Titel: aMP: Deep learning framework to characterize meiosis progression timeline in tetraploid Arabidopsis thaliana
Sprache: Englisch
Autor*in: Ghosh, Ankit
Schlagwörter: Bioinformatics; Machine learning; Single cell analysis; Meiosis; Plant Science
Erscheinungsdatum: 2024
Tag der mündlichen Prüfung: 2024-10-01
Zusammenfassung: 
In plant science, the detailed examination of meiosis progression is hampered by the labour-intensive process of manual image analysis. Our research presents a novel, deep learning-based automation pipeline that significantly streamlines the quantification of meiotic timelines by analysing live-imaging videos. This innovative approach employs segmentation models to stabilize video frames, refines localization techniques to precisely identify individual meiocytes, and adopts a restricted space motion-inspired tracking methodology that effectively reduces computation time and improves tracking accuracy. Crucially, our framework distinguishes itself by generating Z-normalized staging pathways, enabling the construction of a piece-wise timeline of meiotic progression. This is achieved through a meticulously curated landmarking scheme, which our results confirm aligns with established meiosis timelines in both wild-type and heat-shocked Arabidopsis thaliana.

Our study ventures beyond the diploid paradigm, extending the application of our high-throughput pipeline to tetraploid variants. The analyses disclose that while tetraploids exhibit comparable meiosis-I timelines to their diploid counterparts, a pronounced prolongation characterizes their meiosis-II stages. Furthermore, the systematic examination of tcx5;6 mutants and ATM gene insertions in tetraploids provides a quantitative view of the temporal dynamics in meiotic progression, highlighting the potential for chromosomal behaviour and genetic regulation to modulate meiotic efficiency.

By integrating a convolutional neural network (CNN) based methodology with our modular pipeline, we deliver a transformative tool for meiosis analysis. Our work is not only restricted to timeline analysis, but the modular approach shows ability in different segmentation tasks from basics like pollen counting to more structured like DNA double-strand break and BiFC quantization.
URL: https://ediss.sub.uni-hamburg.de/handle/ediss/11301
URN: urn:nbn:de:gbv:18-ediss-123289
Dokumenttyp: Dissertation
Betreuer*in: Schnittger, Arp
Laue, Sören
Enthalten in den Sammlungen:Elektronische Dissertationen und Habilitationen

Dateien zu dieser Ressource:
Datei Prüfsumme GrößeFormat  
PhD_Thesis_AG.pdfa5d30fff3be93fdb68b212fff8c3db3b66.37 MBAdobe PDFÖffnen/Anzeigen
Zur Langanzeige

Info

Seitenansichten

Letzte Woche
Letzten Monat
geprüft am null

Download(s)

Letzte Woche
Letzten Monat
geprüft am null
Werkzeuge

Google ScholarTM

Prüfe