanding due to rare transitions between biologically relevant equilibrium states. In this study, we present DynamicBind, a deep learning method that employs equivariant geometric diffusion networks to construct a smooth energy landscape, promoting efficient transitions between different equilibrium states. DynamicBind accurately recovers ligand-specific conformations from unbound protein structures without the need for holo-structures or extensive sampling. Remarkably, it demonstrates state-of-the-art performance in docking and virtual screening benchmarks. Our experiments reveal that DynamicBind can accommodate a wide range of large protein conformational changes and identify cryptic pockets in unseen protein targets. As a result, DynamicBind shows potential in accelerating the development of small molecules for previously undruggable targets and expanding the horizons of computational drug discovery.

HyperPCM: Robust Task-Conditioned Modeling of Drug-Target Interactions.

A central problem in drug discovery is to identify the interactions between drug-like compounds and protein targets. Over the past few decades, various quantitative structure-activity relationship (QSAR) and proteo-chemometric (PCM) approaches have been developed to model and predict these interactions. While QSAR approaches solely utilize representations of the drug compound, PCM methods incorporate both representations of the protein target and the drug compound, enabling them to achieve above-chance predictive accuracy on previously unseen protein targets. Both QSAR and PCM approaches have recently been improved by machine learning and deep neural networks, that allow the development of drug-target interaction prediction models from measurement data. However, deep neural networks typically require large amounts of training data and cannot robustly adapt to new tasks, such as predicting interaction for unseen protein targets at inference time. In this work, we propose to use HyperNetworks to efficiently transfer information between tasks during inference and thus to accurately predict drug-target interactions on unseen protein targets. Our HyperPCM method reaches state-of-the-art performance compared to previous methods on multiple well-known benchmarks, including Davis, DUD-E, and a ChEMBL derived data set, and particularly excels at zero-shot inference involving unseen protein targets. Our method, as well as reproducible data preparation, is available at https://github.com/ml-jku/hyper-dti.

PILOT: equivariant diffusion for pocket-conditioned de novo ligand generation with multi-objective guidance via importance sampling.

The generation of ligands that both are tailored to a given protein pocket and exhibit a range of desired chemical properties is a major challenge in structure-based drug design. Here, we propose an in silico approach for the de novo generation of 3D ligand structures using the equivariant diffusion model PILOT, combining pocket conditioning with a large-scale pre-training and property guidance. Its multi-objective trajectory-based importance sampling strategy is designed to direct the model towards molecules that not only exhibit desired characteristics such as increased binding affinity for a given protein pocket but also maintains high synthetic accessibility. This ensures the practicality of sampled molecules, thus maximizing their potential for the drug discovery pipeline. PILOT significantly outperforms existing methods across various metrics on the common benchmark dataset CrossDocked2020. Moreover, we employ PILOT to generate novel ligands for unseen protein pockets from the Kinodata-3D dataset, which encompasses a substantial portion of the human kinome. The generated structures exhibit predicted IC50 values indicative of potent biological activity, which highlights the potential of PILOT as a powerful tool for structure-based drug design.

ZeroBind: a protein-specific zero-shot predictor with subgraph matching for drug-target interactions

Existing drug-target interaction (DTI) prediction methods generally fail to generalize well to novel (unseen) proteins and drugs. In this study, we propose a protein-specific meta-learning framework ZeroBind with subgraph matching for predicting protein-drug interactions from their structures. During the meta-training process, ZeroBind formulates training a protein-specific model, which is also considered a learning task, and each task uses graph neural networks (GNNs) to learn the protein graph embedding and the molecular graph embedding. Inspired by the fact that molecules bind to a binding pocket in proteins instead of the whole protein, ZeroBind introduces a weakly supervised subgraph information bottleneck (SIB) module to recognize the maximally informative and compressive subgraphs in protein graphs as potential binding pockets. In addition, ZeroBind trains the models of individual proteins as multiple tasks, whose importance is automatically learned with a task adaptive self-attention module to make final predictions. The results show that ZeroBind achieves superior performance on DTI prediction over existing methods, especially for those unseen proteins and drugs, and performs well after fine-tuning for those proteins or drugs with a few known binding partners.

Cross-protein transfer learning substantially improves disease variant prediction

BackgroundGenetic variation in the human genome is a major determinant of individual disease risk, but the vast majority of missense variants have unknown etiological effects. Here, we present a robust learning framework for leveraging saturation mutagenesis experiments to construct accurate computational predictors of proteome-wide missense variant pathogenicity.ResultsWe train cross-protein transfer (CPT) models using deep mutational scanning (DMS) data from only five proteins and achieve state-of-the-art performance on clinical variant interpretation for unseen proteins across the human proteome. We also improve predictive accuracy on DMS data from held-out proteins. High sensitivity is crucial for clinical applications and our model CPT-1 particularly excels in this regime. For instance, at 95% sensitivity of detecting human disease variants annotated in ClinVar, CPT-1 improves specificity to 68%, from 27% for ESM-1v and 55% for EVE. Furthermore, for genes not used to train REVEL, a supervised method widely used by clinicians, we show that CPT-1 compares favorably with REVEL. Our framework combines predictive features derived from general protein sequence models, vertebrate sequence alignments, and AlphaFold structures, and it is adaptable to the future inclusion of other sources of information. We find that vertebrate alignments, albeit rather shallow with only 100 genomes, provide a strong signal for variant pathogenicity prediction that is complementary to recent deep learning-based models trained on massive amounts of protein sequence data. We release predictions for all possible missense variants in 90% of human genes.ConclusionsOur results demonstrate the utility of mutational scanning data for learning properties of variants that transfer to unseen proteins.

SAM-DTA: a sequence-agnostic model for drug-target binding affinity prediction.

Drug-target binding affinity prediction is a fundamental task for drug discovery and has been studied for decades. Most methods follow the canonical paradigm that processes the inputs of the protein (target) and the ligand (drug) separately and then combines them together. In this study we demonstrate, surprisingly, that a model is able to achieve even superior performance without access to any protein-sequence-related information. Instead, a protein is characterized completely by the ligands that it interacts. Specifically, we treat different proteins separately, which are jointly trained in a multi-head manner, so as to learn a robust and universal representation of ligands that is generalizable across proteins. Empirical evidences show that the novel paradigm outperforms its competitive sequence-based counterpart, with the Mean Squared Error (MSE) of 0.4261 versus 0.7612 and the R-Square of 0.7984 versus 0.6570 compared with DeepAffinity. We also investigate the transfer learning scenario where unseen proteins are encountered after the initial training, and the cross-dataset evaluation for prospective studies. The results reveals the robustness of the proposed model in generalizing to unseen proteins as well as in predicting future data. Source codes and data are available at https://github.com/huzqatpku/SAM-DTA.

Cross-modality and self-supervised protein embedding for compound-protein affinity and contact prediction.

Computational methods for compound-protein affinity and contact (CPAC) prediction aim at facilitating rational drug discovery by simultaneous prediction of the strength and the pattern of compound-protein interactions. Although the desired outputs are highly structure-dependent, the lack of protein structures often makes structure-free methods rely on protein sequence inputs alone. The scarcity of compound-protein pairs with affinity and contact labels further limits the accuracy and the generalizability of CPAC models. To overcome the aforementioned challenges of structure naivety and labeled-data scarcity, we introduce cross-modality and self-supervised learning, respectively, for structure-aware and task-relevant protein embedding. Specifically, protein data are available in both modalities of 1D amino-acid sequences and predicted 2D contact maps that are separately embedded with recurrent and graph neural networks, respectively, as well as jointly embedded with two cross-modality schemes. Furthermore, both protein modalities are pre-trained under various self-supervised learning strategies, by leveraging massive amount of unlabeled protein data. Our results indicate that individual protein modalities differ in their strengths of predicting affinities or contacts. Proper cross-modality protein embedding combined with self-supervised learning improves model generalizability when predicting both affinities and contacts for unseen proteins. Data and source codes are available at https://github.com/Shen-Lab/CPAC. Supplementary data are available at Bioinformatics online.

RAPPPID: towards generalizable protein interaction prediction with AWD-LSTM twin networks.

Computational methods for the prediction of protein-protein interactions (PPIs), while important tools for researchers, are plagued by challenges in generalizing to unseen proteins. Datasets used for modelling protein-protein predictions are particularly predisposed to information leakage and sampling biases. In this study, we introduce RAPPPID, a method for the Regularized Automatic Prediction of Protein-Protein Interactions using Deep Learning. RAPPPID is a twin Averaged Weight-Dropped Long Short-Term memory network which employs multiple regularization methods during training time to learn generalized weights. Testing on stringent interaction datasets composed of proteins not seen during training, RAPPPID outperforms state-of-the-art methods. Further experiments show that RAPPPID's performance holds regardless of the particular proteins in the testing set and its performance is higher for experimentally supported edges. This study serves to demonstrate that appropriate regularization is an important component of overcoming the challenges of creating models for PPI prediction that generalize to unseen proteins. Additionally, as part of this study, we provide datasets corresponding to several data splits of various strictness, in order to facilitate assessment of PPI reconstruction methods by others in the future. Code and datasets are freely available at https://github.com/jszym/rapppid and Zenodo.org. Supplementary data are available at Bioinformatics online.

ScanNet: an interpretable geometric deep learning model for structure-based protein binding site prediction.

Predicting the functional sites of a protein from its structure, such as the binding sites of small molecules, other proteins or antibodies, sheds light on its function in vivo. Currently, two classes of methods prevail: machine learning models built on top of handcrafted features and comparative modeling. They are, respectively, limited by the expressivity of the handcrafted features and the availability of similar proteins. Here, we introduce ScanNet, an end-to-end, interpretable geometric deep learning model that learns features directly from 3D structures. ScanNet builds representations of atoms and amino acids based on the spatio-chemical arrangement of their neighbors. We train ScanNet for detecting protein-protein and protein-antibody binding sites, demonstrate its accuracy-including for unseen protein folds-and interpret the filters learned. Finally, we predict epitopes of the SARS-CoV-2 spike protein, validating known antigenic regions and predicting previously uncharacterized ones. Overall, ScanNet is a versatile, powerful and interpretable model suitable for functional site prediction tasks. A webserver for ScanNet is available from http://bioinfo3d.cs.tau.ac.il/ScanNet/ .

Protein design via deep learning.

Proteins with desired functions and properties are important in fields like nanotechnology and biomedicine. De novo protein design enables the production of previously unseen proteins from the ground up and is believed as a key point for handling real social challenges. Recent introduction of deep learning into design methods exhibits a transformative influence and is expected to represent a promising and exciting future direction. In this review, we retrospect the major aspects of current advances in deep-learning-based design procedures and illustrate their novelty in comparison with conventional knowledge-based approaches through noticeable cases. We not only describe deep learning developments in structure-based protein design and direct sequence design, but also highlight recent applications of deep reinforcement learning in protein design. The future perspectives on design goals, challenges and opportunities are also comprehensively discussed.

An Approach for Predicting Protein-Protein Interactions using Supervised Autoencoders

Identifying protein-protein interactions (PPIs) represents a challenging research problem in computational biology. Even though machine learning methods have significantly advanced the research in this field, current approaches struggle to accurately predict interactions for previously unseen proteins. Supervised autoencoders, which are neural networks trained to simultaneously predict labels and reconstruct their inputs, have been proved to offer generalization guarantees in the linear case. Moreover, the addition of a reconstruction branch was empirically shown to improve the performance of standard neural networks classifiers on several tasks. In this paper, we introduce a two-stage sequence-based PPI prediction method based on supervised autoencoders. The proposed approach consists of initially training a denoising autoencoder on protein sequences, followed by a supervised training stage in which the model learns to both predict whether two proteins interact and to reconstruct the two proteins in the pair. An experimental analysis was performed on two public PPI data sets containing testing pairs formed using both seen and unseen protein sequences. The results show that our approach surpasses, on the two data sets, multiple machine learning classifiers proposed in the literature.

GNINA 1.0: molecular docking with deep learning

Molecular docking computationally predicts the conformation of a small molecule when binding to a receptor. Scoring functions are a vital piece of any molecular docking pipeline as they determine the fitness of sampled poses. Here we describe and evaluate the 1.0 release of the Gnina docking software, which utilizes an ensemble of convolutional neural networks (CNNs) as a scoring function. We also explore an array of parameter values for Gnina 1.0 to optimize docking performance and computational cost. Docking performance, as evaluated by the percentage of targets where the top pose is better than 2Å root mean square deviation (Top1), is compared to AutoDock Vina scoring when utilizing explicitly defined binding pockets or whole protein docking. Gnina, utilizing a CNN scoring function to rescore the output poses, outperforms AutoDock Vina scoring on redocking and cross-docking tasks when the binding pocket is defined (Top1 increases from 58% to 73% and from 27% to 37%, respectively) and when the whole protein defines the binding pocket (Top1 increases from 31% to 38% and from 12% to 16%, respectively). The derived ensemble of CNNs generalizes to unseen proteins and ligands and produces scores that correlate well with the root mean square deviation to the known binding pose. We provide the 1.0 version of Gnina under an open source license for use as a molecular docking tool at https://github.com/gnina/gnina.

Data-driven molecular design for discovery and synthesis of novel ligands: a case study on SARS-CoV-2

Bridging systems biology and drug design, we propose a deep learning framework for de novo discovery of molecules tailored to bind with given protein targets. Our methodology is exemplified by the task of designing antiviral candidates to target SARS-CoV-2 related proteins. Crucially, our framework does not require fine-tuning for specific proteins but is demonstrated to generalize in proposing ligands with high predicted binding affinities against unseen targets. Coupling our framework with the automatic retrosynthesis prediction of IBM RXN for Chemistry, we demonstrate the feasibility of swift chemical synthesis of molecules with potential antiviral properties that were designed against a specific protein target. In particular, we synthesize an antiviral candidate designed against the host protein angiotensin converting enzyme 2 (ACE2); a surface receptor on human respiratory epithelial cells that facilitates SARS-CoV-2 cell entry through its spike glycoprotein.This is achieved as follows. First, we train a multimodal ligand–protein binding affinity model on predicting affinities of bioactive compounds to target proteins and couple this model with pharmacological toxicity predictors. Exploiting this multi-objective as a reward function of a conditional molecular generator that consists of two variational autoencoders (VAE), our framework steers the generation toward regions of the chemical space with high-reward molecules. Specifically, we explore a challenging setting of generating ligands against unseen protein targets by performing a leave-one-out-cross-validation on 41 SARS-CoV-2-related target proteins. Using deep reinforcement learning, it is demonstrated that in 35 out of 41 cases, the generation is biased towards sampling binding ligands, with an average increase of 83% comparing to an unbiased VAE. The generated molecules exhibit favorable properties in terms of target binding affinity, selectivity and drug-likeness. We use molecular retrosynthetic models to provide a synthetic accessibility assessment of the best generated hit molecules. Finally, with this end-to-end framework, we synthesize 3-Bromobenzylamine, a potential inhibitor of the host ACE2 protein, solely based on the recommendations of a molecular retrosynthesis model and a synthesis protocol prediction model. We hope that our framework can contribute towards swift discovery of de novo molecules with desired pharmacological properties.

Capturing Protein Domain Structure and Function Using Self-Supervision on Domain Architectures

Predicting biological properties of unseen proteins is shown to be improved by the use of protein sequence embeddings. However, these sequence embeddings have the caveat that biological metadata do not exist for each amino acid, in order to measure the quality of each unique learned embedding vector separately. Therefore, current sequence embedding cannot be intrinsically evaluated on the degree of their captured biological information in a quantitative manner. We address this drawback by our approach, dom2vec, by learning vector representation for protein domains and not for each amino acid base, as biological metadata do exist for each domain separately. To perform a reliable quantitative intrinsic evaluation in terms of biology knowledge, we selected the metadata related to the most distinctive biological characteristics of a domain, which are its structure, enzymatic, and molecular function. Notably, dom2vec obtains an adequate level of performance in the intrinsic assessment—therefore, we can draw an analogy between the local linguistic features in natural languages and the domain structure and function information in domain architectures. Moreover, we demonstrate the dom2vec applicability on protein prediction tasks, by comparing it with state-of-the-art sequence embeddings in three downstream tasks. We show that dom2vec outperforms sequence embeddings for toxin and enzymatic function prediction and is comparable with sequence embeddings in cellular location prediction.

Towards learning transferable embeddings for protein conformations using Variational Autoencoders

The analysis of the transitions that a protein conformation can undergo is an important step towards achieving a thorough understanding of protein functions. In this context, learning informative and transferable low-dimensional representations of proteins conformations is of great importance for developing robust computational approaches that can model proteins dynamical systems. In this paper we perform an experimental evaluation of the ability of the latent space representations learned by a Variational Autoencoder (VAE) trained on trajectories of proteins belonging to a superfamily to characterize an unseen protein from the same superfamily. In our study, we compare the latent space encodings obtained using two representations for the protein conformations and evaluate the impact of the latent space dimensionality on the embeddings’ quality. The obtained results outline that the VAE’s latent space can capture structural information characterizing a certain protein superfamily and, consequently, generalize to a different testing protein from that superfamily.

Prediction of Protein–Protein Interaction via co-occurring Aligned Pattern Clusters

Predicting Protein–Protein Interaction (PPI) is important for making new discoveries in the molecular mechanisms inside a cell. Traditionally, new PPIs are identified through biochemical experiments but such methods are labor-intensive, expensive, time-consuming and technically ineffective due to high false positive rates. Sequence-based prediction is currently the most readily applicable and cost-effective method. It exploits known PPI Databases to construct classifiers for predicting unknown PPIs based only on sequence data without requiring any other prior knowledge. Among existing sequence-based methods, most feature-based methods use exact sequence patterns with fixed length as features — a constraint which is biologically unrealistic. SVM with Pairwise String Kernel renders better predicting performance. However it is difficult to be biologically interpretable since it is kernel-based where no concrete feature values are computed. Here we have developed a novel method WeMine-P2P to overcome these drawbacks. By assuming that the regions/sites that mediate PPI are more conserved, WeMine-P2P first discovers/locates the conserved sequence patterns in protein sequences in the form of Aligned Pattern Clusters (APCs), allowing pattern variations with variable length. It then pairs up all APCs into a set of Co-Occurring APC (cAPC) pairs, and computes a cAPC-PPI score for each cAPC pair on all PPI pairs. It further constructs a feature vector composed of all cAPC pairs with their cAPC-PPI scores for each PPI pair and uses them for constructing a PPI predictor. Through 40 independent experiments, we showed that (1) WeMine-P2P outperforms the well-known algorithm, PIPE2, which also utilizes co-occurring amino acid sequence segments but does not allow variable lengths and pattern variations; (2) WeMine-P2P achieves satisfactory PPI prediction performance, comparable to the SVM-based methods particularly among unseen protein sequences with a potential reduction of feature dimension of 1280×; (3) Unlike SVM-based methods, WeMine-P2P renders interpretable biological features from which we observed that co-occurring sequence patterns from the compositional bias regions are more discriminative. WeMine-P2P is extendable to predict other biosequence interactions such as Protein–DNA interactions.

PepPSy: a web server to prioritize gene products in experimental and biocuration workflows.

Among the 20 000 human gene products predicted from genome annotation, about 3000 still lack validation at protein level. We developed PepPSy, a user-friendly gene expression-based prioritization system, to help investigators to determine in which human tissues they should look for an unseen protein. PepPSy can also be used by biocurators to revisit the annotation of specific categories of proteins based on the ‘omics’ data housed by the system. In this study, it was used to prioritize 21 dubious protein-coding genes among the 616 annotated in neXtProt for reannotation. PepPSy is freely available at http://peppsy.genouest.org.Database URL: http://peppsy.genouest.org.