Search

An imbalanced classification problem occurs when the distribution of samples among different classes is uneven or biased. Handling small and imbalanced training datasets poses a notable challenge in machine learning, especially in domains such as bioinformatics and medical research. These challenges can result in biased models, leading to poor performance on under-represented classes and an overemphasis on specific features, failing to capture the genuine patterns present in the data. The present study proposes a feature selection approach-based on genes connectivity and a class balancing technique for building a machine leaning model using imbalanced gene expression data. Rheumatic arthritis data composed of 28 normal samples and 152 rheumatic samples was used in testing our proposed model. Through the weighted gene co-expression network analysis (WGCNA) approach, features were reduced to 601 from 27,991 original features. The reduced features were used to build machine learning classification models with imbalanced and later balanced classes using Spread Sub-Sample technique. According to our findings, two classifiers reported higher accuracy with imbalanced data as compared to the balanced data set. This is an indication that most classifiers are biased when trained using imbalanced dataset. Logistic regression returned improved accuracy of 95%. The other two machine learning algorithms used in this study were decision tree and IBK returned reduced accuracy of 81% and 91% respectively. In conclusion, feature selection and class balancing approaches are important in reducing model execution time and accuracy especially for RNASeq gene expression data.

Digital image analysis has brought great benefit to agriculture. Plant diseases are an important impediment to food security because they lead to reduction in the quality and quantity of agricultural produce. Detection of plant diseases relies on the skills of the field personnel and often times lacks accuracy since it depends on the experience of the field team. In some cases, disease symptoms vary from one plant to another and this can lead to misclassifications is that. The objective of this study was to develop a classification model for maize leaves diseases using a Convolutional neural network (CNN). Maize leaves showing variation from normal leaves were collected from farms within Embu County, Kenya. Disease symptoms on each leaf were used to diagnose the underlying disease by a plant pathologist. A total of 405 images were captured by focusing on the diseases sections of each leaf using digital camera. We divided our dataset into three parts which is training, validation and testing sets as required for the deep neural network. We used two deep convolutional neural networks AlexNet and ResNet50 to perform maize disease classification. It was observed that both models (AlexNet and ResNet-50) achieved the highest accuracy when the training data was set at 70% and testing dataset was at 30%. The overall performance indicated that ResNet-50 model performed better than AlexNet. This study will help the farmers to take necessary arrangements based on detected diseases at the earliest to avoid the spread of maize diseases. However, the increase of sample size and training of extra neural networks and an application of Ensemble methods is recommended for a more conclusive model accuracy of maize leaves disease and other plant diseases.

Tsetse flies use antennal expressed genes to navigate their environment. While most canonical genes associated with chemoreception are annotated, potential gaps with important antennal genes are uncharacterized in Glossina morsitans morsitans. We generated antennae-specific transcriptomes from adult male G. m. morsitans flies fed/unfed on bloodmeal and/or exposed to an attractant (ε-nonalactone), a repellant (δ-nonalactone) or paraffin diluent. Using bioinformatics approach, we mapped raw reads onto G. m. morsitans gene-set from VectorBase and collected un-mapped reads (constituting the gaps in annotation). We de novo assembled these reads (un-mapped) into transcript and identified corresponding genes of the transcripts in G. m. morsitans gene-set and protein homologs in UniProt protein database to further annotate the gaps. We predicted potential protein-coding gene regions associated with these transcripts in G. m. morsitans genome, annotated/curated these genes and identified their putative annotated orthologs/homologs in Drosophila melanogaster, Musca domestica or Anopheles gambiae genomes. We finally evaluated differential expression of the novel genes in relation to odor exposures relative to no-odor control (unfed flies). About 45.21% of the sequenced reads had no corresponding transcripts within G. m. morsitans gene-set, corresponding to the gap in existing annotation of the tsetse fly genome. The total reads assembled into 72,428 unique transcripts, most (74.43%) of which had no corresponding genes in the UniProt database. We annotated/curated 592 genes from these transcripts, among which 202 were novel while 390 were improvements of existing genes in the G. m. morsitans genome. Among the novel genes, 94 had orthologs in D. melanogaster, M. domestica or An. gambiae while 88 had homologs in UniProt. These orthologs were putatively associated with oxidative regulation, protein synthesis, transcriptional and/or translational regulation, detoxification and metal ion binding, thus providing insight into their specific roles in antennal physiological processes in male G. m. morsitans. A novel gene (GMOY014237.R1396) was differentially expressed in response to the attractant. We thus established significant gaps in G. m. morsitans genome annotation and identified novel male antennae-expressed genes in the genome, among which > 53% (108) are potentially G. m. morsitans specific.

Background High-throughput sequencing generates large volumes of biological data that must be interpreted to make meaningful inference on the biological function. Problems arise due to the large number of characteristics p (dimensions) that describe each record [n] in the database. Feature selection using a subset of variables extracted from the large datasets is one of the approaches towards solving this problem. Methodology In this study we analyzed the transcriptome of Glossina morsitans morsitans (Tsetsefly) antennae after exposure to either a repellant (δ-nonalactone) or an attractant (ε-nonalactone). We identified 308 genes that were upregulated or downregulated due to exposure to a repellant (δ-nonalactone) or an attractant (ε-nonalactone) respectively. Weighted gene coexpression network analysis was used to cluster the genes into 12 modules and filter unconnected genes. Discretized and association rule mining was used to find association between genes thereby predicting the putative function of unannotated genes. Results and discussion Among the significantly expressed chemosensory genes (FDR < 0.05) in response to Ɛ-nonalactone were gustatory receptors (GrIA and Gr28b), ionotrophic receptors (Ir41a and Ir75a), odorant binding proteins (Obp99b, Obp99d, Obp59a and Obp28a) and the odorant receptor (Or67d). Several non-chemosensory genes with no assigned function in the NCBI database were co-expressed with the chemosensory genes. Exposure to a repellent (δ-nonalactone) did not show any significant change between the treatment and control samples. We generated a coexpression network with 276 edges and 130 nodes. Genes CAH3, Ahcy, Ir64a, Or67c, Ir8a and Or67a had node degree values above 11 and therefore could be regarded as the top hub genes in the network. Association rule mining showed a relation between various genes based on their appearance in the same itemsets as consequent and antecedent.

High throughput sequencing generates large volumes of high dimensional data. Identifying informative features from the generated big data is always a challenge. Feature selection reduces complex data into a smaller number of variables while preserving the information as much as possible. In this study, we used DaMiRseq, DESeq2, edgeR and Limma + voom to identify differentially expressed genes in 79 small cell lung cancer (sclc) and 7 normal controls. A gene network was used to identify any coexpressed genes. Association rule mining was used to identify any association between connected genes in the network. Limma + voom identified the highest number of differentially expressed genes. However, 81 genes were common in the four differential gene expression analysis methods used. After filtering out all nodes with a degree less than 5, the final network had 43 nodes and 63 edges. Association rule mining on the coexpressed genes generated 263 rules. Genes that were common in the rules were: SLC34A2, CAV2, EPAS1, CTSH, AQP1 and LRRK2. These genes have been associated with various types of cancer. Therefore, feature selection using differential gene expression analysis, co-expression networks and association rule mining could help infer relationships among genes and their possibility of having a shared biological function.

Analysis of high-dimensional data, with more features (p) than observations (N) (p>N), places significant demand in cost and memory computational usage attributes. Feature selection can be used to reduce the dimensionality of the data. We used a graph-based approach, principal component analysis (PCA) and recursive feature elimination to select features for classification from RNAseq datasets from two lung cancer datasets. The selected features were discretized for association rule mining where support and lift were used to generate informative rules. Our results show that the graph-based feature selection improved the performance of sequential minimal optimization (SMO) and multilayer perceptron classifiers (MLP) in both datasets. In association rule mining, features selected using the graph-based approach outperformed the other two feature-selection techniques at a support of 0.5 and lift of 2. The non-redundant rules reflect the inherent relationships between features. Biological features are usually related to functions in living systems, a relationship that cannot be deduced by feature selection and classification alone. Therefore, the graph-based feature-selection approach combined with rule mining is a suitable way of selecting and finding associations between features in high-dimensional RNAseq data.