Background: Considering the heterogeneity of proteins across diverse cell types and states, studying protein thermostability at the single-cell level enables a more profound comprehension of cellular function and the mechanisms underlying disease progression. Methods: In this study, we constructed classification and regression models to predict the thermostability difference of homologous protein pairs by integrating implicit features extracted from protein sequences using eight language models, including ProtBERT, AminoBERT, and ProtT5-XL, with explicit sequence features that are manually computed. Results: Our results demonstrate that the fusion of explicit and implicit features significantly enhances prediction performance. In classification tasks, the combination of implicit features extracted by AminoBERT and the optimal explicit feature set achieves an accuracy of 87.1%. In regression tasks, the combination of implicit features extracted by Word2vec and the optimal explicit feature set yields a PCC of 0.864 and a R2 of 0.742, which is better than previously reported results. Discussion: This study reveals the complementary strengths of language models and handcrafted features in predicting protein thermostability. Combining both types of features significantly improves the performance of classification and regression models and helps identify key factors affecting protein stability. However, the study is limited by its reliance on existing datasets, which may reduce its ability to generalize to novel or rare protein families. Conclusion: The integration of implicit and explicit sequence features enables a more comprehensive representation of protein sequences and facilitates the identification of factors influencing the thermostability of orthologous proteins.

Introduction: Alzheimer's disease (AD), a leading cause of dementia, affects millions globally. By 2050, it is expected to impact over 100 million people. Mild cognitive impairment (MCI) is often considered a precursor to AD, but not all MCI patients progress to AD. Therefore, accurately predicting the risk of MCI patients converting to AD is essential. Methods: This study is a cross-sectional study analyzing routine blood test data collected from AD and MCI patients in Hong Kong between 2000 and 2019. To reduce gender and age bias, subjects were divided into four groups. Models were trained using machine learning and routine blood markers. Results: On the independent test set, the model for females aged 65–74 performed best with an AUC of 0.76. For other age groups, the AUCs were as follows: 0.65 for males aged 65–74, 0.66 for females aged 75–89, and 0.67 for males aged 75–89. Based on this, we developed a platform named MAP (http://lab.malab.cn/~lijing/MAP.html) to predict the risk of MCI converting to AD, assisting clinicians and MCI patients in early diagnosis and prevention. Discussion: Routine blood markers combined with machine learning offer a practical, non-invasive approach for predicting the risk of MCI-to-AD conversion. Predictive performance varies by age and gender. Conclusion: This study supports the use of blood-based machine learning models as cost-effective tools for early AD risk screening in MCI patients.

Background: Protein secondary structure prediction is an important task in bioinformatics and structural biology. Protein’s structure is the basis for its corresponding function. Experimental methods for determining the tertiary structure of proteins are both costly and time-consuming. Since the tertiary structure of proteins is further formed by secondary structure, leveraging computational approaches for efficient prediction of protein secondary structure is important. Both local and global interactions between amino acids affect the prediction results. Objective: We propose a module aimed at processing sequence profile features for deep feature extraction and constructing a lightweight network to extract fused features. Methods: To enhance the network’s ability to capture both local and global interactions, we propose an efficient method InConTPSS, which integrates convolution operation with different receptive fields and temporal convolutional networks in the inception architecture. Concurrently, InConTPSS takes into account the issue of distribution imbalance across various states of secondary structures and improves the predictive performance of scarce categories. Results: Experimental results on six benchmark datasets (including CASP12, CASP13, CASP14, CB513, TEST2016, and TEST2018) demonstrate our method achieves state-of-the-art performance with a simpler model on both 3-state and 8-state secondary structure prediction. result: Experimental results on benchmark datasets (including CASP12, CASP13, CASP14, CB513, TEST2016 and TEST2018) demonstrate our method achieves state-of-the-art performance with a simpler model on both 3-state and 8-state secondary structure prediction. Conclusion: Through the combination of the convolutional layer and temporal convolutional network, the inception network structure can effectively process the fused features and improve the prediction results. InConTPSS achieves the most advanced performance in protein secondary structure prediction, and the reasonable use of label-distribution-aware margin loss in our method can effectively improve the prediction accuracy of scarce secondary structures.

Introduction: Alzheimer's disease (AD) is still one of the most challenging neurodegenerative disorders with no available therapeutics. The identification of drug targets is important for effective treatment development. This research introduces an innovative and improved deep learning prediction framework, BiLSTM-AD, aimed toward therapeutic target indication in crucial, complicated illnesses, particularly Alzheimer's disease. Methods: This paper proposes a deep learning framework, BiLSTM-AD, to predict drug-target indications for Alzheimer's Disease using Protein-Protein Interaction (PPI) datasets from String-DB and InTAct databases. The model utilizes Dual Mode Self-Attention to better capture local and global dependencies in the PPI data. This attention mechanism enables the model to learn from the most significant interactions, which has resulted in better prediction accuracy. Results: The BiLSTM-AD model is evaluated with Precision, Recall, and F1-Score, among other metrics, and compared to reactions-based baseline models such as ANN, RNN, CNN, LSTM, and GNN of the same architecture. Testing on PPI datasets, compared to those baseline methods, BiLSTM-AD has achieved 96% and 97.3% prediction accuracy and provides a more robust solution to infer AD-associated targets for potential drugs. Discussion: The model's self-attention mechanism integrates max and average pooling to capture and emphasize both dominating local interactions and global context. Using a dual perspective greatly enhances the model's ability to discover possible therapeutic targets in important proteinprotein interactions. Precision, Recall, F1-Score, AUC, AUPR, and MSE show that the BiLSTMAD model outperforms other benchmark models in prediction accuracy. With 96% accuracy and 0.93 AUC, our technique surpasses strong baselines (CNN, LSTM, GNN). Conclusion: Integrating the BiLSTM and dual mode self-attention mechanism is an especially effective approach to improve the model performance and enables a practical focus on interaction among drug candidates.

Introduction: Microorganisms are widespread and essential to the transformation of substances and organic matter. Researchers studied microorganisms through various conventional methods, such as machine learning (ML), to overcome multiple obstacles. This review aims to highlight the involvement of ML in various aspects of microbiology to provide insightful information, along with advancement challenges. Methods: Concerning the microbiological aspects and the integration of ML and their associated applications, the relevant literature was diligently reviewed to collect meaningful information on the ML involvement in different fields of microbiology and discussed. Results: Due to the complexity of microbiological data, the researchers are using the amalgamation of various stages and diverse ML applications to deal with and organize the data systematics for accurate results and proper hypotheses. Subsequently, navigating these microbiological data requires an extensive feature-based model for the appropriate validation and to obtain accurate results. Discussion: This study mainly summarizes the various applications and development of ML models used in many aspects, especially the fundamentals of ML in microbiological data, clinical applications, microbial ecology, and the surrounding environment. At present, ML's involvement in microbial aspects is widely utilized; however, bulk data and proper information are needed for accurate and informative outcomes. Conclusion: This review sheds light on ML's involvement in microbiological aspects, and briefly discusses the different aspects. The advanced approaches followed by different tools and databases can be a potential lead toward significant research and promising findings.

Background: Genome assembly tools are used to reconstruct genomic sequences from raw sequencing data, which are then used for identifying the organisms present in a metagenomic sample. Methodology: More recently, machine learning approaches have been applied to a variety of bioinformatics problems, and in this paper, we explore their use for organism identification. We start by evaluating several commonly used metagenomic assembly tools, including PhyloFlash, MEGAHIT, MetaSPAdes, Kraken2, Mothur, UniCycler, and PathRacer, and compare them against state-of-theart deep learning-based machine learning classification approaches represented by DNABERT and DeLUCS, in the context of two synthetic mock community datasets. Results: Our analysis focuses on determining whether ensembling metagenome assembly tools with machine learning tools have the potential to improve identification performance relative to using the tools individually. Conclusion: We find that this is indeed the case, and analyze the level of effectiveness of potential tool ensembling for organisms with different characteristics (based on factors such as repetitiveness, genome size, and GC content).

Introduction: Breast carcinoma continues to be a predominant factor contributing to cancer- associated mortality in women across the globe. Despite the significant advancements in medical technology today, there remain challenges in accurately stratifying patients based on their risk profiles and identifying the most effective treatment strategies for breast cancer. The regulation of metabolism and transcription factors is considered to have a close association with cancer progression. Objective: Objective: In this study, the co-expression network was utilized to identify transcription factors associated with metabolic molecule subtypes, and ultimately, a risk scoring model was constructed. The two groups of patients demonstrated markedly different clinical outcomes. Method: The ssGSEA is utilized for the enrichment of metabolic pathways. Additionally, WGCNA is employed to explore related transcription factor modules, and the VIPER method is used to infer the state of transcription factors. A machine learning methodology SVM has been employed to model patient survival outcomes. Result: We found that patients with lower risk scores exhibit extended survival durations and chemotherapy response in comparison to their high-risk counterparts. Meanwhile, high-risk patients exhibited higher levels of chromosomal instability and tumor immunogenicity relative to low-risk patients. In addition, we constructed a ceRNA network and successfully identified 39 master regulators associated with survival. Conclusion: We achieved risk stratification of breast cancer patients and accurately predicted their prognosis. Additionally, based on the differences observed between the two subgroups, the result highlighted various contributors impacting the clinical prognosis of breast cancer patients.

The interaction between circular RNA (circRNA) and RNA binding protein (RBP) plays an important biological role in the occurrence and development of various diseases. Highthroughput biological experimental methods such as CLIP-seq can effectively analyze the interaction between the two, but biological experiments are inefficient and expensive, and they can only capture binding sites of a specific RBP on circRNA in a selected cell environment at a time. These biological experiments still rely on downstream data analysis to understand the mechanisms behind many biological structures and physiological processes. However, the rapid growth of experimental data dimensions and production speed pose challenges to traditional analysis methods. In recent years, deep learning has made great progress in the genome and transcriptome, and some deep learning prediction algorithms for RBP binding sites on circRNA have also emerged. In this paper, we briefly introduce some biological background knowledge related to circRNA-RBP interaction; present relevant deep learning techniques in this field, including the problem formulation, data source, sequence encoding, deep learning model and overall process of RBP binding sites prediction on circRNA; deeply analyze the current deep learning methods. Finally, some problems existing in the current research and the direction of future research are discussed. It is hoped to help researchers without basic knowledge of deep learning or basic biological background quickly understand the RBP binding sites prediction on circRNA.

Background: Endometriosis is a debilitating gynecological disorder characterized by chronic pain, infertility, and the growth of endometrial tissue outside the uterus. Accurate and early detection of this condition is crucial for effective management and treatment. Methods: We developed a gene rank matrix-based model to integrate endometriosis cohorts across multiple platforms. After removing batch effects, we identified 83 genes associated with endometriosis and further refined a diagnostic model using 11 of these genes. The model was trained on two platforms and validated on two others using SVM, Random Forest, Logistic Regression, and gradient-boosting machine learning algorithms. Results: The integration via the gene rank matrix effectively mitigated batch effects. Utilizing a gradient boosting classifier with a subset of 11 genes, the model demonstrated commendable diagnostic efficacy, achieving an Area Under the Curve (AUC) of 0.77, an accuracy of 0.72, and an F1 score of 0.72 for the training dataset. When subjected to validation, the model maintained its performance, yielding an AUC of 0.769, an accuracy of 0.719, and an F1 score of 0.732. These 11 genes were found to be associated with immunosuppression. Conclusion: Our approach to integrating gene rank matrices effectively consolidates endometriosis data across diverse platforms. The diagnostic model, harnessing the predictive power of 11 specific genes, surpasses alternative models, thereby offering promising prospects for aiding clinical diagnosis of endometriosis. Further validation is imperative to elucidate the functional significance of these 11 genes. Our study underscores the potential of data integration coupled with machine learning techniques in advancing the diagnosis of intricate diseases, such as endometriosis.

Background: Hepatocellular carcinoma (HCC) is a highly heterogeneous malignant tumor characterized by elevated mortality rates and poor diagnostic outcomes. Objective: Accurate identification of cancer subtypes is crucial for guiding personalized treatment and improving patient prognosis. Methods: We present a method for precisely identifying HCC subtypes by integrating multi-omics data. By integrating the GRACES dimensionality reduction and hMKL subtype identification models to analyze 266 HCC patients, we more accurately identified two subtypes significantly associated with overall survival. Results: Our findings indicate that the mortality rates for the high-risk liver cancer subtype and the low-risk liver cancer subtype were 55.9% and 27.9%, respectively, over a three-year period. Pathway activity analysis revealed significant differences in the activity of five pathways between the two subtypes. Additionally, immunocell infiltration analysis demonstrated notable variations in the abundance and status of seven types of immune cells across the subtypes. Further exploration of protein- protein interaction (PPI) networks and centrality indicators led to the identification of 13 upregulated and 14 down-regulated hub genes. Conclusion: There is limited research on ENPP3 (up-regulated) and C3 (down-regulated) in the context of HCC, suggesting their potential as biomarkers. Additionally, low expression levels of PIK3R1, KDR, and CYP3A5, along with high expression levels of EGLN3 and EPO, may indicate a higher risk of liver cancer in patients. Single-gene survival analysis highlighted the significant impact of highly expressed genes on HCC prognosis, with PKM, RRM2, and EPO in the risk score playing crucial roles. conclusion: There is limited research on ENPP3 (up-regulated) and C3 (down-regulated) in the context of HCC, suggesting their potential as biomarkers. Additionally, low expression levels of PIK3R1, KDR, and CYP3A5, along with high expression levels of EGLN3 and EPO, may indicate a higher risk of liver cancer in patients. Single-gene survival analysis highlighted the significant impact of highly expressed genes on HCC prognosis, with PKM, RRM2, and EPO in the risk score playing crucial roles.