Multimodal learning on heterogeneous subgraphs and LLMs representation for MHC-peptide binding affinity prediction.

In the drug discovery process and repurposing of existing drugs, accurately identifying ligands with high binding affinity to proteins is a very critical step. However, it sinks a lot of time and resources to detect the protein-ligand binding affinity through biological experiments. Therefore, it is very necessary to develop an accurate and reliable computational method to predict the binding affinity between protein and ligand. At present, some computational methods have been proposed to predict the protein-ligand binding affinity, but the absence of protein-ligand complexes structures restricts some predictive methods that require input the complexes structures. In this paper, a novel deep-learning-based method is proposed, named DPLA, to predict binding affinity by integrating multilevel information of protein and ligand. More specifically, our model extracted some important information, such as sequence representation, structural property representation of amino acids in protein and protein binding pocket, MACCS key ligand molecular fingerprint and ligand molecular network features. This method was tested on the PDBbind core set, and we compared it with some recent state-of-art protein-ligand affinity prediction methods. The excellent performance shows that DPLA is an accurate and reliable method for affinity prediction.

Many multiple alignment methods implicitly or explicitly try to minimize the amount of biological change implied by an alignment. At the level of sequences, biological change is measured along a phylogenetic tree, a structure frequently being predicted only after the multiple alignment instead of together with it. The Generalized Tree Alignment problem addresses both questions simultaneously. It can formally be viewed as a Steiner tree problem in sequence space and our approach merges a path heuristic for the construction of a Steiner tree with a clustering method as usually applied only to distance data. This combination is achieved using sequence graphs, a data structure for efficient representation of similar sequences. Although somewhat slower in practice than an earlier method by Hein (1989) the current approach achieves significantly better results in terms of the underlying scoring function. Furthermore, a variant of the algorithm is introduced that maintains a guaranteed error bound of (2 - 2/n) for n sequences.

The detection of regulatory sequences in DNA is a challenging problem, especially when considered in the context of whole genomes. The degree of sequence conservation of regulatory protein binding sites is often weak, and the sites are obscured by surrounding intergenic sequence. Since structural interactions are vital for protein-DNA interactions, structural representations of regulatory sites can yield a more accurate model and a better understanding of within-site variability. However, the use of multiple alternative representations of DNA introduces a requirement for novel algorithms that can create and test different combinations of DNA features. The Genetic Algorithm Neural Network (GANN) was designed to identify combinations of patterns that can be used to distinguish between different classes of training sequence. GANN trains a set of artificial neural networks to classify sets of sequence using either backpropagation or a genetic algorithm, and uses an ‘outer genetic algorithm’ to choose the best inputs from a pool of DNA features that can include sequence, structure, and weight matrix representations. When trained with a subset of upstream sequences from a whole genome, GANN was able to detect patterns such as the Shine-Dalgarno sequence in Escherichia coli K12, and sequences consistent with archaeal promoters in the archaeon Sulfolobus solfataricus P2. The Motif Genetic Algorithm (MGA) constructs motif representations by concatenating minimal units of DNA sequence and structure. This algorithm was used to model conserved patterns in DNA, including the binding sites for E. coli cyclic AMP activated protein (CAP), integration host factor (IHF), and two different promoter types recognized by alternative bacterial sigma factors. The CAP models were used to detect other putative binding sites in upstream regions of the E. coli K12 genome, while attempts to train an accurate model of IHF binding sites revealed an important role for structural representations in motif modeling.

Jeffrey proposed (1990) a graphic representation of DNA sequences using Barnsley's iterative function systems. In spite of further developments in this direction, the proposed graphic representation of DNA sequences has been lacking a rigorous connection between its spatial scaling characteristics and the statistical characteristics of the DNA sequences themselves. We 1) generalize Jeffrey's graphic representation to accommodate (possibly infinite) sequences over an arbitrary finite number of symbols; 2) establish a direct correspondence between the statistical characterization of symbolic sequences via Renyi entropy spectra (1959) and the multifractal characteristics (Renyi generalized dimensions) of the sequences' spatial representations; 3) show that for general symbolic dynamical systems, the multifractal f/sub H/-spectra in the sequence space coincide with the f/sub H/-spectra on spatial sequence representations.

Multimodal representations play an important role in multimodal learning tasks, including cross-modal retrieval and intra-modal clustering. However, existing multimodal representation learning approaches focus on building one common space by aligning different modalities and ignore the complementary information across the modalities, such as the intra-modal local structures. In other words, they only focus on the object-level alignment and ignore structure-level alignment. To tackle the problem, we propose a novel symmetric multimodal representation learning framework by transferring local structures across different modalities, namely MTLS. A customized soft metric learning strategy and an iterative parameter learning process are designed to symmetrically transfer local structures and enhance the cluster structures in intra-modal representations. The bidirectional retrieval loss based on multi-layer neural networks is utilized to align two modalities. MTLS is instantiated with image and text data and shows its superior performance on image-text retrieval and image clustering. MTLS outperforms the state-of-the-art multimodal learning methods by up to 32% in terms of R@1 on text-image retrieval and 16.4% in terms of AMI onclustering.

Metabolic stability plays a crucial role in the early stages of drug discovery and development. Accurately modeling and predicting molecular metabolic stability has great potential for the efficient screening of drug candidates as well as the optimization of lead compounds. Considering wet-lab experiment is time-consuming, laborious, and expensive, in silico prediction of metabolic stability is an alternative choice. However, few computational methods have been developed to address this task. In addition, it remains a significant challenge to explain key functional groups determining metabolic stability. To address these issues, we develop a novel cross-modality graph contrastive learning model named CMMS-GCL for predicting the metabolic stability of drug candidates. In our framework, we design deep learning methods to extract features for molecules from two modality data, i.e., SMILES sequence and molecule graph. In particular, for the sequence data, we design a multi-head attention BiGRU-based encoder to preserve the context of symbols to learn sequence representations of molecules. For the graph data, we propose a graph contrastive learning-based encoder to learn structure representations by effectively capturing the consistencies between local and global structures. We further exploit fully connected neural networks to combine the sequence and structure representations for model training. Extensive experimental results on two datasets demonstrate that our CMMS-GCL consistently outperforms seven state-of-the-art methods. Furthermore, a collection of case studies on sequence data and statistical analyses of the graph structure module, strengthens the validation of the interpretability of crucial functional groups recognized by CMMS-GCL. Overall, CMMS-GCL can serve as an effective and interpretable tool for predicting metabolic stability, identifying critical functional groups, and thus facilitating the drug discovery process and lead compound optimization. The code and data underlying this article are freely available at https://github.com/dubingxue/CMMS-GCL. Supplementary data are available at Bioinformatics online.

Deep learning-based predictive models, leveraging Electronic Health Records (EHR), are receiving increasing attention in healthcare. An effective representation of a patient's EHR should hierarchically encompass both the temporal relationships between historical visits and medical events, and the inherent structural information within these elements. Existing patient representation methods can be roughly categorized into sequential representation and graphical representation. The sequential representation methods focus only on the temporal relationships among longitudinal visits. On the other hand, the graphical representation approaches, while adept at extracting the graph-structured relationships between various medical events, fall short in effectively integrate temporal information. To capture both types of information, we model a patient's EHR as a novel temporal heterogeneous graph. This graph includes historical visits nodes and medical events nodes. It propagates structured information from medical event nodes to visit nodes and utilizes time-aware visit nodes to capture changes in the patient's health status. Furthermore, we introduce a novel temporal graph transformer (TRANS) that integrates temporal edge features, global positional encoding, and local structural encoding into heterogeneous graph convolution, capturing both temporal and structural information. We validate the effectiveness of TRANS through extensive experiments on three real-world datasets. The results show that our proposed approach achieves state-of-the-art performance.

Precise movement control is of prime importance in almost every kind of sport and greatly influences an athlete's performance. In dancing not only motor but also cognitive skills, e.g. in the form of memorized representational structures, are essential components of the performance. This study investigated different metrics related to the long-term memory of ballet dancers with different skill levels regarding the pas de bourrée using the structural-dimensional analysis of mental representations (SDA-M) method. To this end, the Correct Action Selection Probability Analysis (CASPA) algorithm, a recent SDA-M extension that predicts the individual probabilities of correct action selections within a movement sequence, has been applied in the context of dancing for the first time. Significant positive correlations were found between participants' degree of expertise and the proximity of their mental representation structure to the ideal reference structure, as well as between the degree of expertise and the probability of correct action selection within the movement sequence estimated by CASPA. The results indicate that increased training experience in ballet dancing is not only associated with functionally better structured mental representations of the movement sequence but also with a higher probability of correct action selection. These findings provide further evidence for SDA-M with CASPA as an auspicious tool for individualized task-related memory assessments and diagnostics in different action sequences, e.g. as the basis for mental training.

The recently released Boltz-2 cofolding model is generating high expectations by enabling both protein-ligand structure and binding affinity predictions. When applied to a recently described and challenging data set of ultralarge-virtual-screening hits, Boltz-2 excels at discriminating true from false positives, overcoming by a large margin all scoring functions tested so far on raw docking poses. Strikingly, affinity predictions seem to be relatively independent of pose quality but are not biased by obvious chemical similarity to known compounds sharing comparable binding potencies. To ascertain that Boltz-2 truly relies on the physics of intermolecular interactions, we challenged affinity predictions with biologically meaningful challenges (target mutation and target shuffling). Binary classification of active vs inactive compounds remains insensitive to key binding site mutations and even in some cases to target exchange, raising concerns on the hidden features governing Boltz-2 affinity predictions.

The extensive propagation of industrial Internet of Things (IIoT) technologies has encouraged intruders to initiate a variety of attacks that need to be identified to maintain the security of end-user data and the safety of services offered by service providers. Deep learning (DL), especially recurrent approaches, has been applied successfully to the analysis of IIoT forensics but their key challenge of recurrent DL models is that they struggle with long traffic sequences and cannot be parallelized. Multihead attention (MHA) tried to address this shortfall but failed to capture the local representation of IIoT traffic sequences. In this article, we propose a forensics-based DL model (called Deep-IFS) to identify intrusions in IIoT traffic. The model learns local representations using local gated recurrent unit (LocalGRU), and introduces an MHA layer to capture and learn global representation (i.e., long-range dependencies). A residual connection between layers is designed to prevent information loss. Another challenge facing the current IIoT forensics frameworks is their limited scalability, limiting performance in handling Big IIoT traffic data produced by IIoT devices. This challenge is addressed by deploying and training the proposed Deep-IFS in a fog computing environment. The intrusion identification becomes scalable by distributing the computation and the IIoT traffic data across worker fog nodes for training the model. The master fog node is responsible for sharing training parameters and aggregating worker node output. The aggregated classification output is subsequently passed to the cloud platform for mitigating attacks. Empirical results on the Bot-IIoT dataset demonstrate that the developed distributed Deep-IFS can effectively handle Big IIoT traffic data compared with the present centralized DL-based forensics techniques. Further, the results validate the robustness of the proposed Deep-IFS across various evaluation measures.

The rapid accumulation of molecular data motivates development of innovative approaches to computationally characterize sequences, structures and functions of biological and chemical molecules in an efficient, accessible and accurate manner. Notwithstanding several computational tools that characterize protein or nucleic acids data, there are no one-stop computational toolkits that comprehensively characterize a wide range of biomolecules. We address this vital need by developing a holistic platform that generates features from sequence and structural data for a diverse collection of molecule types. Our freely available and easy-to-use iFeatureOmega platform generates, analyzes and visualizes 189 representations for biological sequences, structures and ligands. To the best of our knowledge, iFeatureOmega provides the largest scope when directly compared to the current solutions, in terms of the number of feature extraction and analysis approaches and coverage of different molecules. We release three versions of iFeatureOmega including a webserver, command line interface and graphical interface to satisfy needs of experienced bioinformaticians and less computer-savvy biologists and biochemists. With the assistance of iFeatureOmega, users can encode their molecular data into representations that facilitate construction of predictive models and analytical studies. We highlight benefits of iFeatureOmega based on three research applications, demonstrating how it can be used to accelerate and streamline research in bioinformatics, computational biology, and cheminformatics areas. The iFeatureOmega webserver is freely available at http://ifeatureomega.erc.monash.edu and the standalone versions can be downloaded from https://github.com/Superzchen/iFeatureOmega-GUI/ and https://github.com/Superzchen/iFeatureOmega-CLI/.

PocketDTA: A pocket-based multimodal deep learning model for drug-target affinity prediction.

The electrostatically embedded generalized molecular fractionation with conjugate caps (EE-GMFCC) method has been successfully utilized for efficient linear-scaling quantum mechanical (QM) calculation of protein energies. In this work, we applied the EE-GMFCC method for calculation of binding affinity of Endonuclease colicin-immunity protein complex. The binding free energy changes between the wild-type and mutants of the complex calculated by EE-GMFCC are in good agreement with experimental results. The correlation coefficient (R) between the predicted binding energy changes and experimental values is 0.906 at the B3LYP/6-31G*-D level, based on the snapshot whose binding affinity is closest to the average result from the molecular mechanics/Poisson-Boltzmann surface area (MM/PBSA) calculation. The inclusion of the QM effects is important for accurate prediction of protein-protein binding affinities. Moreover, the self-consistent calculation of PB solvation energy is required for accurate calculations of protein-protein binding free energies. This study demonstrates that the EE-GMFCC method is capable of providing reliable prediction of relative binding affinities for protein-protein complexes. © 2018 Wiley Periodicals, Inc.

Toxicity is paramount for comprehending compound properties, particularly in the early stages of drug design. Due to the diversity and complexity of toxic effects, it became a challenge to compute compound toxicity tasks. To address this issue, we propose a multimodal representation learning model, termed multimodal graph isomorphism network (MMGIN), to address this challenge for compound toxicity multitask learning. Based on fingerprints and molecular graphs of compounds, our MMGIN model incorporates a multimodal representation learning model to acquire a comprehensive compound representation. This model adopts a two-channel structure to independently learn fingerprint representation and molecular graph representation. Subsequently, two feedforward neural networks utilize the learned multimodal compound representation to perform multitask learning, encompassing compound toxicity classification and multiple compound category classification simultaneously. To test the effectiveness of our model, we constructed a novel data set, termed the compound toxicity multitask learning (CTMTL) data set, derived from the TOXRIC data set. We compare our MMGIN model with other representative machine learning and deep learning models on the CTMTL and Tox21 data sets. The experimental results demonstrate significant advancements achieved by our MMGIN model. Furthermore, the ablation study underscores the effectiveness of the introduced fingerprints, molecular graphs, the multimodal representation learning model, and the multitask learning model, showcasing the model's superior predictive capability and robustness.

Quantitative structure-activity relationships (QSARs) are developed to describe the ability of 6-azasteroids to inhibit human type 1 5alpha-reductase. Models are generated using a set of 93 compounds with known binding affinities (K(i)) to 5alpha-reductase and 3beta-hydroxy-Delta(5)-steroid dehydrogenase/3-keto-Delta(5)-steroid isomerase (3-BHSD). QSARs are generated to predict K(i) values for inhibitors of 5alpha-reductase and to predict selectivity (S(i)) of compound binding to 3-BHSD relative to 5alpha-reductase. Log(K(i)) values range from -0.70 log units to 4.69 log units, and log(S(i)) values range from -3.00 log units to 3.84 log units. Topological, geometric, electronic, and polar surface descriptors are used to encode molecular structure. Information-rich subsets of descriptors are identified using evolutionary optimization procedures. Predictive models are generated using linear regression, computational neural networks (CNNs), principal components regression, and partial least squares. Compounds in an external prediction set are used for model validation. A 10-3-1 CNN is developed for prediction of binding affinity to 5alpha-reductase that produces root-mean-square error (RMSE) of 0.293 log units (R(2) = 0.97) for compounds in the external prediction set. Additionally, an 8-3-1 CNN is generated for prediction of inhibitor selectivity that produces RMSE = 0.513 log units (R(2) = 0.89) for the external prediction set. Models are further validated through Monte Carlo experiments in which models are generated after dependent variable values have been scrambled.