Candidate disease gene prediction : a deep learning approach

Carvalho, Matilde Baptista Neves de

http://hdl.handle.net/10400.5/95452

Use this identifier to reference this record.

Name:	Description:	Size:	Format:
TM_Matilde_Carvalho.pdf		1.22 MB	Adobe PDF	Download

Send Feedback

Authors

Carvalho, Matilde Baptista Neves de

Advisor(s)

Barros, Márcia Cristina Afonso

Abstract(s)

The recommendation of candidate genes plays a crucial role in genetic research and medicine, given the vast and complex nature of the human genome, which harbors a multitude of genes potentially associated with various diseases. In this context, recommendation systems assume particular relevance by streamlining the identification process, acting as tools that suggest specific candidate genes related to a particular disease. Consequently, this approach expedites the phases of scientific investigation and discovery. Additionally, in the current landscape marked by the information age and big data, the implementation of recommendation systems enhances efficiency in genetic studies. Given the continuous accumulation of genetic information at an unprecedented rate, these systems assist researchers in navigating and prioritizing potential candidates, optimizing the use of time and resources. Healthcare professionals can also, with the aid of these systems, personalize treatments based on individual genetic composition, leading to more tailored and effective medical interventions and therapies. Given the intrinsic relevance of the subject, this study proposes a comprehensive approach to address contemporary challenges in genomic research. The creation of the recommendation system named RecSysModel in PyTorch stands out, with its architecture inspired by Neuronal Collaborative Filtering, aiming to recommend candidate genes for diseases listed in the Disease Ontology based on the level of scientific evidence in the literature. Data is imported directly from the SQLite database, designated DiseaseGene, into the system. It was designed to map diseases in the Disease Ontology, aligning with those recorded in the DisGeNET database, to contain detailed information about genes, diseases present in the ontology, and interactions between them, providing a robust approach for the analysis and recommendation of genes associated with these conditions. It is crucial to highlight that the approach adopted for the proposed model stands out for its innovation, as it was not identified during the state-of-the-art review. This absence underscores the unique and original contribution of this methodology, filling a gap in the scientific literature, emphasizing the singularity of the approach in this study, and the innovation presented to the scientific community in this research field.