MG-BERT: leveraging unsupervised atomic representation learning for molecular property prediction

PepExplainer: An explainable deep learning model for selection-based macrocyclic peptide bioactivity prediction and optimization

Macrocyclic peptides possess unique features, making them highly promising as a drug modality. However, evaluating their bioactivity through wet lab experiments is generally resource-intensive and time-consuming. Despite advancements in artificial intelligence (AI) for bioactivity prediction, challenges remain due to limited data availability and the interpretability issues in deep learning models, often leading to less-than-ideal predictions. To address these challenges, we developed PepExplainer, an explainable graph neural network based on substructure mask explanation (SME). This model excels at deciphering amino acid substructures, translating macrocyclic peptides into detailed molecular graphs at the atomic level, and efficiently handling non-canonical amino acids and complex macrocyclic peptide structures. PepExplainer's effectiveness is enhanced by utilizing the correlation between peptide enrichment data from selection-based focused library and bioactivity data, and employing transfer learning to improve bioactivity predictions of macrocyclic peptides against IL-17C/IL-17 RE interaction. Additionally, PepExplainer underwent further validation for bioactivity prediction using an additional set of thirteen newly synthesized macrocyclic peptides. Moreover, it enabled the optimization of the IC50 of a macrocyclic peptide, reducing it from 15 nM to 5.6 nM based on the contribution score provided by PepExplainer. This achievement underscores PepExplainer's skill in deciphering complex molecular patterns, highlighting its potential to accelerate the discovery and optimization of macrocyclic peptides.

3MTox: A motif-level graph-based multi-view chemical language model for toxicity identification with deep interpretation

Toxicity identification plays a key role in maintaining human health, as it can alert humans to the potential hazards caused by long-term exposure to a wide variety of chemical compounds. Experimental methods for determining toxicity are time-consuming, and costly, while computational methods offer an alternative for the early identification of toxicity. For example, some classical ML and DL methods, which demonstrate excellent performance in toxicity prediction. However, these methods also have some defects, such as over-reliance on artificial features and easy overfitting, etc. Proposing novel models with superior prediction performance is still an urgent task. In this study, we propose a motifs-level graph-based multi-view pretraining language model, called 3MTox, for toxicity identification. The 3MTox model uses Bidirectional Encoder Representations from Transformers (BERT) as the backbone framework, and a motif graph as input. The results of extensive experiments showed that our 3MTox model achieved state-of-the-art performance on toxicity benchmark datasets and outperformed the baseline models considered. In addition, the interpretability of the model ensures that the it can quickly and accurately identify toxicity sites in a given molecule, thereby contributing to the determination of the status of toxicity and associated analyses. We think that the 3MTox model is among the most promising tools that are currently available for toxicity identification.

Crystal Composition Transformer: Self-Learning Neural Language Model for Generative and Tinkering Design of Materials

Self-supervised neural language models have recently achieved unprecedented success from natural language processing to learning the languages of biological sequences and organic molecules. These models have demonstrated superior performance in the generation, structure classification, and functional predictions for proteins and molecules with learned representations. However, most of the masking-based pre-trained language models are not designed for generative design, and their black-box nature makes it difficult to interpret their design logic. Here a Blank-filling Language Model for Materials (BLMM) Crystal Transformer is proposed, a neural network-based probabilistic generative model for generative and tinkering design of inorganic materials. The model is built on the blank-filling language model for text generation and has demonstrated unique advantages in learning the "materials grammars" together with high-quality generation, interpretability, and data efficiency. It can generate chemically valid materials compositions with as high as 89.7% charge neutrality and 84.8% balanced electronegativity, which are more than four and eight times higher compared to a pseudo-random sampling baseline. The probabilistic generation process of BLMM allows it to recommend materials tinkering operations based on learned materials chemistry, which makes it useful for materials doping. The model is applied to discover a set of new materials as validated using the Density Functional Theory (DFT) calculations. This work thus brings the unsupervised transformer language models based generative artificial intelligence to inorganic materials. A user-friendly web app for tinkering materials design has been developed and can be accessed freely at www.materialsatlas.org/blmtinker.

Hybrid fragment-SMILES tokenization for ADMET prediction in drug discovery

BACKGROUND

Drug discovery and development is the extremely costly and time-consuming process of identifying new molecules that can interact with a biomarker target to interrupt the disease pathway of interest. In addition to binding the target, a drug candidate needs to satisfy multiple properties affecting absorption, distribution, metabolism, excretion, and toxicity (ADMET). Artificial intelligence approaches provide an opportunity to improve each step of the drug discovery and development process, in which the first question faced by us is how a molecule can be informatively represented such that the in-silico solutions are optimized.

RESULTS

This study introduces a novel hybrid SMILES-fragment tokenization method, coupled with two pre-training strategies, utilizing a Transformer-based model. We investigate the efficacy of hybrid tokenization in improving the performance of ADMET prediction tasks. Our approach leverages MTL-BERT, an encoder-only Transformer model that achieves state-of-the-art ADMET predictions, and contrasts the standard SMILES tokenization with our hybrid method across a spectrum of fragment library cutoffs.

CONCLUSION

The findings reveal that while an excess of fragments can impede performance, using hybrid tokenization with high frequency fragments enhances results beyond the base SMILES tokenization. This advancement underscores the potential of integrating fragment- and character-level molecular features within the training of Transformer models for ADMET property prediction.

Advancing drug discovery with deep attention neural networks

In the dynamic field of drug discovery, deep attention neural networks are revolutionizing our approach to complex data. This review explores the attention mechanism and its extended architectures, including graph attention networks (GATs), transformers, bidirectional encoder representations from transformers (BERT), generative pre-trained transformers (GPTs) and bidirectional and auto-regressive transformers (BART). Delving into their core principles and multifaceted applications, we uncover their pivotal roles in catalyzing de novo drug design, predicting intricate molecular properties and deciphering elusive drug-target interactions. Despite challenges, these attention-based architectures hold unparalleled promise to drive transformative breakthroughs and accelerate progress in pharmaceutical research.

Multi-task pretrained language model with novel application domains enables more comprehensive health and ecological toxicity prediction

In silico models for screening substances of healthy and ecological concern are essential for effective chemical management. However, current data-driven toxicity prediction models confront formidable challenges related to expressive capacity, data scarcity, and reliability issues. Thus, this study introduces TOX-BERT, a SMILES-based pretrained model for screening health and ecological toxicity. Results show that masked atom recovery pretraining and multi-task learning offer promising solutions to enhance model capacity and address data scarcity issues. Two novel application domain (AD) parameters, termed PCA-AD and LDS, were proposed to improve prediction reliability of TOX-BERT with accuracy surpassing 90 % and mean absolute error (MAE) below 0.52. TOX-BERT was applied to 18,905 IECSC chemicals, revealing distinct toxicity relationships that align with experimental studies such as those between cardiotoxicity and acute ecotoxicity. In addition to previous PBT screening, 156 potential high-risk chemicals for specific endpoint were identified covering 7 categories. Furthermore, a SMILES-based toxicity site detection approach was developed for structural toxicity analysis. These advancements carry profound implications to address challenges faced by current data-driven toxicity prediction models. TOX-BERT emerges as a valuable tool for more comprehensive, reliable, and applicable predictions of health and ecological toxicity in chemical risk assessment and management.

Quantum-Informed Molecular Representation Learning Enhancing ADMET Property Prediction

We examined pretraining tasks leveraging abundant labeled data to effectively enhance molecular representation learning in downstream tasks, specifically emphasizing graph transformers to improve the prediction of ADMET properties. Our investigation revealed limitations in previous pretraining tasks and identified more meaningful training targets, ranging from 2D molecular descriptors to extensive quantum chemistry simulations. These data were seamlessly integrated into supervised pretraining tasks. The implementation of our pretraining strategy and multitask learning outperforms conventional methods, achieving state-of-the-art outcomes in 7 out of 22 ADMET tasks within the Therapeutics Data Commons by utilizing a shared encoder across all tasks. Our approach underscores the effectiveness of learning molecular representations and highlights the potential for scalability when leveraging extensive data sets, marking a significant advancement in this domain.

Can large language models understand molecules?

PURPOSE

Large Language Models (LLMs) like Generative Pre-trained Transformer (GPT) from OpenAI and LLaMA (Large Language Model Meta AI) from Meta AI are increasingly recognized for their potential in the field of cheminformatics, particularly in understanding Simplified Molecular Input Line Entry System (SMILES), a standard method for representing chemical structures. These LLMs also have the ability to decode SMILES strings into vector representations.

METHOD

We investigate the performance of GPT and LLaMA compared to pre-trained models on SMILES in embedding SMILES strings on downstream tasks, focusing on two key applications: molecular property prediction and drug-drug interaction prediction.

RESULTS

We find that SMILES embeddings generated using LLaMA outperform those from GPT in both molecular property and DDI prediction tasks. Notably, LLaMA-based SMILES embeddings show results comparable to pre-trained models on SMILES in molecular prediction tasks and outperform the pre-trained models for the DDI prediction tasks.

CONCLUSION

The performance of LLMs in generating SMILES embeddings shows great potential for further investigation of these models for molecular embedding. We hope our study bridges the gap between LLMs and molecular embedding, motivating additional research into the potential of LLMs in the molecular representation field. GitHub: https://github.com/sshaghayeghs/LLaMA-VS-GPT .

Enhancing Molecular Property Prediction through Task-Oriented Transfer Learning: Integrating Universal Structural Insights and Domain-Specific Knowledge

Precisely predicting molecular properties is crucial in drug discovery, but the scarcity of labeled data poses a challenge for applying deep learning methods. While large-scale self-supervised pretraining has proven an effective solution, it often neglects domain-specific knowledge. To tackle this issue, we introduce Task-Oriented Multilevel Learning based on BERT (TOML-BERT), a dual-level pretraining framework that considers both structural patterns and domain knowledge of molecules. TOML-BERT achieved state-of-the-art prediction performance on 10 pharmaceutical datasets. It has the capability to mine contextual information within molecular structures and extract domain knowledge from massive pseudo-labeled data. The dual-level pretraining accomplished significant positive transfer, with its two components making complementary contributions. Interpretive analysis elucidated that the effectiveness of the dual-level pretraining lies in the prior learning of a task-related molecular representation. Overall, TOML-BERT demonstrates the potential of combining multiple pretraining tasks to extract task-oriented knowledge, advancing molecular property prediction in drug discovery.

Predicting ADMET Properties from Molecule SMILE: A Bottom-Up Approach Using Attention-Based Graph Neural Networks

Understanding the pharmacokinetics, safety and efficacy of candidate drugs is crucial for their success. One key aspect is the characterization of absorption, distribution, metabolism, excretion and toxicity (ADMET) properties, which require early assessment in the drug discovery and development process. This study aims to present an innovative approach for predicting ADMET properties using attention-based graph neural networks (GNNs). The model utilizes a graph-based representation of molecules directly derived from Simplified Molecular Input Line Entry System (SMILE) notation. Information is processed sequentially, from substructures to the whole molecule, employing a bottom-up approach. The developed GNN is tested and compared with existing approaches using six benchmark datasets and by encompassing regression (lipophilicity and aqueous solubility) and classification (CYP2C9, CYP2C19, CYP2D6 and CYP3A4 inhibition) tasks. Results show the effectiveness of our model, which bypasses the computationally expensive retrieval and selection of molecular descriptors. This approach provides a valuable tool for high-throughput screening, facilitating early assessment of ADMET properties and enhancing the likelihood of drug success in the development pipeline.

Advancing Drug-Target Interaction prediction with BERT and subsequence embedding

Exploring the relationship between proteins and drugs plays a significant role in discovering new synthetic drugs. The Drug-Target Interaction (DTI) prediction is a fundamental task in the relationship between proteins and drugs. Unlike encoding proteins by amino acids, we use amino acid subsequence to encode proteins, which simulates the biological process of DTI better. For this research purpose, we proposed a novel deep learning framework based on Bidirectional Encoder Representation from Transformers (BERT), which integrates high-frequency subsequence embedding and transfer learning methods to complete the DTI prediction task. As the first key module, subsequence embedding allows to explore the functional interaction units from drug and protein sequences and then contribute to finding DTI modules. As the second key module, transfer learning promotes the model learn the common DTI features from protein and drug sequences in a large dataset. Overall, the BERT-based model can learn two kinds features through the multi-head self-attention mechanism: internal features of sequence and interaction features of both proteins and drugs, respectively. Compared with other methods, BERT-based methods enable more DTI-related features to be discovered by means of attention scores which associated with tokenized protein/drug subsequences. We conducted extensive experiments for the DTI prediction task on three different benchmark datasets. The experimental results show that the model achieves an average prediction metrics higher than most baseline methods. In order to verify the importance of transfer learning, we conducted an ablation study on datasets, and the results show the superiority of transfer learning. In addition, we test the scalability of the model on the dataset in unseen drugs and proteins, and the results of the experiments show that it is acceptable in scalability.

Large language models for science and medicine

Large language models (LLMs) are a type of machine learning model that learn statistical patterns over text, such as predicting the next words in a sequence of text. Both general purpose and task-specific LLMs have demonstrated potential across diverse applications. Science and medicine have many data types that are highly suitable for LLMs, such as scientific texts (publications, patents and textbooks), electronic medical records, large databases of DNA and protein sequences and chemical compounds. Carefully validated systems that can understand and reason across all these modalities may maximize benefits. Despite the inevitable limitations and caveats of any new technology and some uncertainties specific to LLMs, LLMs have the potential to be transformative in science and medicine.

Complementary multi-modality molecular self-supervised learning via non-overlapping masking for property prediction

Self-supervised learning plays an important role in molecular representation learning because labeled molecular data are usually limited in many tasks, such as chemical property prediction and virtual screening. However, most existing molecular pre-training methods focus on one modality of molecular data, and the complementary information of two important modalities, SMILES and graph, is not fully explored. In this study, we propose an effective multi-modality self-supervised learning framework for molecular SMILES and graph. Specifically, SMILES data and graph data are first tokenized so that they can be processed by a unified Transformer-based backbone network, which is trained by a masked reconstruction strategy. In addition, we introduce a specialized non-overlapping masking strategy to encourage fine-grained interaction between these two modalities. Experimental results show that our framework achieves state-of-the-art performance in a series of molecular property prediction tasks, and a detailed ablation study demonstrates efficacy of the multi-modality framework and the masking strategy.

Gram matrix: an efficient representation of molecular conformation and learning objective for molecular pretraining

Accurate prediction of molecular properties is fundamental in drug discovery and development, providing crucial guidance for effective drug design. A critical factor in achieving accurate molecular property prediction lies in the appropriate representation of molecular structures. Presently, prevalent deep learning-based molecular representations rely on 2D structure information as the primary molecular representation, often overlooking essential three-dimensional (3D) conformational information due to the inherent limitations of 2D structures in conveying atomic spatial relationships. In this study, we propose employing the Gram matrix as a condensed representation of 3D molecular structures and for efficient pretraining objectives. Subsequently, we leverage this matrix to construct a novel molecular representation model, Pre-GTM, which inherently encapsulates 3D information. The model accurately predicts the 3D structure of a molecule by estimating the Gram matrix. Our findings demonstrate that Pre-GTM model outperforms the baseline Graphormer model and other pretrained models in the QM9 and MoleculeNet quantitative property prediction task. The integration of the Gram matrix as a condensed representation of 3D molecular structure, incorporated into the Pre-GTM model, opens up promising avenues for its potential application across various domains of molecular research, including drug design, materials science, and chemical engineering.

Advances in machine intelligence-driven virtual screening approaches for big-data

Virtual screening (VS) is an integral and ever-evolving domain of drug discovery framework. The VS is traditionally classified into ligand-based (LB) and structure-based (SB) approaches. Machine intelligence or artificial intelligence has wide applications in the drug discovery domain to reduce time and resource consumption. In combination with machine intelligence algorithms, VS has emerged into revolutionarily progressive technology that learns within robust decision orders for data curation and hit molecule screening from large VS libraries in minutes or hours. The exponential growth of chemical and biological data has evolved as "big-data" in the public domain demands modern and advanced machine intelligence-driven VS approaches to screen hit molecules from ultra-large VS libraries. VS has evolved from an individual approach (LB and SB) to integrated LB and SB techniques to explore various ligand and target protein aspects for the enhanced rate of appropriate hit molecule prediction. Current trends demand advanced and intelligent solutions to handle enormous data in drug discovery domain for screening and optimizing hits or lead with fewer or no false positive hits. Following the big-data drift and tremendous growth in computational architecture, we presented this review. Here, the article categorized and emphasized individual VS techniques, detailed literature presented for machine learning implementation, modern machine intelligence approaches, and limitations and deliberated the future prospects.

INTransformer: Data augmentation-based contrastive learning by injecting noise into transformer for molecular property prediction

Molecular property prediction plays an essential role in drug discovery for identifying the candidate molecules with target properties. Deep learning models usually require sufficient labeled data to train good prediction models. However, the size of labeled data is usually small for molecular property prediction, which brings great challenges to deep learning-based molecular property prediction methods. Furthermore, the global information of molecules is critical for predicting molecular properties. Therefore, we propose INTransformer for molecular property prediction, which is a data augmentation method via contrastive learning to alleviate the limitations of the labeled molecular data while enhancing the ability to capture global information. Specifically, INTransformer consists of two identical Transformer sub-encoders to extract the molecular representation from the original SMILES and noisy SMILES respectively, while achieving the goal of data augmentation. To reduce the influence of noise, we use contrastive learning to ensure the molecular encoding of noisy SMILES is consistent with that of the original input so that the molecular representation information can be better extracted by INTransformer. Experiments on various benchmark datasets show that INTransformer achieved competitive performance for molecular property prediction tasks compared with the baselines and state-of-the-art methods.

Pre-training molecular representation model with spatial geometry for property prediction

AI-enhanced bioinformatics and cheminformatics pivots on generating increasingly descriptive and generalized molecular representation. Accurate prediction of molecular properties needs a comprehensive description of molecular geometry. We design a novel Graph Isomorphic Network (GIN) based model integrating a three-level network structure with a dual-level pre-training approach that aligns the characteristics of molecules. In our Spatial Molecular Pre-training (SMPT) Model, the network can learn implicit geometric information in layers from lower to higher according to the dimension. Extensive evaluations against established baseline models validate the enhanced efficacy of SMPT, with notable accomplishments in classification tasks. These results emphasize the importance of spatial geometric information in molecular representation modeling and demonstrate the potential of SMPT as a valuable tool for property prediction.

A deep learning framework for predicting molecular property based on multi-type features fusion

Extracting expressive molecular features is essential for molecular property prediction. Sequence-based representation is a common representation of molecules, which ignores the structure information of molecules. While molecular graph representation has a weak ability in expressing the 3D structure. In this article, we try to make use of the advantages of different type representations simultaneously for molecular property prediction. Thus, we propose a fusion model named DLF-MFF, which integrates the multi-type molecular features. Specifically, we first extract four different types of features from molecular fingerprints, 2D molecular graph, 3D molecular graph and molecular image. Then, in order to learn molecular features individually, we use four essential deep learning frameworks, which correspond to four distinct molecular representations. The final molecular representation is created by integrating the four feature vectors and feeding them into prediction layer to predict molecular property. We compare DLF-MFF with 7 state-of-the-art methods on 6 benchmark datasets consisting of multiple molecular properties, the experimental results show that DLF-MFF achieves state-of-the-art performance on 6 benchmark datasets. Moreover, DLF-MFF is applied to identify potential anti-SARS-CoV-2 inhibitor from 2500 drugs. We predict probability of each drug being inferred as a 3CL protease inhibitor and also calculate the binding affinity scores between each drug and 3CL protease. The results show that DLF-MFF product better performance in the identification of anti-SARS-CoV-2 inhibitor. This work is expected to offer novel research perspectives for accurate prediction of molecular properties and provide valuable insights into drug repurposing for COVID-19.

ChemMORT: an automatic ADMET optimization platform using deep learning and multi-objective particle swarm optimization

Drug discovery and development constitute a laborious and costly undertaking. The success of a drug hinges not only good efficacy but also acceptable absorption, distribution, metabolism, elimination, and toxicity (ADMET) properties. Overall, up to 50% of drug development failures have been contributed from undesirable ADMET profiles. As a multiple parameter objective, the optimization of the ADMET properties is extremely challenging owing to the vast chemical space and limited human expert knowledge. In this study, a freely available platform called Chemical Molecular Optimization, Representation and Translation (ChemMORT) is developed for the optimization of multiple ADMET endpoints without the loss of potency (https://cadd.nscc-tj.cn/deploy/chemmort/). ChemMORT contains three modules: Simplified Molecular Input Line Entry System (SMILES) Encoder, Descriptor Decoder and Molecular Optimizer. The SMILES Encoder can generate the molecular representation with a 512-dimensional vector, and the Descriptor Decoder is able to translate the above representation to the corresponding molecular structure with high accuracy. Based on reversible molecular representation and particle swarm optimization strategy, the Molecular Optimizer can be used to effectively optimize undesirable ADMET properties without the loss of bioactivity, which essentially accomplishes the design of inverse QSAR. The constrained multi-objective optimization of the poly (ADP-ribose) polymerase-1 inhibitor is provided as the case to explore the utility of ChemMORT.

Attention is all you need: utilizing attention in AI-enabled drug discovery

Recently, attention mechanism and derived models have gained significant traction in drug development due to their outstanding performance and interpretability in handling complex data structures. This review offers an in-depth exploration of the principles underlying attention-based models and their advantages in drug discovery. We further elaborate on their applications in various aspects of drug development, from molecular screening and target binding to property prediction and molecule generation. Finally, we discuss the current challenges faced in the application of attention mechanisms and Artificial Intelligence technologies, including data quality, model interpretability and computational resource constraints, along with future directions for research. Given the accelerating pace of technological advancement, we believe that attention-based models will have an increasingly prominent role in future drug discovery. We anticipate that these models will usher in revolutionary breakthroughs in the pharmaceutical domain, significantly accelerating the pace of drug development.

Recent Applications of Machine Learning in Molecular Property and Chemical Reaction Outcome Predictions

Burgeoning developments in machine learning (ML) and its rapidly growing adaptations in chemistry are noteworthy. Motivated by the successful deployments of ML in the realm of molecular property prediction (MPP) and chemical reaction prediction (CRP), herein we highlight some of its most recent applications in predictive chemistry. We present a nonmathematical and concise overview of the progression of ML implementations, ranging from an ensemble-based random forest model to advanced graph neural network algorithms. Similarly, the prospects of various feature engineering and feature learning approaches that work in conjunction with ML models are described. Highly accurate predictions reported in MPP tasks (e.g., lipophilicity, solubility, distribution coefficient), using methods such as D-MPNN, MolCLR, SMILES-BERT, and MolBERT, offer promising avenues in molecular design and drug discovery. Whereas MPP pertains to a given molecule, ML applications in chemical reactions present a different level of challenge, primarily arising from the simultaneous involvement of multiple molecules and their diverse roles in a reaction setting. The reported RMSEs in MPP tasks range from 0.287 to 2.20, while those for yield predictions are well over 4.9 in the lower end, reaching thresholds of >10.0 in several examples. Our Review concludes with a set of persisting challenges in dealing with reaction data sets and an overall optimistic outlook on benefits of ML-driven workflows for various MPP as well as CRP tasks.

3D-SMGE: a pipeline for scaffold-based molecular generation and evaluation

In the process of drug discovery, one of the key problems is how to improve the biological activity and ADMET properties starting from a specific structure, which is also called structural optimization. Based on a starting scaffold, the use of deep generative model to generate molecules with desired drug-like properties will provide a powerful tool to accelerate the structural optimization process. However, the existing generative models remain challenging in extracting molecular features efficiently in 3D space to generate drug-like 3D molecules. Moreover, most of the existing ADMET prediction models made predictions of different properties through a single model, which can result in reduced prediction accuracy on some datasets. To effectively generate molecules from a specific scaffold and provide basis for the structural optimization, the 3D-SMGE (3-Dimensional Scaffold-based Molecular Generation and Evaluation) work consisting of molecular generation and prediction of ADMET properties is presented. For the molecular generation, we proposed 3D-SMG, a novel deep generative model for the end-to-end design of 3D molecules. In the 3D-SMG model, we designed the cross-aggregated continuous-filter convolution (ca-cfconv), which is used to achieve efficient and low-cost 3D spatial feature extraction while ensuring the invariance of atomic space rotation. 3D-SMG was proved to generate valid, unique and novel molecules with high drug-likeness. Besides, the proposed data-adaptive multi-model ADMET prediction method outperformed or maintained the best evaluation metrics on 24 out of 27 ADMET benchmark datasets. 3D-SMGE is anticipated to emerge as a powerful tool for hit-to-lead structural optimizations and accelerate the drug discovery process.

FG-BERT: a generalized and self-supervised functional group-based molecular representation learning framework for properties prediction

Artificial intelligence-based molecular property prediction plays a key role in molecular design such as bioactive molecules and functional materials. In this study, we propose a self-supervised pretraining deep learning (DL) framework, called functional group bidirectional encoder representations from transformers (FG-BERT), pertained based on ~1.45 million unlabeled drug-like molecules, to learn meaningful representation of molecules from function groups. The pretrained FG-BERT framework can be fine-tuned to predict molecular properties. Compared to state-of-the-art (SOTA) machine learning and DL methods, we demonstrate the high performance of FG-BERT in evaluating molecular properties in tasks involving physical chemistry, biophysics and physiology across 44 benchmark datasets. In addition, FG-BERT utilizes attention mechanisms to focus on FG features that are critical to the target properties, thereby providing excellent interpretability for downstream training tasks. Collectively, FG-BERT does not require any artificially crafted features as input and has excellent interpretability, providing an out-of-the-box framework for developing SOTA models for a variety of molecule (especially for drug) discovery tasks.

TransFoxMol: predicting molecular property with focused attention

Predicting the biological properties of molecules is crucial in computer-aided drug development, yet it's often impeded by data scarcity and imbalance in many practical applications. Existing approaches are based on self-supervised learning or 3D data and using an increasing number of parameters to improve performance. These approaches may not take full advantage of established chemical knowledge and could inadvertently introduce noise into the respective model. In this study, we introduce a more elegant transformer-based framework with focused attention for molecular representation (TransFoxMol) to improve the understanding of artificial intelligence (AI) of molecular structure property relationships. TransFoxMol incorporates a multi-scale 2D molecular environment into a graph neural network + Transformer module and uses prior chemical maps to obtain a more focused attention landscape compared to that obtained using existing approaches. Experimental results show that TransFoxMol achieves state-of-the-art performance on MoleculeNet benchmarks and surpasses the performance of baselines that use self-supervised learning or geometry-enhanced strategies on small-scale datasets. Subsequent analyses indicate that TransFoxMol's predictions are highly interpretable and the clever use of chemical knowledge enables AI to perceive molecules in a simple but rational way, enhancing performance.

Prediction of Multiple Types of RNA Modifications via Biological Language Model

It has been demonstrated that RNA modifications play essential roles in multiple biological processes. Accurate identification of RNA modifications in the transcriptome is critical for providing insights into the biological functions and mechanisms. Many tools have been developed for predicting RNA modifications at single-base resolution, which employ conventional feature engineering methods that focus on feature design and feature selection processes that require extensive biological expertise and may introduce redundant information. With the rapid development of artificial intelligence technologies, end-to-end methods are favorably received by researchers. Nevertheless, each well-trained model is only suitable for a specific RNA methylation modification type for nearly all of these approaches. In this study, we present MRM-BERT by feeding task-specific sequences into the powerful BERT (Bidirectional Encoder Representations from Transformers) model and implementing fine-tuning, which exhibits competitive performance to the state-of-the-art methods. MRM-BERT avoids repeated de novo training of the model and can predict multiple RNA modifications such as pseudouridine, m6A, m5C, and m1A in Mus musculus, Arabidopsis thaliana, and Saccharomyces cerevisiae. In addition, we analyse the attention heads to provide high attention regions for the prediction, and conduct saturated in silico mutagenesis of the input sequences to discover potential changes of RNA modifications, which can better assist researchers in their follow-up research.

Similarity-based pairing improves efficiency of siamese neural networks for regression tasks and uncertainty quantification

Siamese networks, representing a novel class of neural networks, consist of two identical subnetworks sharing weights but receiving different inputs. Here we present a similarity-based pairing method for generating compound pairs to train Siamese neural networks for regression tasks. In comparison with the conventional exhaustive pairing, it reduces the algorithm complexity from O(n2) to O(n). It also results in a better prediction performance consistently on the three physicochemical datasets, using a multilayer perceptron with the circular fingerprint as a proof of concept. We further include into a Siamese neural network the transformer-based Chemformer, which extracts task-specific features from the simplified molecular-input line-entry system representation of compounds. Additionally, we propose a means to measure the prediction uncertainty by utilizing the variance in predictions from a set of reference compounds. Our results demonstrate that the high prediction accuracy correlates with the high confidence. Finally, we investigate implications of the similarity property principle in machine learning.

PepScaf: Harnessing Machine Learning with In Vitro Selection toward De Novo Macrocyclic Peptides against IL-17C/IL-17RE Interaction

The combination of library-based screening and artificial intelligence (AI) has been accelerating the discovery and optimization of hit ligands. However, the potential of AI to assist in de novo macrocyclic peptide ligand discovery has yet to be fully explored. In this study, an integrated AI framework called PepScaf was developed to extract the critical scaffold relative to bioactivity based on a vast dataset from an initial in vitro selection campaign against a model protein target, interleukin-17C (IL-17C). Taking the generated scaffold, a focused macrocyclic peptide library was rationally constructed to target IL-17C, yielding over 20 potent peptides that effectively inhibited IL-17C/IL-17RE interaction. Notably, the top two peptides displayed exceptional potency with IC50 values of 1.4 nM. This approach presents a viable methodology for more efficient macrocyclic peptide discovery, offering potential time and cost savings. Additionally, this is also the first report regarding the discovery of macrocyclic peptides against IL-17C/IL-17RE interaction.

MVIL6: Accurate identification of IL-6-induced peptides using multi-view feature learning

Interleukin-6 (IL-6) is a potential therapeutic target for many diseases, and it is of great significance in accurately predicting IL-6-induced peptides for IL-6 research. However, the cost of traditional wet experiments to detect IL-6-induced peptides is huge, and the discovery and design of peptides by computer before the experimental stage have become a promising technology. In this study, we developed a deep learning model called MVIL6 for predicting IL-6-inducing peptides. Comparative results demonstrated the outstanding performance and robustness of MVIL6. Specifically, we employ a pre-trained protein language model MG-BERT and the Transformer model to process two different sequence-based descriptors and integrate them with a fusion module to improve the prediction performance. The ablation experiment demonstrated the effectiveness of our fusion strategy for the two models. In addition, to provide good interpretability of our model, we explored and visualized the amino acids considered important for IL-6-induced peptide prediction by our model. Finally, a case study presented using MVIL6 to predict IL-6-induced peptides in the SARS-CoV-2 spike protein shows that MVIL6 achieves higher performance than existing methods and can be useful for identifying potential IL-6-induced peptides in viral proteins.

SMG-BERT: integrating stereoscopic information and chemical representation for molecular property prediction

Molecular property prediction is a crucial task in various fields and has recently garnered significant attention. To achieve accurate and fast prediction of molecular properties, machine learning (ML) models have been widely employed due to their superior performance compared to traditional methods by trial and error. However, most of the existing ML models that do not incorporate 3D molecular information are still in need of improvement, as they are mostly poor at differentiating stereoisomers of certain types, particularly chiral ones. Also,routine featurization methods using only incomplete features are hard to obtain explicable molecular representations. In this paper, we propose the Stereo Molecular Graph BERT (SMG-BERT) by integrating the 3D space geometric parameters, 2D topological information, and 1D SMILES string into the self-attention-based BERT model. In addition, nuclear magnetic resonance (NMR) spectroscopy results and bond dissociation energy (BDE) are integrated as extra atomic and bond features to improve the model's performance and interpretability analysis. The comprehensive integration of 1D, 2D, and 3D information could establish a unified and unambiguous molecular characterization system to distinguish conformations, such as chiral molecules. Intuitively integrated chemical information enables the model to possess interpretability that is consistent with chemical logic. Experimental results on 12 benchmark molecular datasets show that SMG-BERT consistently outperforms existing methods. At the same time, the experimental results demonstrate that SMG-BERT is generalizable and reliable.

PharmBERT: a domain-specific BERT model for drug labels

Human prescription drug labeling contains a summary of the essential scientific information needed for the safe and effective use of the drug and includes the Prescribing Information, FDA-approved patient labeling (Medication Guides, Patient Package Inserts and/or Instructions for Use), and/or carton and container labeling. Drug labeling contains critical information about drug products, such as pharmacokinetics and adverse events. Automatic information extraction from drug labels may facilitate finding the adverse reaction of the drugs or finding the interaction of one drug with another drug. Natural language processing (NLP) techniques, especially recently developed Bidirectional Encoder Representations from Transformers (BERT), have exhibited exceptional merits in text-based information extraction. A common paradigm in training BERT is to pretrain the model on large unlabeled generic language corpora, so that the model learns the distribution of the words in the language, and then fine-tune on a downstream task. In this paper, first, we show the uniqueness of language used in drug labels, which therefore cannot be optimally handled by other BERT models. Then, we present the developed PharmBERT, which is a BERT model specifically pretrained on the drug labels (publicly available at Hugging Face). We demonstrate that our model outperforms the vanilla BERT, ClinicalBERT and BioBERT in multiple NLP tasks in the drug label domain. Moreover, how the domain-specific pretraining has contributed to the superior performance of PharmBERT is demonstrated by analyzing different layers of PharmBERT, and more insight into how it understands different linguistic aspects of the data is gained.

NoiseMol: A noise-robusted data augmentation via perturbing noise for molecular property prediction

Simplified Molecular-Input Line-Entry System (SMILES) is one of a widely used molecular representation methods for molecular property prediction. We conjecture that all the characters in the SMILES string of a molecule are essential for making up the molecules, but most of them make little contribution to determining a particular property of the molecule. Therefore, we verified the conjecture in the pre-experiment. Motivated by the result, we propose to inject proper noisy information into the SMILES to augment the training data by increasing the diversity of the labeled molecules. To this end, we explore injecting perturbing noise into the original labeled SMILES strings to construct augmented data for alleviating the limitation of the labeled compound data and enhancing the model to extract more useful molecular representation for molecular property prediction. Specifically, we directly adopt mask, swap, deletion, and fusion operations on SMILES strings to randomly mask, swap, and delete atoms in SMILES strings. Then, the augmented data is used by two strategies: each epoch alternately feeds the original and perturbing noisy molecules, or each batch alternately feeds the original and perturbing noisy molecules. We conduct experiments on both Transformer and BiGRU models to validate the effectiveness by adopting widely used datasets from MoleculeNet and ZINC. Experimental results demonstrate that the proposed method outperforms strong baselines on all the datasets. NoiseMol obtains the best performance on BBBP and FDA when compared with state-of-the-art methods. Besides, NoiseMol achieves the best accuracy on LogP. Therefore, injecting perturbing noise into the labeled SMILES strings is an effective and efficient method, which improves the prediction performance, generalization, and robustness of the deep learning models.

Drug Discovery Targeting Post-Translational Modifications in Response to DNA Damages Induced by Space Radiation

DNA damage in astronauts induced by cosmic radiation poses a major barrier to human space exploration. Cellular responses and repair of the most lethal DNA double-strand breaks (DSBs) are crucial for genomic integrity and cell survival. Post-translational modifications (PTMs), including phosphorylation, ubiquitylation, and SUMOylation, are among the regulatory factors modulating a delicate balance and choice between predominant DSB repair pathways, such as non-homologous end joining (NHEJ) and homologous recombination (HR). In this review, we focused on the engagement of proteins in the DNA damage response (DDR) modulated by phosphorylation and ubiquitylation, including ATM, DNA-PKcs, CtIP, MDM2, and ubiquitin ligases. The involvement and function of acetylation, methylation, PARylation, and their essential proteins were also investigated, providing a repository of candidate targets for DDR regulators. However, there is a lack of radioprotectors in spite of their consideration in the discovery of radiosensitizers. We proposed new perspectives for the research and development of future agents against space radiation by the systematic integration and utilization of evolutionary strategies, including multi-omics analyses, rational computing methods, drug repositioning, and combinations of drugs and targets, which may facilitate the use of radioprotectors in practical applications in human space exploration to combat fatal radiation hazards.

The prediction of molecular toxicity based on BiGRU and GraphSAGE

The prediction of molecules toxicity properties plays an crucial role in the realm of the drug discovery, since it can swiftly screen out the expected drug moleculars. The conventional method for predicting toxicity is to use some in vivo or in vitro biological experiments in the laboratory, which can easily pose a threat significant time and financial waste and even ethical issues. Therefore, using computational approaches to predict molecular toxicity has become a common strategy in modern drug discovery. In this article, we propose a novel model named MTBG, which primarily makes use of both SMILES (Simplified molecular input line entry system) strings and graph structures of molecules to extract drug molecular feature in the field of drug molecular toxicity prediction. To verify the performance of the MTBG model, we opt the Tox21 dataset and several widely used baseline models. Experimental results demonstrate that our model can perform better than these baseline models.

Learning with uncertainty to accelerate the discovery of histone lysine-specific demethylase 1A (KDM1A/LSD1) inhibitors

Machine learning including modern deep learning models has been extensively used in drug design and screening. However, reliable prediction of molecular properties is still challenging when exploring out-of-domain regimes, even for deep neural networks. Therefore, it is important to understand the uncertainty of model predictions, especially when the predictions are used to guide further experiments. In this study, we explored the utility and effectiveness of evidential uncertainty in compound screening. The evidential Graphormer model was proposed for uncertainty-guided discovery of KDM1A/LSD1 inhibitors. The benchmarking results illustrated that (i) Graphormer exhibited comparative predictive power to state-of-the-art models, and (ii) evidential regression enabled well-ranked uncertainty estimates and calibrated predictions. Subsequently, we leveraged time-splitting on the curated KDM1A/LSD1 dataset to simulate out-of-distribution predictions. The retrospective virtual screening showed that the evidential uncertainties helped reduce false positives among the top-acquired compounds and thus enabled higher experimental validation rates. The trained model was then used to virtually screen an independent in-house compound set. The top 50 compounds ranked by two different ranking strategies were experimentally validated, respectively. In general, our study highlighted the importance to understand the uncertainty in prediction, which can be recognized as an interpretable dimension to model predictions.

CasANGCL: pre-training and fine-tuning model based on cascaded attention network and graph contrastive learning for molecular property prediction

MOTIVATION

Molecular property prediction is a significant requirement in AI-driven drug design and discovery, aiming to predict the molecular property information (e.g. toxicity) based on the mined biomolecular knowledge. Although graph neural networks have been proven powerful in predicting molecular property, unbalanced labeled data and poor generalization capability for new-synthesized molecules are always key issues that hinder further improvement of molecular encoding performance.

RESULTS

We propose a novel self-supervised representation learning scheme based on a Cascaded Attention Network and Graph Contrastive Learning (CasANGCL). We design a new graph network variant, designated as cascaded attention network, to encode local-global molecular representations. We construct a two-stage contrast predictor framework to tackle the label imbalance problem of training molecular samples, which is an integrated end-to-end learning scheme. Moreover, we utilize the information-flow scheme for training our network, which explicitly captures the edge information in the node/graph representations and obtains more fine-grained knowledge. Our model achieves an 81.9% ROC-AUC average performance on 661 tasks from seven challenging benchmarks, showing better portability and generalizations. Further visualization studies indicate our model's better representation capacity and provide interpretability.

Transformer-based deep learning method for optimizing ADMET properties of lead compounds

A successful drug needs to exhibit both effective pharmacodynamics (PD) and safe pharmacokinetics (PK). However, the coordinated optimization of PD and PK properties in molecule generation tasks remains a great challenge for most existing methods, especially when they focus on the pursuit of affinity and selectivity for the lead compound. Thus, molecular optimization for PK properties is a critical step in the drug discovery pipeline, in which absorption, distribution, metabolism, excretion and toxicity (ADMET) property predictive models play an increasingly important role by providing an effective method to assess multiple PK properties of compounds. Here, we proposed a Graph Bert-based ADMET prediction model that achieves state-of-the-art performance on the public dataset Therapeutics Data Commons (TDC) by combining molecular graph features and descriptor features, with 11 tasks ranked first and 20 tasks ranked in the top 3. Based on this prediction model, we trained a Transformer model with multiple properties as constraints for learning the structural transformations involved in MMP and the accompanying property changes. The experimental results show that the trained Constraints-Transformer can implement targeted modifications to the starting molecule, while preserving the core scaffold. Moreover, molecular docking and binding mode analysis demonstrate that the optimized molecules still retain the activity and selectivity for biological targets. Therefore, the proposed method accounts for biological activity and ADMET properties simultaneously. Finally, a webserver containing ADMET property prediction and molecular optimization functions is provided, enabling chemists to improve the properties of starting molecules individually.

Kinome-wide polypharmacology profiling of small molecules by multi-task graph isomorphism network approach

Prediction of the interactions between small molecules and their targets play important roles in various applications of drug development, such as lead discovery, drug repurposing and elucidation of potential drug side effects. Therefore, a variety of machine learning-based models have been developed to predict these interactions. In this study, a model called auxiliary multi-task graph isomorphism network with uncertainty weighting (AMGU) was developed to predict the inhibitory activities of small molecules against 204 different kinases based on the multi-task Graph Isomorphism Network (MT-GIN) with the auxiliary learning and uncertainty weighting strategy. The calculation results illustrate that the AMGU model outperformed the descriptor-based models and state-of-the-art graph neural networks (GNN) models on the internal test set. Furthermore, it also exhibited much better performance on two external test sets, suggesting that the AMGU model has enhanced generalizability due to its great transfer learning capacity. Then, a naïve model-agnostic interpretable method for GNN called edges masking was devised to explain the underlying predictive mechanisms, and the consistency of the interpretability results for 5 typical epidermal growth factor receptor (EGFR) inhibitors with their structure‒activity relationships could be observed. Finally, a free online web server called KIP was developed to predict the kinome-wide polypharmacology effects of small molecules (http://cadd.zju.edu.cn/kip).

Traditional Machine and Deep Learning for Predicting Toxicity Endpoints

Molecular structure property modeling is an increasingly important tool for predicting compounds with desired properties due to the expensive and resource-intensive nature and the problem of toxicity-related attrition in late phases during drug discovery and development. Lately, the interest for applying deep learning techniques has increased considerably. This investigation compares the traditional physico-chemical descriptor and machine learning-based approaches through autoencoder generated descriptors to two different descriptor-free, Simplified Molecular Input Line Entry System (SMILES) based, deep learning architectures of Bidirectional Encoder Representations from Transformers (BERT) type using the Mondrian aggregated conformal prediction method as overarching framework. The results show for the binary CATMoS non-toxic and very-toxic datasets that for the former, almost equally balanced, dataset all methods perform equally well while for the latter dataset, with an 11-fold difference between the two classes, the MolBERT model based on a large pre-trained network performs somewhat better compared to the rest with high efficiency for both classes (0.93-0.94) as well as high values for sensitivity, specificity and balanced accuracy (0.86-0.87). The descriptor-free, SMILES-based, deep learning BERT architectures seem capable of producing well-balanced predictive models with defined applicability domains. This work also demonstrates that the class imbalance problem is gracefully handled through the use of Mondrian conformal prediction without the use of over- and/or under-sampling, weighting of classes or cost-sensitive methods.

Fusing 2D and 3D molecular graphs as unambiguous molecular descriptors for conformational and chiral stereoisomers

The rapid progress of machine learning (ML) in predicting molecular properties enables high-precision predictions being routinely achieved. However, many ML models, such as conventional molecular graph, cannot differentiate stereoisomers of certain types, particularly conformational and chiral ones that share the same bonding connectivity but differ in spatial arrangement. Here, we designed a hybrid molecular graph network, Chemical Feature Fusion Network (CFFN), to address the issue by integrating planar and stereo information of molecules in an interweaved fashion. The three-dimensional (3D, i.e., stereo) modality guarantees precision and completeness by providing unabridged information, while the two-dimensional (2D, i.e., planar) modality brings in chemical intuitions as prior knowledge for guidance. The zipper-like arrangement of 2D and 3D information processing promotes cooperativity between them, and their synergy is the key to our model's success. Experiments on various molecules or conformational datasets including a special newly created chiral molecule dataset comprised of various configurations and conformations demonstrate the superior performance of CFFN. The advantage of CFFN is even more significant in datasets made of small samples. Ablation experiments confirm that fusing 2D and 3D molecular graphs as unambiguous molecular descriptors can not only effectively distinguish molecules and their conformations, but also achieve more accurate and robust prediction of quantum chemical properties.

Gene-centric multi-omics integration with convolutional encoders for cancer drug response prediction

MOTIVATION

Tumor heterogeneity, including genetic and transcriptomic characteristics, can reduce the efficacy of anticancer pharmacological therapy, resulting in clinical variability in patient response to therapeutic medications. Multi-omics integration can allow in silico models to provide an additional perspective on a biological system.

METHODS

In this study, we propose a gene-centric multi-channel (GCMC) architecture to integrate multi-omics for predicting cancer drug response. GCMC transformed multi-omics profiles into a three-dimensional tensor with an additional dimension for omics types. GCMC's convolutional encoders captures multi-omics profiles for each gene and yields gene-centric features to predict drug responses.

RESULTS

We evaluated GCMC on various datasets, including The Cancer Genome Atlas (TCGA) patients, patient-derived xenografts (PDX) mice models, and the Genomics of Drug Sensitivity in Cancer (GDSC) cell line datasets. GCMC achieved better performance than baseline models, including single-omics models, in more than 75% of 265 drugs from GDSC cell line datasets. Furthermore, as for the clinical applicability of GCMC, it achieved the best performance on TCGA and PDX datasets in terms of both AUPR and AUC. We also analyzed models' capability of integrating multi-omics profiles by measuring the contribution ratio of omics types. GCMC can incorporate multi-omics profiles in various manners to enhance performance for each drug type. These results suggested that GCMC can improve performance and feature extraction capability by integrating multi-omics profiles in a gene-centric manner.

MultiGran-SMILES: multi-granularity SMILES learning for molecular property prediction

MOTIVATION

Extracting useful molecular features is essential for molecular property prediction. Atom-level representation is a common representation of molecules, ignoring the sub-structure or branch information of molecules to some extent; however, it is vice versa for the substring-level representation. Both atom-level and substring-level representations may lose the neighborhood or spatial information of molecules. While molecular graph representation aggregating the neighborhood information of a molecule has a weak ability in expressing the chiral molecules or symmetrical structure. In this article, we aim to make use of the advantages of representations in different granularities simultaneously for molecular property prediction. To this end, we propose a fusion model named MultiGran-SMILES, which integrates the molecular features of atoms, sub-structures and graphs from the input. Compared with the single granularity representation of molecules, our method leverages the advantages of various granularity representations simultaneously and adjusts the contribution of each type of representation adaptively for molecular property prediction.

RESULTS

The experimental results show that our MultiGran-SMILES method achieves state-of-the-art performance on BBBP, LogP, HIV and ClinTox datasets. For the BACE, FDA and Tox21 datasets, the results are comparable with the state-of-the-art models. Moreover, the experimental results show that the gains of our proposed method are bigger for the molecules with obvious functional groups or branches.

AVAILABILITY AND IMPLEMENTATION

The code and data underlying this work are available on GitHub at https://github. com/Jiangjing0122/MultiGran.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Deep learning methods for molecular representation and property prediction

With advances in artificial intelligence (AI) methods, computer-aided drug design (CADD) has developed rapidly in recent years. Effective molecular representation and accurate property prediction are crucial tasks in CADD workflows. In this review, we summarize contemporary applications of deep learning (DL) methods for molecular representation and property prediction. We categorize DL methods according to the format of molecular data (1D, 2D, and 3D). In addition, we discuss some common DL models, such as ensemble learning and transfer learning, and analyze the interpretability methods for these models. We also highlight the challenges and opportunities of DL methods for molecular representation and property prediction.

Multidrug representation learning based on pretraining model and molecular graph for drug interaction and combination prediction

MOTIVATION

Approaches for the diagnosis and treatment of diseases often adopt the multidrug therapy method because it can increase the efficacy or reduce the toxic side effects of drugs. Using different drugs simultaneously may trigger unexpected pharmacological effects. Therefore, efficient identification of drug interactions is essential for the treatment of complex diseases. Currently proposed calculation methods are often limited by the collection of redundant drug features, a small amount of labeled data and low model generalization capabilities. Meanwhile, there is also a lack of unique methods for multidrug representation learning, which makes it more difficult to take full advantage of the originally scarce data.

RESULTS

Inspired by graph models and pretraining models, we integrated a large amount of unlabeled drug molecular graph information and target information, then designed a pretraining framework, MGP-DR (Molecular Graph Pretraining for Drug Representation), specifically for drug pair representation learning. The model uses self-supervised learning strategies to mine the contextual information within and between drug molecules to predict drug-drug interactions and drug combinations. The results achieved promising performance across multiple metrics compared with other state-of-the-art methods. Our MGP-DR model can be used to provide a reliable candidate set for the combined use of multiple drugs.

AVAILABILITY AND IMPLEMENTATION

Code of the model, datasets and results can be downloaded from GitHub (https://github.com/LiangYu-Xidian/MGP-DR).

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Attention-wise masked graph contrastive learning for predicting molecular property

MOTIVATION

Accurate and efficient prediction of the molecular property is one of the fundamental problems in drug research and development. Recent advancements in representation learning have been shown to greatly improve the performance of molecular property prediction. However, due to limited labeled data, supervised learning-based molecular representation algorithms can only search limited chemical space and suffer from poor generalizability.

RESULTS

In this work, we proposed a self-supervised learning method, ATMOL, for molecular representation learning and properties prediction. We developed a novel molecular graph augmentation strategy, referred to as attention-wise graph masking, to generate challenging positive samples for contrastive learning. We adopted the graph attention network as the molecular graph encoder, and leveraged the learned attention weights as masking guidance to generate molecular augmentation graphs. By minimization of the contrastive loss between original graph and augmented graph, our model can capture important molecular structure and higher order semantic information. Extensive experiments showed that our attention-wise graph mask contrastive learning exhibited state-of-the-art performance in a couple of downstream molecular property prediction tasks. We also verified that our model pretrained on larger scale of unlabeled data improved the generalization of learned molecular representation. Moreover, visualization of the attention heatmaps showed meaningful patterns indicative of atoms and atomic groups important to specific molecular property.

Organic Compound Synthetic Accessibility Prediction Based on the Graph Attention Mechanism

Accurate estimation of the synthetic accessibility of small molecules is needed in many phases of drug discovery. Several expert-crafted scoring methods and descriptor-based quantitative structure-activity relationship (QSAR) models have been developed for synthetic accessibility assessment, but their practical applications in drug discovery are still quite limited because of relatively low prediction accuracy and poor model interpretability. In this study, we proposed a data-driven interpretable prediction framework called GASA (Graph Attention-based assessment of Synthetic Accessibility) to evaluate the synthetic accessibility of small molecules by distinguishing compounds to be easy- (ES) or hard-to-synthesize (HS). GASA is a graph neural network (GNN) architecture that makes self-feature deduction by applying an attention mechanism to automatically capture the most important structural features related to synthetic accessibility. The sampling around the hypothetical classification boundary was used to improve the ability of GASA to distinguish structurally similar molecules. GASA was extensively evaluated and compared with two descriptor-based machine learning methods (random forest, RF; eXtreme gradient boosting, XGBoost) and four existing scores (SYBA: SYnthetic Bayesian Accessibility; SCScore: Synthetic Complexity score; RAscore: Retrosynthetic Accessibility score; SAscore: Synthetic Accessibility score). Our analysis demonstrates that GASA achieved remarkable performance in distinguishing similar molecules compared with other methods and had a broader applicability domain. In addition, we show how GASA learns the important features that affect molecular synthetic accessibility by assigning attention weights to different atoms. An online prediction service for GASA was offered at http://cadd.zju.edu.cn/gasa/.

Systematic optimization of host-directed therapeutic targets and preclinical validation of repositioned antiviral drugs

Inhibition of host protein functions using established drugs produces a promising antiviral effect with excellent safety profiles, decreased incidence of resistant variants and favorable balance of costs and risks. Genomic methods have produced a large number of robust host factors, providing candidates for identification of antiviral drug targets. However, there is a lack of global perspectives and systematic prioritization of known virus-targeted host proteins (VTHPs) and drug targets. There is also a need for host-directed repositioned antivirals. Here, we integrated 6140 VTHPs and grouped viral infection modes from a new perspective of enriched pathways of VTHPs. Clarifying the superiority of nonessential membrane and hub VTHPs as potential ideal targets for repositioned antivirals, we proposed 543 candidate VTHPs. We then presented a large-scale drug-virus network (DVN) based on matching these VTHPs and drug targets. We predicted possible indications for 703 approved drugs against 35 viruses and explored their potential as broad-spectrum antivirals. In vitro and in vivo tests validated the efficacy of bosutinib, maraviroc and dextromethorphan against human herpesvirus 1 (HHV-1), hepatitis B virus (HBV) and influenza A virus (IAV). Their drug synergy with clinically used antivirals was evaluated and confirmed. The results proved that low-dose dextromethorphan is better than high-dose in both single and combined treatments. This study provides a comprehensive landscape and optimization strategy for druggable VTHPs, constructing an innovative and potent pipeline to discover novel antiviral host proteins and repositioned drugs, which may facilitate their delivery to clinical application in translational medicine to combat fatal and spreading viral infections.

An efficient curriculum learning-based strategy for molecular graph learning

Computational methods have been widely applied to resolve various core issues in drug discovery, such as molecular property prediction. In recent years, a data-driven computational method-deep learning had achieved a number of impressive successes in various domains. In drug discovery, graph neural networks (GNNs) take molecular graph data as input and learn graph-level representations in non-Euclidean space. An enormous amount of well-performed GNNs have been proposed for molecular graph learning. Meanwhile, efficient use of molecular data during training process, however, has not been paid enough attention. Curriculum learning (CL) is proposed as a training strategy by rearranging training queue based on calculated samples' difficulties, yet the effectiveness of CL method has not been determined in molecular graph learning. In this study, inspired by chemical domain knowledge and task prior information, we proposed a novel CL-based training strategy to improve the training efficiency of molecular graph learning, called CurrMG. Consisting of a difficulty measurer and a training scheduler, CurrMG is designed as a plug-and-play module, which is model-independent and easy-to-use on molecular data. Extensive experiments demonstrated that molecular graph learning models could benefit from CurrMG and gain noticeable improvement on five GNN models and eight molecular property prediction tasks (overall improvement is 4.08%). We further observed CurrMG’s encouraging potential in resource-constrained molecular property prediction. These results indicate that CurrMG can be used as a reliable and efficient training strategy for molecular graph learning.

Availability: The source code is available in https://github.com/gu-yaowen/CurrMG.

Chemformer: a pre-trained transformer for computational chemistry

Transformer models coupled with a simplified molecular line entry system (SMILES) have recently proven to be a powerful combination for solving challenges in cheminformatics. These models, however, are often developed specifically for a single application and can be very resource-intensive to train. In this work we present the Chemformer model—a Transformer-based model which can be quickly applied to both sequence-to-sequence and discriminative cheminformatics tasks. Additionally, we show that self-supervised pre-training can improve performance and significantly speed up convergence on downstream tasks. On direct synthesis and retrosynthesis prediction benchmark datasets we publish state-of-the-art results for top-1 accuracy. We also improve on existing approaches for a molecular optimisation task and show that Chemformer can optimise on multiple discriminative tasks simultaneously. Models, datasets and code will be made available after publication.

Out-of-the-box deep learning prediction of quantum-mechanical partial charges by graph representation and transfer learning

Accurate prediction of atomic partial charges with high-level quantum mechanics (QM) methods suffers from high computational cost. Numerous feature-engineered machine learning (ML)-based predictors with favorable computability and reliability have been developed as alternatives. However, extensive expertise effort was needed for feature engineering of atom chemical environment, which may consequently introduce domain bias. In this study, SuperAtomicCharge, a data-driven deep graph learning framework, was proposed to predict three important types of partial charges (i.e. RESP, DDEC4 and DDEC78) derived from high-level QM calculations based on the structures of molecules. SuperAtomicCharge was designed to simultaneously exploit the 2D and 3D structural information of molecules, which was proved to be an effective way to improve the prediction accuracy of the model. Moreover, a simple transfer learning strategy and a multitask learning strategy based on self-supervised descriptors were also employed to further improve the prediction accuracy of the proposed model. Compared with the latest baselines, including one GNN-based predictor and two ML-based predictors, SuperAtomicCharge showed better performance on all the three external test sets and had better usability and portability. Furthermore, the QM partial charges of new molecules predicted by SuperAtomicCharge can be efficiently used in drug design applications such as structure-based virtual screening, where the predicted RESP and DDEC4 charges of new molecules showed more robust scoring and screening power than the commonly used partial charges. Finally, two tools including an online server (http://cadd.zju.edu.cn/deepchargepredictor) and the source code command lines (https://github.com/zjujdj/SuperAtomicCharge) were developed for the easy access of the SuperAtomicCharge services.