BACKGROUND: Predicting immune checkpoint inhibitor (ICI) response remains a significant challenge in cancer immunotherapy. Many existing approaches rely on differential gene expression analysis or predefined immune signatures, which may fail to capture the complex regulatory mechanisms underlying immune response. Network-based models attempt to integrate biological interactions, but they often lack a quantitative framework to assess how individual genes contribute within pathways, limiting the specificity and interpretability of biomarkers. Given these limitations, we developed PathNetDRP, a framework that integrates biological pathways, protein-protein interaction networks, and machine learning to identify functionally relevant biomarkers for ICI response prediction. RESULTS: We introduce PathNetDRP, a novel biomarker discovery approach that applies the PageRank algorithm to prioritize ICI-associated genes, maps them to relevant biological pathways, and calculates PathNetGene scores to quantify their contribution to immune response. Unlike conventional methods that focus solely on gene expression differences, PathNetDRP systematically incorporates biological context to improve biomarker selection. Validation across multiple independent cancer cohorts showed that PathNetDRP achieved strong predictive performance, with cross-validation the area under the receiver operating characteristic curves increasing from 0.780 to 0.940. Interestingly, PathNetDRP did not merely improve predictive accuracy; it also provided insights into key immune-related pathways, reinforcing its potential for identifying clinically relevant biomarkers. CONCLUSION: The biomarkers identified by PathNetDRP demonstrated robust predictive performance across cross-validation and independent validation datasets, suggesting their potential utility in clinical applications. Furthermore, enrichment analysis highlighted key immune-related pathways, providing a deeper understanding of their role in ICI response regulation. While these findings underscore the promise of PathNetDRP, future work will explore the integration of additional predictive features, such as tumor mutational burden and microsatellite instability, to further refine its applicability.

Digitizing molecular structure information of diverse compounds is essential for AI-driven drug discovery. While a significant amount of molecular structure data is already stored and managed in digital formats for in silico predictions of drug stability and side effects, a significant portion remains in analog documents. To address this, deep learning-based optical chemical structure recognition (OCSR) technologies have been developed. Although these models excel with abundant digitally rendered chemical structure images, they underperform on hand-drawn images due to lack of sufficient training data for such images. This study proposes OCSAug, a diffusion model-based data augmentation technique to improve OCSR performance on hand-drawn molecular images. Using a denoising diffusion probabilistic model to capture irregularities and distortions, and the RePaint algorithm to generate augmented data, OCSAug significantly enhances recognition accuracy. Evaluation on the DECIMER dataset shows an improvement of 1.918–3.820 times over results without augmentation.

The accurate prediction of drug responses based on the genomic profile of a patient is essential to progress in the field of precision medicine. The advent of various deep-learning algorithms based on publicly available large-scale omics datasets is the driving force behind research in this field. The characteristics of biological datasets, characterized by high dimensions and low sample sizes, pose challenges of overfitting and limited generalization in prediction models. Additionally, constructing prediction models using biological data such as gene expression is further complicated by the need to account for the complex relationships among genes, which exacerbates the aforementioned challenges. To address these challenges, we propose a drug response prediction framework (DrDiff) that integrates a denoising diffusion probabilistic model (DDPM) based data augmentation module with a graph attention network based drug response prediction module. The proposed model showed a 10% higher AUC than the state-of-the-art models for drug response prediction for the six drugs considered in the study, suggesting the superior generalization performance of DrDiff over other baseline models. Furthermore, we demonstrated the feasibility of generative models, which form one of the modules of the proposed framework, in overcoming the fundamental limitations of omics datasets. Further experiments bear out the feasibility of generative models, which form one of the modules of the proposed framework, in augmenting gene expression data.

Protein-ligand interactions play an essential role in many biological processes, and prior knowledge of ligand binding sites is necessary for successful drug design. Many 3D structure- and sequence-based methods have been proposed for identifying ligand binding sites. The 3D structure-based methods typically achieve better binding site prediction than the sequence-based methods. However, as deep-learning techniques that can extract structural information from large-scale sequence data have been developed, the performance gap between 3D structure- and sequence-based methods is narrowing. Nonetheless, there remains room for improvement in sequence-based prediction. We propose Pseq2Sites, a sequence-based deep-learning model for predicting ligand binding sites. Pseq2Sites comprises a 1D convolutional neural network that extracts local features from the protein sequence, and a position-based attention mechanism that captures long-distance dependencies between binding residues. To verify the effectiveness of the proposed method, we compared it with other state-of-the-art methods using three public datasets: COACH420, HOLO4K, and CSAR-NRC HiQ. Utilizing solely protein sequence information, Pseq2Sites outperformed 3D structure-based state-of-the-art methods on external test datasets; within the COACH420 dataset, Pseq2Sites remarkably identified 97% of the binding pockets (at a significance level δ = 0.5), which was 27% higher than the second highest-performing model. Pseq2Sites also achieved outstanding binding site prediction, even for proteins with low similarity to the training dataset. Our code is available at https://github.com/Blue1993/Pseq2Sites.

Abstract Structure-constrained molecular optimisation aims to improve the target pharmacological properties of input molecules through small perturbations of the molecular structures. Previous studies have exploited various optimisation techniques to satisfy the requirements of structure-constrained molecular optimisation tasks. However, several studies have encountered difficulties in producing property-improved and synthetically feasible molecules. To achieve both property improvement and synthetic feasibility of molecules, we proposed a molecular structure editing model called SELF-EdiT that uses self-referencing embedded strings (SELFIES) and Levenshtein transformer models. The SELF-EdiT generates new molecules that resemble the seed molecule by iteratively applying fragment-based deletion-and-insertion operations to SELFIES. The SELF-EdiT exploits a grammar-based SELFIES tokenization method and the Levenshtein transformer model to efficiently learn deletion-and-insertion operations for editing SELFIES. Our results demonstrated that SELF-EdiT outperformed existing structure-constrained molecular optimisation models by a considerable margin of success and total scores on the two benchmark datasets. Furthermore, we confirmed that the proposed model could improve the pharmacological properties without large perturbations of the molecular structures through edit-path analysis. Moreover, our fragment-based approach significantly relieved the SELFIES collapse problem compared to the existing SELFIES-based model. SELF-EdiT is the first attempt to apply editing operations to the SELFIES to design an effective editing-based optimisation, which can be helpful for fellow researchers planning to utilise the SELFIES.

Deep generative models for molecular generation have accelerated the development of de novo drug design by introducing how to generate novel molecular structures expressed in simplified molecular-input line-entry system (SMILES) or molecular graph formats. Numerous drug design studies have proposed combinations of variational autoencoder (VAE) and autoregressive generators such as recurrent neural networks (RNNs) to generate SMILES strings. However, RNN-VAE has one notorious issue, called posterior collapse, in which different latent vectors produce indistinguishable molecular distributions. In this study, we proposed a Gumbel-Softmax-based generative model, MolBit, and a genetic algorithm-based molecular property optimization method. We confirmed that the proposed model avoided the posterior collapse problem and outperformed the existing drug design models with SMILES.

BACKGROUND: Accurate prediction of protein-ligand binding affinity is important for lowering the overall cost of drug discovery in structure-based drug design. For accurate predictions, many classical scoring functions and machine learning-based methods have been developed. However, these techniques tend to have limitations, mainly resulting from a lack of sufficient energy terms to describe the complex interactions between proteins and ligands. Recent deep-learning techniques can potentially solve this problem. However, the search for more efficient and appropriate deep-learning architectures and methods to represent protein-ligand complex is ongoing. RESULTS: In this study, we proposed a deep-neural network model to improve the prediction accuracy of protein-ligand complex binding affinity. The proposed model has two important features, descriptor embeddings with information on the local structures of a protein-ligand complex and an attention mechanism to highlight important descriptors for binding affinity prediction. The proposed model performed better than existing binding affinity prediction models on most benchmark datasets. CONCLUSIONS: We confirmed that an attention mechanism can capture the binding sites in a protein-ligand complex to improve prediction performance. Our code is available at https://github.com/Blue1993/BAPA .

MOTIVATION: Identification of genes that can be used to predict prognosis in patients with cancer is important in that it can lead to improved therapy, and can also promote our understanding of tumor progression on the molecular level. One of the common but fundamental problems that render identification of prognostic genes and prediction of cancer outcomes difficult is the heterogeneity of patient samples. RESULTS: To reduce the effect of sample heterogeneity, we clustered data samples using K-means algorithm and applied modified PageRank to functional interaction (FI) networks weighted using gene expression values of samples in each cluster. Hub genes among resulting prioritized genes were selected as biomarkers to predict the prognosis of samples. This process outperformed traditional feature selection methods as well as several network-based prognostic gene selection methods when applied to Random Forest. We were able to find many cluster-specific prognostic genes for each dataset. Functional study showed that distinct biological processes were enriched in each cluster, which seems to reflect different aspect of tumor progression or oncogenesis among distinct patient groups. Taken together, these results provide support for the hypothesis that our approach can effectively identify heterogeneous prognostic genes, and these are complementary to each other, improving prediction accuracy. AVAILABILITY AND IMPLEMENTATION: https://github.com/mathcom/CPR. CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.