Briefings in Bioinformatics

BackgroundSpatial transcriptomics (ST) enables high-resolution mapping of gene expression within tissue slices, providing detailed insights into tissue architecture and cellular interactions. However, batch effects, arising from non-biological variations in sample collection, processing, sequencing platforms, or experimental protocols, can obscure biological signals, hinder data integration, and impact downstream analyses. Despite their critical impact, batch effects in ST datasets remain poorly defined and insufficiently explored. To address this gap, we propose a framework to categorize and define batch effects in ST and systematically evaluate the performance of ST methods with batch effect correction capabilities. ResultsWe categorized batch effects in ST into four types based on their sources: (1) Inter-slice, (2) Inter-sample, (3) Cross-protocol/platform, and (4) Intra-slice. Seven ST integration methods--DeepST, STAligner, GraphST, STitch3D, PRECAST, spatiAlign, and SPIRAL--were evaluated on benchmark datasets from human and mouse tissues. Using metrics such as graph connectivity, kBET, ASW, and iLISI, we assessed both the preservation of biological neighborhoods and the effectiveness of these methods in batch correction. Additionally, we applied STAligner for downstream analysis to compare results before and after batch correction, further highlighting the importance of batch effect correction in ST analysis. ConclusionNo single method is universally optimal. GraphST, PRECAST, SPIRAL, and STAligner performed well for same-platform integration, whereas SPIRAL and STAligner excelled in cross-platform settings. These findings highlight the need for robust and generalizable ST approaches with effective batch correction capabilities to facilitate the integration of multi-platform ST datasets in future research.

The prediction of protein contact map needs enough normalized number of effective sequence (Nf) in multiple sequence alignment (MSA). When Nf is small, the predicted contact maps are often not satisfactory. To solve this problem, we randomly selected a small part of sequence homologs for proteins with large Nf to generate MSAs with small Nf. From these MSAs, input features were generated and were passed through a consistency learning network, aiming to get the same results when using the features generated from the MSA with large Nf. The results showed that this method effectively improves the prediction accuracy of protein contact maps with small Nf.

Functional peptides are one kind of short protein fragments that have a wide range of beneficial functions for living organisms. The majority of previous research focused on mono-functional peptides, but a growing number of multi-functional peptides have been discovered. Although enormous experimental efforts endeavor to assay multi-functional peptides, only a small fraction of millions of known peptides have been explored. Effective and precise techniques for identifying multi-functional peptides can facilitate their discovery and mechanistic understanding. In this article, we presented a novel method, called iMFP-LG, for identifying multi-functional peptides based on protein language models (pLMs) and graph attention networks (GATs). Comparison results showed iMFP-LG significantly outperforms state-of-the-art methods on both multifunctional bioactive peptides and multi-functional therapeutic peptides datasets. The interpretability of iMFP-LG was also illustrated by visualizing attention patterns in pLMs and GATs. Regarding to the outstanding performance of iMFP-LG on the identification of multi-functional peptides, we employed iMFP-LG to screen novel candidate peptides with both ACP and AMP functions from millions of known peptides in the UniRef90. As a result, 8 candidate peptides were identified, and 1 candidate that exhibits significant antibacterial and anticancer effect was confirmed through molecular structure alignment and biological experiments. We anticipate iMFP-LG can assist in the discovery of multi-functional peptides and contribute to the advancement of peptide drug design. Availability and implementationThe models and associated code are available at: https://github.com/chen-bioinfo/iMFP-LG. Supplementary informationSupplementary data are available online.

Cell type annotation is critical to understand the cell population heterogeneity in the single-cell RNA sequencing (scRNA-seq) analysis. Due to their fast, precise, and user-friendly advantages, automatic annotation methods are gradually replacing traditional unsupervised clustering approaches in cell type identification practice. However, current supervised annotation tools are easily overfitting, thus favoring large cell populations but failing to learn the information of smaller populations. This drawback will significantly mislead biological analysis, especially when the rare cell types are important. Here, we present scBalance, an integrated sparse neural network framework that leverages the adaptive weight sampling and dropout techniques for the auto-annotation task. Using 20 scRNA-seq datasets with different scales and different imbalance degrees, we systematically validate the strong performance of scBalance for both intra-dataset and inter-dataset annotation tasks. Furthermore, we also demonstrate the scalability of scBalance on identifying rare cell types in million-level datasets by uncovering the immune landscape in bronchoalveolar cells. Up to now, scBalance is the first and only auto-annotation tool that expands scalability to 1.5 million cells dataset. In addition, scBalance also shows a fast and stable speed outperforming commonly used tools across all scales of datasets. We implemented scBalance in a user-friendly manner that can easily interact with Scanpy, which makes scBalance a superior tool in the increasingly important Python-based platform.

Protein-nucleic acid interactions play a fundamental and critical role in a wide range of life activities. Accurate identification of nucleic acid-binding residues helps to understand the intrinsic mechanisms of the interactions. However, the accuracy and interpretability of existing computational methods for recognizing nucleic acid-binding residues need to be further improved. Here, we propose a novel method called GeSite based the domain adaptive protein language model and explainable E(3)-equivariant graph convolution neural network. Prediction results across multiple benchmark test sets demonstrate that GeSite is superior or comparable to state-of-the-art prediction methods. The performance comparison on low structure similarity and newly released test proteins demonstrates the robustness and generalization of the method. Detailed experimental results suggest that the advanced performance of GeSite lies in the well-designed nucleic acid-binding protein adaptive language model. Meanwhile, interpretability analysis exposes the perception of the prediction model on various remote and close functional domains, which is the source of its discernment. The data and source code of GeSite are freely accessible at https://github.com/pengsl-lab/GeSite.

We have entered the multi-omics era, and we can measure cells from different aspects. When dealing with such multi-omics data, the first step is to determine the correspondence among different omics. In other words, we should match data from different spaces corresponding to the same object. This problem is particularly challenging in the single-cell multi-omics scenario because such data are very sparse with extremely high dimensions. Secondly, matched single-cell multi-omics data are rare and hard to collect. Furthermore, due to the limitations of the experimental environment, the data are usually highly noisy. To promote the single-cell multi-omics research, we overcome the above challenges, proposing a novel framework to align and integrate single-cell RNA-seq data and single-cell ATAC-seq data. Our approach can efficiently map the above data with high sparsity and noise from different spaces to a low-dimensional manifold in a unified space, making the downstream alignment and integration straightforward. Compared with the other state-of-the-art methods, our method performs better on both simulated and real single-cell data. On the real data, the performance improvement on accuracy over the previous methods is up to 55.7% regarding scRNA-seq and scATAC-seq data integration. Downstream trajectory inference analysis shows that our tool can transfer the labels from scRNA-seq to scATAC-seq with very high accuracy, which indicates our methods effectiveness.

Analyzing cell types of origin of cell-free RNA can enhance the resolution of liquid biopsies, thereby deepening the understanding of molecular and cellular changes in development and disease processes. Existing deconvolution methods typically rely on meticulously curated gene expression profiles or employ deep neural network with vast and complex solution spaces that are difficult to interpret. These approaches overlook the synergistic and co-expression effects among genes in biological signaling pathways, compromising their generalizability and robustness. we developed Deconformer, a Transformer-based deconvolution model that integrates biological signaling pathways at the embedding stage, to address these issues. Compared to popular methods on multiple datasets, Deconformer demonstrates superior performance and robustness, and is capable of tracking the developmental process of the fetal and placenta. Additionally, pathway-level interpretability of Deconformer offers new insights into crosstalk, dependencies, and other interactions within cell-free RNA pathways, supporting further biological discoveries. We posit that Deconformer represents a significant advancement in the precise analysis of the cell-free transcriptome. It holds the promise of describing disease progression and severity with a new level of accuracy, focusing on the contributions of originating cell types and their pathway dependencies. This model has the potential to catalyze the development of non-invasive diagnostic tools and enhance our understanding of the underlying biology of diseases.

Clustering is one of the most commonly used methods in single-cell RNA sequencing (scRNA-seq) data analysis and other fields of biology. Traditional clustering methods usually use data from a single source as the input (e.g. scRNA-seq data). However, as the data become more and more complex and contain information from multiple sources, a clustering method that could integrate multiple data is required. Here, we present inClust (integrated clustering), a clustering method that integrates information from multiple sources based on variational autoencoder and vector arithmetic in latent space. inClust perform information integration and clustering jointly, meanwhile it could utilize the labeling information from data as regulation information. It is a flexible framework that can accomplish different tasks under different modes, ranging from supervised to unsupervised. We demonstrate the capability of inClust in the tasks of conditional out-of-distribution generation under supervised mode; label transfer under semi-supervised mode and guided clustering mode; spatial domain identification under unsupervised mode. inClust performs well in all tasks, indicating that it is an excellent general framework for clustering and task-related clustering in the era of multi-omics.

AlphaFold2 is expected to be able to predict protein complex structures as long as a multiple sequence alignment (MSA) of the interologs of the target protein-protein interaction (PPI) can be provided. In this study, a simplified phylogeny-based approach was applied to generate the MSA of interologs, which was then used as the input to AlphaFold2 for protein complex structure prediction. Extensively benchmarked this protocol on non-redundant PPI dataset including 107 bacterial PPIs and 442 eukaryotic PPIs, we show complex structures of 79.5% of the bacterial PPIs and 49.8% of the eukaryotic PPIs can be successfully predicted, which yielded significantly better performance than the application of MSA of interologs prepared by two existing approaches. Considering PPIs may not be conserved in species with long evolutionary distances, we further restricted interologs in the MSA to different taxonomic ranks of the species of the target PPI in protein complex structure prediction. We found the success rates can be increased to 87.9% for the bacterial PPIs and 56.3% for the eukaryotic PPIs if interologs in the MSA are restricted to a specific taxonomic rank of the species of each target PPI. Finally, we show the optimal taxonomic ranks for protein complex structure prediction can be selected with the application of the predicted TM-scores of the output models.

Cell-type identification plays a fundamental role in single-cell RNA-Seq analytics. Thanks to the recent success of the contrastive learning paradigm, the accuracy of automatic cell-type identification has also been improved. In this work, we propose a novel contrastive learning-based cell-type identification method, namely GAN-RCL, which conducts hard positive sample selection on real and conditional generative adversarial networks-created synthetic instances to improve the performance of augmentation-free contrastive learning. Experimental results confirm that GAN-RCL successfully outperforms other recently-proposed contrastive learning-based methods and shows the state-of-the-art predictive performance on multiple single-cell RNA-Seq cell-type identification tasks.

Contrastive learning has already been widely used to handle single-cell RNA-Seq data due to its outstanding performance in transforming original data distributions into hypersphere feature spaces. In this work, we conduct a large-scale empirical evaluation to investigate the generative encoder networks that are learned by five different state-of-the-art single-cell RNA-Seq contrastive learning methods. Unlike the conventional discriminative model-based cell-type prediction studies, this work is focused on the performance of contrastive learning-based generative encoder networks in terms of their capacity to estimate the intrinsic distributions of different cell-types - a fundamental property that directly affects the performance of any downstream single-cell RNA-Seq data analytics. The experimental results confirm that supervised contrastive learning-based encoder networks lead to better performance than self-supervised contrastive learning-based encoder networks, and the recently proposed Gaussian noise augmentation-based single-cell RNA-Seq contrastive learning method shows the best performance on estimating the intrinsic distribution of different cell-types.

The latest developments in spatial transcriptomics technology provide an unprecedented opportunity for in situ elucidation of tissue structure and function. Spatial transcriptomics can provide simultaneous, multi-modal, and complementary information, including gene expression profiles, spatial positions, and histological images. Despite these capabilities, current methodologies often fall short in fully integrating these multi-modal datasets, thereby limiting their ability to fully understand tissue heterogeneity. In this study, we propose XVGAE (cross-view graph autoencoders), a novel approach that integrates gene expression data, spatial coordinates, and histological images to identify spatial domains. XVGAE constructs two distinct graphs: a spatial graph from spatial coordinates and a histological graph from histological images, and these graphs enable XVGAE to learn specific representations for each view and propagate information between them using cross-view graph convolutional networks. The experiments on benchmark datasets of the human dorsolateral prefrontal cortex show demonstrate that the XVGAE could achieve better clustering accuracy than state-of-the-art methods, and further experiments on four real spatial transcriptomics datasets on different sequencing platforms show that the XVGAE could identify biologically meaningful spatial domains with smoother boundary than other methods.

Accurate prediction of protein binding residues (PBRs) from sequence is important for the understanding of cellular activity and helpful for the design of novel drug. However, experimental methods are time-consuming and expensive. In recent years, a lot of computational predictors based on machine learning and deep learning models are proposed to reduce such consumption. But those methods often use MSA tools such as PSI-BLAST or NetSurfP to generate some statistical features and enter them into predictive models as necessary supplementary input. The input generation process normally takes long time, and there is no standard to specify which and how many statistic results should be provided to a prediction model. In addition, prediction of PBRs relies on residue local context, but the most appropriate scale is undetermined. Most works pre-selected certain residue features as input and a scale size based on expertise for certain type of PBRs. In this study, we propose a general tool-free end-to-end framework that can be applied to all types of PBRs, Multi-scale Protein Binding Residues Prediction using language model (MsPBRsP). We adopt a pre-trained language model ProtTrans to save the large consumption caused by MSA tools, and use protein sequence alone as input to our model. To ease scale size uncertainty, we construct multi-size windows in attention layer and multi-size kernels in convolutional layer. We test our framework on various benchmark datasets including PBRs from protein-protein, protein-nucleotide, protein-small ligand, heterodimer, homodimer and antibody-antigen interactions. Compared with existing state-of-the-art methods, MsPBRsP achieves superior performance with less running time and higher prediction rates on every PBRs prediction task. Specifically, we boost F1 score by 27.1% and AUPRC score by 7.6% on NSP448 dataset and decrease running time from over 10 minutes to under 0.1s on average. The source code and datasets are available at https://github.com/biolushuai/MsPBRsP-for-multiple-PBRs-prediction.

In recent years, computational methods for quantifying cell type proportions from transcription data have gained significant attention, particularly those reference-based methods which have demonstrated high accuracy. However, there is currently a lack of comprehensive evaluation and guidance for available reference-based deconvolution methods in cell proportion deconvolution analysis. In this study, we propose a comprehensive evaluation toolkit, called Deconer, specifically designed for reference-based deconvolution methods. Deconer provides various simulated and real gene expression datasets, including both bulk and single-cell sequencing data, and offers multiple visualization interfaces. By utilizing Deconer, we conducted systematic comparisons of 14 reference-based deconvolution methods from different perspectives, including method robustness, accuracy in deconvolving rare components, signature gene selection, and building external reference. We also performed an in-depth analysis of the application scenarios and challenges in cell proportion deconvolution methods. Finally, we provided constructive suggestions for users in selecting and developing cell proportion deconvolution algorithms. This work presents novel insights to researchers, assisting them in choosing appropriate toolkits, applying solutions in clinical contexts, and advancing the development of deconvolution tools tailored to gene expression data.

Natural products produced by microorganisms constitute an important source of essential pharmaceuticals, including antimicrobial and anti-tumor drugs. These bioactive molecules are microbial secondary metabolites synthesized by co-localized genes termed Biosynthetic Gene Clusters (BGCs). The rapid increase of microbial genomics resources, due to the availability of high-throughput sequencing technologies, has spurred the development of computational methods for microbial genome mining for BGC discovery. Current machine learning methods, however, have limited successes in uncovering novel BGCs due to an excessive number of false positives in their predictions. To this end, we propose Deep-BGCpred, a framework that effectively addresses the aforementioned issue by improving a deep learning model termed DeepBGC. The new model embeds multi-source protein family domains and employs a stacked Bidirectional Long Short-Term Memory model to boost accuracy for BGC identifications. In particular, it integrates two customized strategies, sliding window strategy and dual-model serial screening, to improve the models performance stability and reduce the number of false positive in BGC predictions. We compare the proposed model against other well-established methods on common benchmarks and achieve new state-of-the-art results with convincing evidences. We expect that researchers working on genome mining for natural products may be greatly benefited from our newly proposed method, Deep-BGCpred.

B-cell epitopes (BCEs) are a set of specific sites on the surface of an antigen that binds to an antibody produced by B-cell. The recognition of BCEs is a major challenge for drug design and vaccines development. Compared with experimental methods, computational approaches have strong potential for BCEs prediction at much lower cost. Moreover, most of the currently methods focus on using local information around target residue without taking the global information of the whole antigen sequence into consideration. We propose a novel deep leaning method through combing local features and global features for BCEs prediction. In our model, two parallel modules are built to extract local and global features from the antigen separately. For local features, we use Graph Convolutional Networks(GCNs) to capture information of spatial neighbors of a target residue. For global features, Attention-Based Bidirectional Long Short-Term Memory(Att-BLSTM) networks are applied to extract information from the whole antigen sequence. Then the local and global features are combined to predict BCEs. The experiments show that the proposed method achieves superior performance over the state-of-the-art BCEs prediction methods on benchmark datasets. Also, we compare the performance differences between data with or without global features. The experimental results show that global features play an important role in BCEs prediction. Our detailed case study on the BCEs prediction for SARS-Cov-2 receptor binding domain confirms that our method is effective for predicting and clustering true BCEs.

RationaleMany blood-based transcriptional gene signatures for tuberculosis (TB) have been developed with potential use to diagnose disease, predict risk of progression from infection to disease, and monitor TB treatment outcomes. However, an unresolved issue is whether gene set enrichment analysis (GSEA) of the signature transcripts alone is sufficient for prediction and differentiation, or whether it is necessary to use the original statistical model created when the signature was derived. Intra-method comparison is complicated by the unavailability of original training data, missing details about the original trained model, and inadequate publicly-available software tools or source code implementing models. To facilitate these signatures replicability and appropriate utilization in TB research, comprehensive comparisons between gene set scoring methods with cross-data validation of original model implementations are needed. ObjectivesWe compared the performance of 19 TB gene signatures across 24 transcriptomic datasets using both re-rebuilt original models and gene set scoring methods to evaluate whether gene set scoring is a reasonable proxy to the performance of the original trained model. We have provided an open-access software implementation of the original models for all 19 signatures for future use. MethodsWe considered existing gene set scoring and machine learning methods, including ssGSEA, GSVA, PLAGE, Singscore, and Zscore, as alternative approaches to profile gene signature performance. The sample-size-weighted mean area under the curve (AUC) value was computed to measure each signatures performance across datasets. Correlation analysis and Wilcoxon paired tests were used to analyze the performance of enrichment methods with the original models. Measurement and Main ResultsFor many signatures, the predictions from gene set scoring methods were highly correlated and statistically equivalent to the results given by the original diagnostic models. PLAGE outperformed all other gene scoring methods. In some cases, PLAGE outperformed the original models when considering signatures weighted mean AUC values and the AUC results within individual studies. ConclusionGene set enrichment scoring of existing blood-based biomarker gene sets can distinguish patients with active TB disease from latent TB infection and other clinical conditions with equivalent or improved accuracy compared to the original methods and models. These data justify using gene set scoring methods of published TB gene signatures for predicting TB risk and treatment outcomes, especially when original models are difficult to apply or implement.

The development of spatial transcriptomics technologies makes it possible to study tissue heterogeneity at the scale of spatial expressed microenvironment. However, most of the previous methods collapse the spatial patterns in the low spatial resolution. Existing reference based deconvolution methods integrate single-cell reference and spatial transcriptomics data to predict the proportion of cell-types, but the availability of suitable single-cell reference is often limited. In this paper, we propose a novel Transformer based model (TransfromerST) to integrate the spatial gene expression measurements and their spatial patterns in the histology image (if available) without single cell reference. TransfromerST enables the learning of the locally realistic and globally consistent constituents at nearly single cell resolution. TransfromerST firstly uses a transformer based variational autoencoder to explore the latent representation of gene expression, which is further embedded with the spatial relationship learned from adaptive graph Transformer model. The super-resolved cross-scale graph network improves the model-fit to enhanced structure-functional interactions. The public and in-house experimental results with multimodal spatial transcriptomics data demonstrate TransfromerST could highlight the tissue structures at nearly single cell resolution and detect the spatial variable genes and meta gene for each spatial domain. In summary, TransfromerST provides an effective and efficient alternative for spatial transcriptomics tissue clustering, super-resolution and gene expression prediction from histology image.

The rapid advance of spatially resolved transcriptomics technologies has yielded substantial spatial transcriptomics data. Deriving biological insights from these data poses non-trivial computational and analysis challenges, of which the most fundamental step is spatial domain detection (or spatial clustering). Although a number of tools for spatial domain detection have been proposed in recent years, their performance varies across datasets and experimental platforms. It is thus an important task to take full advantage of different tools to get a more accurate and stable result through consensus strategy. In this work, we developed STCC, a novel consensus clustering framework for spatial transcriptomics data that aggregates outcomes from state-of-the-art tools using a variety of consensus strategies, including Onehot-based, Average-based, Hypergraph-based and wNMF-based methods. Comprehensive assessments on simulated and real data from distinct experimental platforms show that consensus clustering significantly improves clustering accuracy over individual methods under varied input parameters. For normal tissue samples exhibiting clear layered structure, consensus clustering by integrating multiple baseline methods leads to improved results. Conversely, when analyzing tumor samples that display scattered cell type distribution patterns, integration of a single baseline method yields satisfactory performance. For consensus strategies, Average-based and Hypergraph-based approaches demonstrated optimal precision and stability. Overall, STCC provides a scalable and practical solution for spatial domain detection in spatial transcriptomic data, laying a solid foundation for future research and applications in spatial transcriptomics.

MotivationIn single cell analyses, cell types are conventionally identified based on known marker gene expressions. Such approaches are time-consuming and irreproducible. Therefore, many new supervised methods have been developed to identify cell types for target datasets using the rapid accumulation of public datasets. However, these approaches are sensitive to batch effects or biological variations since the data distributions are different in cross-platforms or species predictions. ResultsWe developed scAdapt, a virtual adversarial domain adaptation network to transfer cell labels between datasets with batch effects. scAdapt used both the labeled source and unlabeled target data to train an enhanced classifier, and aligned the labeled source centroid and pseudo-labeled target centroid to generate a joint embedding. We demonstrate that scAdapt outperforms existing methods for classification in simulated, cross-platforms, cross-species, and spatial transcriptomic datasets. Further quantitative evaluations and visualizations for the aligned embeddings confirm the superiority in cell mixing and preserving discriminative cluster structure present in the original datasets. Availabilityhttps://github.com/zhoux85/scAdapt. Contactangyd25@mail.sysu.edu.cn or luojinx5@mail.sysu.edu.cn