Marija Rakić: Prediction and Validation of the Temperature Factors of Protein Structures

☰ Meni

V ponedeljek, 25. avgusta 2025, bo ob 16:00 uri izvedeno
predavanje v okviru PONEDELJKOVEGA SEMINARJA RAČUNALNIŠTVA IN INFORMATIKE
Oddelkov za Informacijske znanosti in tehnologije UP FAMNIT in UP IAM.

ČAS/PROSTOR: 25. avgust 2025 ob 16.00 prek Zoom-a (https://upr-si.zoom.us/j/297328207?pwd=S3Zpdk1VR3pjckNtWkQwKzlvcDR5UT09).

——————————————–
PREDAVATELJICA: Marija RAKIĆ
——————————————–

Marija Rakić is a 2nd-year Data Science master’s student at UP FAMNIT, writing her master thesis under the supervision of Assoc. Prof. Jure Pražnikar, PhD. She is also working at UP FAMNIT, as part of the pilot project GDI UP.

——————————————————————————————————————
NASLOV: Prediction and Validation of the Temperature Factors of Protein Structures
——————————————————————————————————————

POVZETEK:

Experimental B-factors, while routinely deposited in crystallographic models, are often inconsistent, resolution-dependent, or entirely missing in predicted structures. To overcome these limitations, we employed a model for the goal of predicting and validating protein B-factors. The model is based on graphlet degree vectors (GDVs), which capture the local topological environment of each atom within a protein structure. The model was used on a dataset of nearly 69,000 protein structures from the Protein Data Bank (PDB), where GDVs were used as features in a multiple linear regression framework. We showed that this approach enables the prediction of B-factors without relying on resolution-specific parameters or refinement metadata. Evaluation of model performance shows strong agreement between predicted and experimental values, with a mean Spearman correlation of 0.72 across the dataset and consistent performance across resolution groups. As a key outcome, a web-based tool named ProtWeb was developed. It allows users to upload protein structures in PDB format and receive two modified files: predicted.pdb, containing normalized B-factor predictions, and rescaled.pdb, where predicted values are scaled to match the experimental distribution.

Seminar bo potekal v angleškem jeziku.

Seminar bo izveden online prek Zoom-a na sledeči povezavi:

https://upr-si.zoom.us/j/297328207?pwd=S3Zpdk1VR3pjckNtWkQwKzlvcDR5UT09

Meeting ID: 297 328 207
Passcode: 123456789

Vabljeni!