PhD Scientific Days 2023

Molecular Sciences - Posters B

IDPredict: Predicting disordered proteins as potential predictive biomarkers for targeted therapies with a machine learning approach based on network topological data

Előadó neve

Dr. Schulc, Klára

Neptun code

CM68W8

Előadó munkahelye

Semmelweis University, Department of Molecular Biology

Előadó telefonszáma

06301811341

Előadó e-mail címe

rako.klari@gmail.com

Az előadás címe

IDPredict: Predicting disordered proteins as potential predictive biomarkers for targeted therapies with a machine learning approach based on network topological data

Szerző(k) neve és munkahelye

Klára Schulc1
Dávid Keresztes1

Department of Molecular Biology

Bemutatás módja

Poszter

Szekció

Molecular Sciences - Posters B

Language of the presentation

English

Preferred session

Molecular Sciences

Összefoglaló szövege

Introduction:
Identificating predictive oncotherapeutic biomarkers is a promising area of cancer research. With the IDPredict project, we suggest a machine learning method based on network topological motif analysis to the prediction for intrinsically disordered proteins (IDPs) with predictive biomarker properties. Analysing motifs containing IDPs and oncotherapeutic targets, their complex regulation can be modelled. A Biomarker Probability Score was developed to assess the biomarker potential of each disordered protein-target pair.
Material and method:
Motifs were identified on the directed edges of three networks using the FANMOD program. Cytoscape plugins were used for network analysis. IDPs and biomarkers were annotated based on the DisProt and CIViCmine databases, respectively. Machine learning models were developed with the Python scikit-learn, XGBoost and SHAP packages. A Biomarker Probability Score was developed based on the ranks of eight independent predictions.
Results and discussion:
We have shown that IDPs form common motifs with known oncotherapeutic targets, where many IDPs are predictive biomarkers for the given target (Human Cancer Signaling Network: 23%). Thus, machine learning models were developed based on topological data and biological characteristics of 109 target-IDP pairs, whose predictive biomarker properties were previously known. After multiple cross-validations and testing, eight different models with >0.88 LOOCV accuracy were used to predict potential predictive biomarker properties for the other identified 695 pairs in our networks. The resulting Biomarker Probability Score was showed to perform well in the discrimination of potential predictive biomarkers. As one of the examples, a clinically relevant predictive biomarker was identified for ponatinib, a multikinase-inhibitor with BCR-ABL as its main target.
Conclusion:
We conclude based on network topology analysis that intrinsically disordered proteins have great potential as predictive biomarkers. These results imply that investigation of their motifs helps to identify novel predictive biomarkers, which can be both further studied individually and validated experimentally. As predictive biomarkers are vital for therapeutical decision making, developing a tool for predictive biomarker identification may have effect on the daily clinical practice.

University and Doctoral School

Semmelweis University, Doctoral School of Molecular Medicine

Supervisor

Prof. Csermely Péter, Dr. Veres Dániel

Publication of my abstract

I do not give consent to the publication of my abstract on the website of the congress.

Kind

Szabad

Status

elfogadva

Accepted presentation method

poszter

Előadás fájl jóváhagyás

nem rendelkezett róla

Előadó

6836

Start

12:06

End

12:11

Authors (legacy)

Klára Schulc1
Dávid Keresztes1

Department of Molecular Biology