Translating nucleic acid binding protein function from model species to minor crops using transfer learning
Publication Date
July 22, 2018
Creator
Abstract
Genomic elements such as proteins or genes are the basic unit of the genome and involved in the functioning of every biological process. Predicting, therefore, the function of these genomic elements is the first step in the understanding of functioning of plants under various stress conditions. To date, various types of computational methods have been developed to predict the function of a given protein sequence. The recent increase in the development of a number of methods has created its own set of problems leading to difficulty in applying on newly sequenced genomes especially non-model crops. Due to these reasons, the immediate requirement for development of sophisticated computational methods to predict the function of a given protein sequence is raised. This thesis presents three novel computational tools developed based on transfer learning algorithms to predict the function of a given protein sequence and these tools are: 1) TL-RBPPred, for prediction of RNA-binding proteins, outperformed SPOT-Seq, RNApred, RBPPred and BLASTp on HumanSet (AUC of 0.977), YeastSet (AUC of 0.971), ArabidopsisSet (AUC of 0.972) and GlymaxSet (AUC of 0.97); 2) TL-DBPPred, for prediction of DNA-binding proteins, outperformed DNABP, enDNA-Prot, iDNA-Prot, nDNAProt, iDNA-Prot|Dis, DNAbinder and BLASTp on an testing dataset (AUC of 0.988); and 3) TL-TFPred, for prediction of transcription factors, outperformed PlantTFcat, iTAK and BLASTp on testing dataset (AUC of 0.999) in terms of prediction accuracy. Further, both TL-RBPPred and TL-DBPPred were tested on the transcriptome of the non-model crop, Bambara groundnut (Vigna subterranea (L.) Verdc.), to identify RNA-binding and DNA-binding proteins, respectively. The results obtained from these tests indicated that these two methods outperformed in terms of prediction accuracy (AUC) as compared to existing current state-of-the art tools such as SPOT-Seq, RBPPred, iDNA-Prot and iDNA-Prot|Dis. Based on the performance, the developed methods will be useful in predicting the function of given protein sequences (DNA, RNA-binding and transcription factor) of model species as well as non-model crops.
Item Type
ethesis
Thesis Type
PhD
Supervisors
Subjects (LC)
Associated Schools / Departments
School of Biosciences (MY)
eprints ID
52289
UoN Repository URI
Except where otherwise noted, this item's license is described as
File(s)![Thumbnail Image]()
Name
Bonthala_Venkata_Suresh_Corrected_Thesis.pdf
Type
Full-text
Description
Examined
Size
3.72 MB
Format
Adobe PDF
Checksum (MD5)
3bd99d7bdfc7f02ab49ef1adf3a3789a