Year: 2023
Type: dissertation
Abstract: Genomic disorders are pathologies that result from the rearrangement/variations of the human genome, that lead to DNA sequence alteration. These are recurrently manifested as developmental disorders, including developmental delay and intellectual disability. Currently these genetic variations, such like copy number variants (CNVs), are identified via Next Generation Sequencing technologies (NGS), which is a less cumbersome and time-consuming methodology than the previously ones. Nevertheless, NGS data analysis is still prone to errors, which are frequently associated to mapping errors, especially in low complexity regions such as telomeres and centromeres. To identify CNVs through NGS data, the sequenced read-pairs are mapped against the human reference genome. Then, the improper pairs for deletions and duplications (with higher than expected insert size or aligned in an outward facing orientation, respectively) are selected and clustered together, defining the variant regions. Concurrently, CNVs can also be identified by chimeric reads, which overlap the CNV breakpoint, corresponding to a junction fragment, or by differences of sequence coverage along the genome. Nevertheless, as the identification of CNVs depends on the read mapping, genomic characteristics, as low complexity regions, lead to the identification of False positive (FP) CNVs which, without manual verification, are difficult to distinguish from real CNVs (True positives - TP). The manual validation of hundreds of CNVs per study case can be a laborious task, therefore automation is vital. In this study, we proposed the development of an automated validation pipeline, established on Machine Learning Classification Model, able to make predictions based on patterns identified on the input data. Thus, in this project a machine learning model able of distinguish FP and TP CNVs based on the genomic characteristics of the specific mapping regions, such as GC content and distance to telomere and centromere, was developed.
Author: Matos, José Pedro Parracha de
Institution: University of Lisbon
Language: Portuguese
Subfield: Genetics
Domain: Life Sciences
Open Access status: green
Funder: Indian National Science Academy
OpenAlex ID: W6990425476
