Optimizing Biomarkers From an Ensemble Learning Pipeline
Prashant Kumar Kuntala · OhioLink ETD Center (Ohio Library and Information Network) · 2017
Understanding gene expression pattern is crucial in deciphering any observed biological phenotypes.Transcription factors (TF) are proteins that regulate genes by binding to a transcription factor binding site (TFBS) within the promoter region of a gene.Motif discovery is a computational approach that conventionally uses stochastic models, enumeration methods and many other techniques to report candidate motifs (TFBS).These methods generate similar motifs for a TF due to various reasons.Motif selection algorithms successfully identify a small set of motifs that address the specificity problem and coverage problem in motif discovery.However, these selected motifs do not always capture all the binding site preferences for a TF.This study verifies the hypothesis that motif discovery tools generate similar motifs for a transcription factor and once these variants (similar motifs) are identified, they can be used to form a super motif set, which may improve the accuracy of motif discovery.This study introduces the concept of Super motif set, a new model to accurately predict the binding sites for a TF.Two heuristic algorithms are introduced to identify Super motif sets, utilizing motif selection algorithms and a motif comparison tool.These super motif sets identified, capture the biological diversity in TFBS preferences of a TF.The algorithms are evaluated on ChIP-seq data for 54 TF factor groups from the ENCODE project.Moreover, the proposed algorithms are used to optimize the motifs that are reported by motif selection algorithms and to report super motif sets in three case studies: Chagas disease, pollen specific HRGP genes in Arabidopsis thaliana and Shigellosis.On an average two motif variants are added to the selected motifs, which improve the accuracy of motif discovery.I would like to thank and express my sincere gratitude to Dr. Lonnie Welch, who introduced me to bioinformatics and guided me to pursue this area of study.I am grateful to Dr. Lonnie Welch for accepting me into the bioinformatics lab at Ohio University and guiding me in learning the ropes of research, for the valuable and constructive suggestions during the planning and development of this research work.I am particularly