Use of Deep Multi-Target Prediction to Identify Learning Styles

It is possible to classify students according to the manner they recognize, process, and store information. This classification should be considered when developing adaptive e-learning systems. It also creates a comprehension of the different styles students demonstrate while in the process of learning, which can help adaptive e-learning systems offer advice and instructions to students, teachers, administrators, and parents in order to optimize students’ learning processes. Moreover, e-learning systems using computational and statistical algorithms to analyze students’ learning may offer the opportunity to complement traditional learning evaluation methods with new ones based on analytical intelligence. In this work, we propose a method based on deep multi-target prediction algorithm using Felder–Silverman learning styles model to improve students’ learning evaluation using feature selection, learning styles models, and multiple target classification. As a result, we present a set of features and a model based on an artificial neural network to investigate the possibility of improving the accuracy of automatic learning styles identification. The obtained results show that learning styles allow adaptive e-learning systems to improve the learning processes of students.

Study of miRNA – mRNA interactions related with cancer

MicroRNAs belong to the large family of small non coding RNAs. They regulateprotein synthesis by binding to their mRNA targets causing mRNA degradationor translational repression. A large number of miRNAs have been associated withcancer because they are often found to be located within cancer associatedgenomic region (CAGRs/FRA) to target cancer-related genes, and to bedifferentially expressed in tumor compared to normal tissues. Previous work inthe Computational Biology lab had identified four new putative miRNA genesthat were located within CAGR. However their mature molecules and theirassociation with cancer phenotypes were unknown. My thesis focuses onresolving these two issues, using a combination of theoretical and experimentaltechniques. The specific aims of this work are: The development of a mature miRNA prediction algorithm(Chapter II, III) The identification of the mature miRNA molecules of the newly identifiedmiRNA genes via a combination of computational and experimentalmethods (Chapter IV) The utilization of a target prediction algorithm to predict andexperimentally verify interactions between the mature molecules andcancer-related genes Chapter IV).

ANIMAL MICRORNA TARGET PREDICTION USING DIVERSE SEQUENCE-SPECIFIC DETERMINANTS

Many recent studies have shown that access of animal microRNAs (miRNAs) to their complementary sites in target mRNAs is determined by several sequence-specific determinants beyond the seed regions in the 5′ end of miRNAs. These factors have been related to the repressive power of miRNAs and used in some programs to predict the efficacy of miRNA complementary sites. However, these factors have not been systematically examined regarding their capacities for improving miRNA target prediction. We develop a new miRNA target prediction algorithm, called Hitsensor, by incorporating many sequence-specific features that determine complementarities between miRNAs and their targets, in addition to the canonical seed regions in the 5′ ends of miRNAs. We evaluate the performance of our algorithm on 720 known animal miRNA:target pairs in four species, Homo sapiens, Mus musculus, Drosophila melanogaster and Caenorhabditis elegans. Our experimental results show that Hitsensor outperforms five popular existing algorithms, indicating that our unique scheme for quantifying the determinants of complementary sites is effective in improving the performance of a miRNA target prediction algorithm. We also examine the effectiveness of miRNA-mediated repression for the predicted targets by using a published quantitative protein expression dataset of miR-223 knockout in mouse neutrophils. Hitsensor identifies more targets than the existing algorithms, and the predicted targets of Hitsensor show comparable protein level changes to those of the existing algorithms.