Precise prediction of perturbation responses is essential in systems biology research, as it plays a pivotal role in characterizing cellular identities and elucidating the regulatory mechanisms of biological pathways. Existing perturbation-responses prediction approaches are predominantly confined to single-modality transcriptomic data, limiting their capacity to capture cross-layer molecular effects. Here, we present MultiPert, a deep learning framework specifically designed for predicting perturbation responses in single-cell multi-omics data. MultiPert employs modality-specific encoders with dedicated pretraining, integrates perturbation through a dual-attention mechanism, and achieves cross-modal alignment via adversarial training. Benchmarking on human THP-1 and kidney multi-omics datasets demonstrates that MultiPert reliably predicts both perturbed gene expression and protein abundance profiles, achieving superior accuracy and stability compared to state-of-the-art strategies. MultiPert generalizes to unseen perturbations and uncovers regulatory mechanisms of immune checkpoint molecules based on perturbed proteomic predictions. In addition, enrichment analyses of perturbed transcriptomic predictions reveal immune-related pathways. By providing an integrated and interpretable framework, MultiPert expands the scope of perturbation modeling at the multi-omics level, thereby offering a robust methodological foundation for comprehensive research into pathogenesis and drug discovery.

The emergence of single-cell multi-omics sequencing technology has enabled the simultaneous profiling of diverse omics data within individual cells. It offers a more comprehensive perspective on cellular phenotypes and heterogeneity. However, single-cell multi-omics data are inherently high-dimensional and heterogeneous. Due to technical limitations and scarce starting materials, the data are often affected by noise and dropout effects. To address these challenges, we propose a novel multitask driven multi-level dynamical fusion algorithm for single-cell multi-omics cell type annotation, named scMMDyn. Our approach incorporates reconstruction and classification auxiliary tasks to guide the training of trustworthy modules at both the feature and modality levels. It executes dynamical fusion during these stages and finally achieves cross-modality fusion via an attention mechanism. This method effectively mitigates data quality issues through reconstruction tasks and feature-level dynamical fusion while providing interpretability at both feature and modality levels. Experimental results across diverse single-cell multi-omics datasets show that our method surpasses existing approaches in cell type annotation.

The advent of single-cell transcriptomics has revolutionized our ability to analyze cellular heterogeneity and dynamics at a fine resolution, yet covering the vast array of potential perturbations remains challenging due to biological variability. To address this, we propose scCADE, a novel computational approach utilizing contrastive learning and an attention mechanism to decouple gene expression signatures and predict cellular responses to perturbations. scCADE excels in predicting responses in cells to perturbations observed in other cells but not yet seen in the target cells. Through rigorous ablation studies and validation across three datasets involving drug and gene editing perturbations, scCADE consistently outperformed existing methods, underscoring its efficacy and potential to advance genomics and personalized medicine by accurately forecasting responses to novel perturbations.

Transcription factors (TFs) regulation is required for the vast majority of biological processes in living organisms. Some diseases may be caused by improper transcriptional regulation. Identifying the target genes of TFs is thus critical for understanding cellular processes and analyzing disease molecular mechanisms. Computational approaches can be challenging to employ when attempting to predict potential interactions between TFs and target genes. In this paper, we present a novel graph model (PPRTGI) for detecting TF-target gene interactions using DNA sequence features. Feature representations of TFs and target genes are extracted from sequence embeddings and biological associations. Then, by combining the aggregated node feature with graph structure, PPRTGI uses a graph neural network with personalized PageRank to learn interaction patterns. Finally, a bilinear decoder is applied to predict interaction scores between TF and target gene nodes. We designed experiments on six datasets from different species. The experimental results show that PPRTGI is effective in regulatory interaction inference, with our proposed model achieving an area under receiver operating characteristic score of 93.87% and an area under precision-recall curves score of 88.79% on the human dataset. This paper proposes a new method for predicting TF-target gene interactions, which provides new insights into modeling molecular networks and can thus be used to gain a better understanding of complex biological systems.

Brain tumor segmentation is pivotal in the diagnosis and treatment of brain tumors. As functional imaging technologies like CT and MR advance, analyzing 3D medical image data becomes more time-consuming. Several challenges exist in 3D medical image segmentation: 1) 2D networks, when applied to 3D segmentation tasks, suffer from a lack of 3D structural information. 2) Pure 3D networks, due to their vast parameter count and smaller training sample, are susceptible to overfitting. 3) Current 2.5D networks do not fully leverage the available 3D structural information. In this study, we introduce the Mix-UNet, a multi-branch network that synergizes 2D and 3D networks. This design preserves essential 3D structural details for precise segmentation while ensuring computational efficiency. Our model comprises two main branches and a fusion module: a 2D branch for coarse segmentation without 3D structural information, a 3D branch to capture comprehensive 3D structural details, and a fusion module for pixel-level integration to produce the final segmentation. Experimental results demonstrate the model’s ability to reduce parameter count, increase robustness, and maintain high precision. When tested on the BraTS 2020 validation dataset, our model achieved mean dice coefficients of 90.4%, 80.7%, and 71.2% for the whole tumor, tumor core, and enhancing tumor, respectively, with only 2.2M parameters.

Inferring gene regulatory networks based on single-cell transcriptomes is critical for systematically understanding cell-specific regulatory networks and discovering drug targets in tumor cells. Here we show that existing methods mainly perform co-expression analysis and apply the image-based model to deal with the non-euclidean scRNA-seq data, which may not reasonably handle the dropout problem and not fully take advantage of the validated gene regulatory topology. We propose a graph-based end-to-end deep learning model for GRN inference (GRNInfer) with the help of known regulatory relations through transductive learning. The robustness and superiority of the model are demonstrated by comparative experiments.

Identification of drug-target interaction (DTI) is the most important issue in the broad field of drug discovery. Using purely biological experiments to verify drug-target binding profiles takes lots of time and effort, so computational technologies for this task obviously have great benefits in reducing the drug search space. Most of computational methods to predict DTI are proposed to solve a binary classification problem, which ignore the influence of binding strength. Therefore, drug-target binding affinity prediction is still a challenging issue. Currently, lots of studies only extract sequence information that lacks feature-rich representation, but we consider more spatial features in order to merge various data in drug and target spaces. In this study, we propose a two-stage deep neural network ensemble model for detecting drug-target binding affinity, called DeepFusionDTA, via various information analysis modules. First stage is to utilize sequence and structure information to generate fusion feature map of candidate protein and drug pair through various analysis modules based deep learning. Second stage is to apply bagging-based ensemble learning strategy for regression prediction, and we obtain outstanding results by combining the advantages of various algorithms in efficient feature abstraction and regression calculation. Importantly, we evaluate our novel method, DeepFusionDTA, which delivers 1.5 percent CI increase on KIBA dataset and 1.0 percent increase on Davis dataset, by comparing with existing prediction tools, DeepDTA. Furthermore, the ideas we have offered can be applied to in-silico screening of the interaction space, to provide novel DTIs which can be experimentally pursued. The codes and data are available from https://github.com/guofei-tju/DeepFusionDTA.

Major histocompatibility complex (MHC) possesses important research value in the treatment of complex human diseases. A plethora of computational tools has been developed to predict MHC class I binders. Here, we comprehensively reviewed 27 up-to-date MHC I binding prediction tools developed over the last decade, thoroughly evaluating feature representation methods, prediction algorithms and model training strategies on a benchmark dataset from Immune Epitope Database. A common limitation was identified during the review that all existing tools can only handle a fixed peptide sequence length. To overcome this limitation, we developed a bilateral and variable long short-term memory (BVLSTM)-based approach, named BVLSTM-MHC. It is the first variable-length MHC class I binding predictor. In comparison to the 10 mainstream prediction tools on an independent validation dataset, BVLSTM-MHC achieved the best performance in six out of eight evaluated metrics. A web server based on the BVLSTM-MHC model was developed to enable accurate and efficient MHC class I binder prediction in human, mouse, macaque and chimpanzee.

Quantifying DNA properties is a challenging task in the broad field of human genomics. Since the vast majority of non-coding DNA is still poorly understood in terms of function, this task is particularly important to have enormous benefit for biology research. Various DNA sequences should have a great variety of representations, and specific functions may focus on corresponding features in the front part of learning model. Currently, however, for multi-class prediction of non-coding DNA regulatory functions, most powerful predictive models do not have appropriate feature extraction and selection approaches for specific functional effects, so that it is difficult to gain a better insight into their internal correlations. Hence, we design a category attention layer and category dense layer in order to select efficient features and distinguish different DNA functions. In this study, we propose a hybrid deep neural network method, called DeepATT, for identifying $919$ regulatory functions on nearly $5$ million DNA sequences. Our model has four built-in neural network constructions: convolution layer captures regulatory motifs, recurrent layer captures a regulatory grammar, category attention layer selects corresponding valid features for different functions and category dense layer classifies predictive labels with selected features of regulatory functions. Importantly, we compare our novel method, DeepATT, with existing outstanding prediction tools, DeepSEA and DanQ. DeepATT performs significantly better than other existing tools for identifying DNA functions, at least increasing $1.6\%$ area under precision recall. Furthermore, we can mine the important correlation among different DNA functions according to the category attention module. Moreover, our novel model can greatly reduce the number of parameters by the mechanism of attention and locally connected, on the basis of ensuring accuracy.

An enhancer is a short region of DNA with the ability to recruit transcription factors and their complexes, increasing the likelihood of the transcription of a particular gene. Considering the importance of enhancers, enhancer identification is a prevailing problem in computational biology. In this paper, we propose a novel two-layer enhancer predictor called iEnhancer-KL, using computational biology algorithms to identify enhancers and then classify these enhancers into strong or weak types. Kullback-Leibler (KL) divergence is creatively taken into consideration to improve the feature extraction method PSTNPss. Then, LASSO is used to reduce the dimension of features and finally helps to get better prediction performance. Furthermore, the selected features are tested on several machine learning models, and the SVM algorithm achieves the best performance. The rigorous cross-validation indicates that our predictor is remarkably superior to the existing state-of-the-art methods with an Acc of 84.23 percent and the MCC of 0.6849 for identifying enhancers. Our code and results can be freely downloaded from https://github.com/Not-so-middle/iEnhancer-KL.git.

Auotmatic melanoma classification in dermoscopy images is a very important task, which can help improve diagnostic accuracy and reduce mortality. Deep convolutional neural network (DCNN) has developed rapidly in recent years, but it is still a challenging task due to the intra-class variation and inter-class similarity of melanoma. We proposed a novel neural network integration model, which is composed of three parts: First, we use U-net segmentation network to generate masks and use the masks to crop original images; Second, we use five state-of-the-art DCNNs to extract features of cropped images, and add the squeeze-excitation block (SE block) to emphasize useful features; Finally, we construct a new neural network with local connection to integrate the classification results, extract features of different class of results, and integrate the results of each class separately. Local connection can integrate each class separately, maximizing the advantages of different networks in various classes. We evaluate our model on ISIC 2017 challenge dataset, and the result shows that our method has better performance compared with the existing methods.

Antiviral peptides (AVPs) have been experimentally verified to block virus into host cells, which have antiviral activity with decapeptide amide. Therefore, utilization of experimentally validated antiviral peptides is a potential alternative strategy for targeting medically important viruses. In this article, we propose a dual-channel deep neural network ensemble method for analyzing variable-length antiviral peptides. The LSTM channel can capture long-term dependencies for effectively studying original variable-length sequence data. The CONV channel can build dynamic neural network for analyzing the local evolution information. Also, our model can fine-tune the substitution matrix for specifically functional peptides. Applying it to a novel experimentally verified dataset, our AVPs predictor, DeepAVP, demonstrates state-of-the-art performance of [Formula: see text] accuracy and 0.85 MCC, which is far better than existing prediction methods for identifying antiviral peptides. Therefore, DeepAVP, web server for predicting the effective AVPs, would make significantly contributions to peptide-based antiviral research.