Journal of Computational Biology最新文献

In scRNA-seq analysis, cell clusters are typically defined by a single round of feature extraction and clustering. This approach may miss phenotypic differences in cell types that are characterized by genes not sufficiently represented in the feature set derived using all cells, such as rare cell types. This work explores an alternative approach, where cell clusters are identified by recursively performing feature extraction and clustering on previously identified clusters, such that each subclustering step uses features that are more specific to distinguishing the higher-resolution subclusters. We benchmark this recursive approach against the conventional, nonrecursive clustering approach and demonstrate that the recursive method results in robust improvement in cell type detection on four scRNA-seq datasets across a wide range of clustering resolution parameters. We apply the recursive approach to cluster scRNA-seq data obtained from patients with Crohn's disease belonging to three clinical phenotypes and observe that recursive clustering captures phenotypic differences only visible at specific levels of granularity within an interpretable hierarchical framework while defining cell clusters within a gene expression feature space more specific to each cluster.

A key step in most viral infections is the binding of a viral protein to a host receptor, leading to the virus entering the host cell. Disrupting this protein-protein interaction is an effective strategy for preventing infection and subsequent disease. Building on recent advances in computational tools for structural biology, we introduce Virus Inhibition via Peptide Engineering and Receptor Mimicry (VIPER), a novel approach for the automatic derivation and optimization of biomimetic decoy peptides that mimic binding sites of human proteins. VIPER leverages structural data from human-pathogen protein complexes, yielding peptides that can competitively inhibit viral entry by mimicking the natural receptor. We computationally validated VIPER using molecular dynamics simulations and showcased its applicability on three clinically relevant viruses, highlighting its potential to accelerate therapeutic development. With a focus on reproducibility and extensibility, VIPER can facilitate the rapid development of antiviral inhibitors by automating the design and optimization of biomimetic compounds.

Maintaining homeostasis, the regulation of internal physiological parameters, is essential for health and well-being. Deviations from optimal levels, or 'sweet spots,' can lead to health deterioration and disease. Identifying biomarkers with sweet spots requires both change-point detection and variance effect analysis. Traditional approaches involve separate tests for change-points and heteroskedasticity, which can yield inaccurate results if model assumptions are violated. To address these challenges, we propose a unified approach: Bayesian Testing for Heteroskedasticity and Sweet Spots (BTHS). This framework integrates sampling-based parameter estimation and Bayes factor computation to enhance change-point detection, heteroskedasticity quantification, and testing in change-point regression settings, and extends previous Bayesian approaches. BTHS eliminates the need for separate analyses and provides detailed insights into both the magnitude and shape of heteroskedasticity, enabling robust identification of sweet spots without strong assumptions. We applied BTHS to blood elements from the Canadian Longitudinal Study on Aging identifying nine blood elements with significant sweet spot variance effects.

Cellular appearance and its dynamics frequently serve as a proxy measurement of live-cell physiological properties. The computational analysis of cell properties is considered to be a significant endeavor in biological and biomedical research. Deep learning has garnered considerable success across various fields. In light of this, various neural networks have been developed to analyze live-cell microscopic videos and capture cellular dynamics with biological significance. Specifically, cellular dynamic grading (CDG) is the task that provides a predefined dynamic grade for a live-cell according to the speed of cellular deformation and intracellular movement. This task involves recording the morphological and cytoplasmic dynamics in live-cell microscopic videos. Similar to other medical image processing tasks, CDG faces challenges in collecting and annotating cellular videos. These deficiencies in medical data limit the performance of deep learning models. In this article, we propose a novel self-supervised framework to overcome these limitations for the CDG task. Our framework relies on the assumption that increasing or decreasing cell dynamic grades is consistent with accelerating or decelerating cell appearance change in videos, respectively. This consistency is subsequently incorporated as a constraint in the loss function for the self-supervised training strategy. Our framework is implemented by formulating a probability transition matrix based on the Earth Mover's Distance and imposing a loss constraint on the elements of this matrix. Experimental results demonstrate that our proposed framework enhances the model's ability to learn spatiotemporal dynamics. Furthermore, our framework outperforms the existing methods on our cell video database.

The development of computational models for the prediction of cardiac cellular dynamics remains a challenge due to the lack of first-principled mathematical models. We develop a novel machine-learning approach hybridizing physics simulation and graph networks to deliver robust predictions of cardiomyocyte dynamics. Embedded with inductive physical priors, the proposed constraint-based interaction neural projection (CINP) algorithm can uncover hidden physical constraints from sparse image data on a small set of beating cardiac cells and provide robust predictions for heterogenous large-scale cell sets. We also implement an in vitro culture and imaging platform for cellular motion and calcium transient analysis to validate the model. We showcase our model's efficacy by predicting complex organoid cellular behaviors in both in silico and in vitro settings.

With the continuous evolution of single-cell RNA sequencing technology, it has become feasible to reconstruct cell development processes using computational methods. Trajectory inference is a crucial downstream analytical task that provides valuable insights into understanding cell cycle and differentiation. During cell development, cells exhibit both stable and transition states, which makes it challenging to accurately identify these cells. To address this challenge, we propose a novel single-cell trajectory inference method using fuzzy clustering, named scFCTI. By introducing fuzzy clustering and quantifying cell uncertainty, scFCTI can identify transition cells within unstable cell states. Moreover, scFCTI can obtain refined cell classification by characterizing different cell stages, which gain more accurate single-cell trajectory reconstruction containing transition paths. To validate the effectiveness of scFCTI, we conduct experiments on five real datasets and four different structure simulation datasets, comparing them with several state-of-the-art trajectory inference methods. The results demonstrate that scFCTI outperforms these methods by successfully identifying unstable cell clusters and obtaining more accurate cell paths with transition states. Especially the experimental results demonstrate that scFCTI can reconstruct the cell trajectory more precisely.

With the rapid advancement of big data and artificial intelligence technologies, the limitations inherent in traditional storage media for accommodating vast amounts of data have become increasingly evident. DNA storage is an innovative approach harnessing DNA and other biomolecules as storage mediums, endowed with superior characteristics including expansive capacity, remarkable density, minimal energy requirements, and unparalleled longevity. Central to the efficient DNA storage is the process of DNA coding, whereby digital information is converted into sequences of DNA bases. A novel encoding method based on adaptive arithmetic coding (AAC) has been introduced, delineating the encoding process into three distinct phases: compression, error correction, and mapping. Prediction by Partial Matching (PPM)-based AAC in the compression phase serves to compress data and enhance storage density. Subsequently, the error correction phase relies on octal Hamming code to rectify errors and safeguard data integrity. The mapping phase employs a "3-2 code" mapping relationship to ensure adherence to biochemical constraints. The proposed method was verified by encoding different formats of files such as text, pictures, and audio. The results indicated that the average coding density of bases can be up to 3.25 per nucleotide, the GC content (which includes guanine [G] and cytosine [C]) can be stabilized at 50% and the homopolymer length is restricted to no more than 2. Simulation experimental results corroborate the method's efficacy in preserving data integrity during both reading and writing operations, augmenting storage density, and exhibiting robust error correction capabilities.

Developing a new drug is a long and expensive process that typically takes 10-15 years and costs billions of dollars. This has led to an increasing interest in drug repositioning, which involves finding new therapeutic uses for existing drugs. Computational methods become an increasingly important tool for identifying associations between drugs and new diseases. Graph- and hypergraph-based approaches are a type of computational method that can be used to identify potential associations between drugs and new diseases. Here, we present a drug repurposing method based on hypergraph neural network for predicting drug-disease association in three stages. First, it constructs a heterogeneous graph that contains drug and disease nodes and links between them; in the second stage, it converts the heterogeneous simple graph to a hypergraph with only disease nodes. This is achieved by grouping diseases that use the same drug into a hyperedge. Indeed, all the diseases that are the common therapeutic goal of a drug are placed on a hyperedge. Finally, a graph neural network is used to predict drug-disease association based on the structure of the hypergraph. This model is more efficient than other methods because it uses a hypergraph to model relationships more effectively than graphs. Furthermore, it constructs the hypergraph using only a drug-disease association matrix, eliminating the need for extensive amounts of data. Experimental results show that the hypergraph-based approach effectively captures complex interrelationships between drugs and diseases, leading to improved accuracy of drug-disease association prediction compared to state-of-the-art methods.

Accurately predicting drug response depending on a patient's genomic profile is critical for advancing personalized medicine. Deep learning approaches rise and especially the rise of graph neural networks leveraging large-scale omics datasets have been a key driver of research in this area. However, these biological datasets, which are typically high dimensional but have small sample sizes, present challenges such as overfitting and poor generalization in predictive models. As a complicating matter, gene expression (GE) data must capture complex inter-gene relationships, exacerbating these issues. In this article, we tackle these challenges by introducing a drug response prediction method, called drug response graph attention network (DRGAT), which combines a denoising diffusion implicit model for data augmentation with a recently introduced graph attention network (GAT) with high-order neighbor propagation (HO-GATs) prediction module. Our proposed approach achieved almost 5% improvement in the area under receiver operating characteristic curve compared with state-of-the-art models for the many studied drugs, indicating our method's reasonable generalization capabilities. Moreover, our experiments confirm the potential of diffusion-based generative models, a core component of our method, to mitigate the inherent limitations of omics datasets by effectively augmenting GE data.

Pseudoprogression (PSP) is a related reaction of glioblastoma treatment, and misdiagnosis can lead to unnecessary intervention. Magnetic resonance imaging (MRI) provides cross-modality images for PSP prediction studies. However, how to effectively use the complementary information between the cross-modality MRI to improve PSP prediction is still a challenging task. To address this challenge, we propose a cross-modality feature interaction network for PSP prediction. Firstly, we propose a triple-branch multi-scale module to extract low-order feature representations and a skip-connection multi-scale module to extract high-order feature representations. Then, a cross-modality interaction module based on attention mechanism is designed to make the complementary information between cross-modality MRI fully interact. Finally, the high-order cross-modality interaction information is fed into a multi-layer perceptron to achieve the PSP prediction task. We evaluate the proposed network on a private dataset with 52 subjects from Hunan Cancer Hospital and validate it on a private dataset with 30 subjects from Xiangya Hospital. The accuracy of our proposed network on the datasets is 0.954 and 0.929, respectively, which is better than most typical convolutional neural network and interaction methods.