Current Bioinformatics - Volume 18, Issue 9, 2023

With the rapid advancement in analyzing high-volume and complex data, machine learning has become one of the most critical and essential tools for classification and prediction. This study reviews machine learning (ML) and deep learning (DL) methods for the classification and prediction of biological signals. The effective utilization of the latest technology in numerous applications, along with various challenges and possible solutions, is the main objective of this present study. A PICO-based systematic review is performed to analyze the applications of ML and DL in different biomedical signals, viz. electroencephalogram (EEG), electromyography (EMG), electrocardiogram (ECG), and wrist pulse signal from 2015 to 2022. From this analysis, one can measure machine learning's effectiveness and key characteristics of deep learning. This literature survey finds a clear shift toward deep learning techniques compared to machine learning used in the classification of biomedical signals.

Introduction: Tuberculosis has been studied as a potentially serious bacterial infection affecting the lung caused by Mycobacterium tuberculosis. In addition to its severe impact on health, resistance to existing drugs has also been seen as a rising concern in the space of medicinal solutions. Therapeutic peptides have the potential to complement existing drug designs to provide effective outcomes against Mycobacterium tuberculosis-resistant strains. Methods: This study illustrated a computational approach to design and test peptides against NADHdependent enoyl-acyl carrier protein reductase of Mycobacterium tuberculosis. A human antimicrobial peptide LL-37 was used as a template, and a further 6 peptides were designed, and their binding and interactions against NADH-dependent enoyl-acyl carrier protein reductase were examined. Further, toxicity, immunogenicity, and a broad spectrum of physicochemical properties were calculated to evaluate the therapeutic and safety profile of these peptides. Results: These peptides were structurally modelled and docked with the protein to determine their binding poses and affinity. The molecular interaction of LL-37 with protein was treated as a reference to evaluate the effectiveness of designed peptides. Solvent accessible surface area (SASA) and ΔG binding free energy of docked complexes assisted in the ranking of these peptides. Eventually, peptides P1: LLGDFFRKSKEK, P3: LLFGDRFLLKEK and P7: LLGDFFRLLKEK were selected for 100 ns molecular dynamic simulation as they showed predicted dissociation constants of 8.7×10-4 M, 3.3×10-4 M and 1.2×10-4 M, respectively. These peptides showed direct hydrogen bond formation with ILE21 and LYS165, which are critical active site residues of the protein. The structural variation pattern collected from the MD simulation suggested a strong and stable binding of P3 and P1 with the protein with RMSD 4-5 Å with the starting conformation under the non-fluctuating state. These two peptides showed relatively similar binding results compared with the control peptide LL-37. Comprehensive structural analysis was performed for the middle structures of the most populated cluster generated from 100 ns MD simulation trajectory. Conclusion: Later, MMPBSA binding energies of these structures were computed, where the average binding free energies of P1, P3, and P7 peptides were -146.93 kcal/mole, -161.16 kcal/mole, and - 151.44 kcal/mole, respectively. These energies suggested that P3 is strongly bound to the active site of NADH-dependent enoyl-acyl carrier protein reductase. Overall, this study proposed the application of these peptides as a possible therapeutic solution to inhibit the growth of Mycobacterium tuberculosis.

Background: Current identification of chronic myelogenous leukemia markers tends to mine diagnostic or prognostic biomarkers, ignoring susceptibility markers in normal samples. Objective: We aim to identify possible susceptibility markers for preventing chronic myelogenous leukemia. Methods: Functional links of H3K79me2 patterns and gene expression changes were inferred by correlation analyses. DNase-seq read distribution, transcription factor motifs, and their binding data were acquired via ceasBW and HOMER. Normalized transcription factor binding signals were submitted to a random forest algorithm to predict susceptibility gene expression changes. Three strategies were performed to validate the influence of low H3K79me2 signals on gene expression changes. Results: The gene-body H3K79me2 signals in normal samples were negatively related to gene expression changes during leukemogenesis (ρ=-0.92), regardless of gene lengths and expression levels. Characterization revealed that genes with lower H3K79me2 signals in normal samples have more open environments. Transcription factors GATA3, GATA4, TEAD1, TEAD3, TEAD4, and TRPS1 may induce the upregulation of up-susceptibility genes (ρ=0.95), and ASCL2, IRF4, IRF3, E2A, OCT4, and ZEB2 may mediate the downregulation of down-susceptibility genes (ρ=0.97). Enrichment analysis implied that the screened susceptibility genes were involved in leukemia-related pathways, and about 50% of leukemia stem cell differentially expressed genes were included in these genes. Besides, all hub genes extracted from susceptibility genes were well documented in different leukemia subtypes. Finally, the effect of H3K79me2 signals on gene expression changes were validated in a mouse model and three cell models. Conclusion: Low gene-body H3K79me2 signals in normal samples may serve as susceptibility markers for chronic myelogenous leukemia.

Background: Metabolic chemical reaction is one of the main types of fundamental processes to maintain life. Generally, each reaction needs an enzyme. The metabolic pathway collects a series of chemical reactions at the system level. As compounds and enzymes are two important components in each metabolic pathway, identification of metabolic pathways that a given compound or enzyme can participate is the first important step for understanding the mechanism of metabolic pathways. Objective: The purpose of this study was to build efficient computational methods to predict the metabolic pathways of compounds and enzymes. Methods: Novel multi-label classifiers were proposed to identify metabolic pathway types, reported in KEGG, of compounds and enzymes. Three heterogeneous networks defining compounds and enzymes as nodes were constructed. To extract more informative features of compounds and enzymes, we generalized the powerful network embedding algorithm, Mashup, to its heterogeneous network version, named MashupH. RAndom k-labELsets (RAKEL) was employed to build the classifiers and support vector machine or random forest was selected as the base classification algorithm. Results: The 10-fold cross-validation results indicated the good performance of the proposed classifiers and such performance was superior to the previous classifier that adopted features yielded by Mashup. Furthermore, some key parameters of MashupH that might contribute to or influence the classifiers were analyzed. Conclusion: The features yielded by MashupH were more informative than those produced by Mashup on heterogeneous networks. This was the main reason the new classifiers were superior to those using features yielded by Mashup.

Background: COVID-19 is still widespread due to the rapidly mutating disposition of the virus, rendering vaccines and previously elicited antibodies ineffective in many cases. The integral membrane Envelope (E) protein which is 75 amino acid residues long, has also acquired several mutations. Objective: In this work, we have adopted a high-throughput approach incorporating patient gene expression patterns to identify drug repurposing candidates for COVID-19. We have come up with a list of FDA-approved drugs that can not only prevent E protein oligomerization in both its wild type and a mutational state but can also regulate gene targets responsible for inducing COVID symptoms. Methods: We performed an exhaustive analysis of the available gene expression profiles corresponding to a spectrum of COVID patient samples, followed by drug-gene interaction mapping. This revealed a set of drugs that underwent further efficacy tests through in silico molecular docking with the wild-type E-protein. We also built the molecular models of mutant E-protein by considering the important non-synonymous mutations affecting E-protein structure to check the activities of the screened set of drugs against the mutated E-protein. Finally, blind molecular docking simulations were performed to obtain unbiased docking results. Results: Interestingly, this work revealed a set of 8 drugs that have the potential to be effective for a wider spectrum of asymptomatic to severely symptomatic COVID patients. Conclusion: The varied stages of infection and rapid rate of mutation motivated us to search for a set of drugs that can be effective for a wider spectrum of asymptomatic to severely symptomatic COVID patients. Further, the efficiency of these drugs against mutated E-protein increases another level of confidence to fight against this rapidly changing deadly RNA virus and subsequently needs to be validated in clinical settings.

Introduction: The prevalence of childhood obesity has been increasing in recent decades, and epigenetics is a great process to detect the relationship between children’s obesity and their mothers’ obesity. To investigate the epigenetic connections of obesity between mother and child, we analyzed the saliva DNA methylation profiles from 96 mother-child families. The BMI of both mother and child was measured. Methods: MCFS (Monte Carlo Feature Selection) and IFS (Incremental Feature Selection) methods were used to select the obesity prediction biomarkers. MCFS analysis indicated that if the child's BMI was greater than 17.46, the mother was very likely to be obese. In other words, the obesity of child and mother were highly connected. 17 obesity marker probes corresponding to 18 genes: ADGRA1, CRYBA2, SRRM4, VIPR2, GRIK2, SLC27A1, CLUHP3, THNSL2, F10, PLEC, HTR3C, ESRRG, PTPRM, ANKRD11, ZFAND2A, RTN2/PPM1N, TEX101, were selected. Most of them were found to be related to obesity in literature. Results: The results showed whether mothers are obese can be concluded through their children's BMI and methylation patterns. They can help understand the molecular mechanism of obesity. Conclusion: Epigenetics is a great indicator of obesity. Our results suggested that the obesity status between child and mother was highly correlated. Obesity-related epigenetics changes from the mother remained in the DNA methylation profile of the child's salivary. DNA methylation can partially reflect the living environment and lifestyles.