Alfredo Barbaro
Recovering Gene Dependency Structure from Perturbation Data with Gradient-Boosted Trees.
Rel. Stefano Quer. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Informatica (Computer Engineering), 2026
|
|
PDF (Tesi_di_laurea)
- Tesi
Accesso limitato a: Solo utenti staff fino al 2 Ottobre 2029 (data di embargo). Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (1MB) |
Abstract
Large-scale Perturb-seq and CRISPR screens measure how a cell's transcriptome responds to thousands of single-gene knockouts, yet standard analysis treats each perturbation in isolation and never reconstructs the global structure of dependencies among genes. This thesis introduces a method that recovers directed, weighted gene dependency graphs from such data: for each perturbed gene, a gradient-boosted regressor predicts its expression from the rest of the transcriptome, its strongest predictors become directed edges weighted by signed fold change, and only edges that recur across many class-balanced resamples are kept, yielding reproducible rather than single-fit graphs. Across five cell types, regulatory hubs prove strongly context specific, with high directional agreement among reliable pairs but no edge shared by all five.
The structure transfers across cell types where biology is conserved but not for context-specific programs such as cell-cycle control, whose prediction error tracks the proliferative state of the target cell
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Corso di laurea
Classe di laurea
Ente in cotutela
Aziende collaboratrici
URI
![]() |
Modifica (riservato agli operatori) |
