PhD Scientific Days 2023

Pharmaceutical Sciences - Posters E

Generative Machine Learning Framework for the Prediction of Links in PPI Networks

Előadó neve

Balogh, Olivér, MSc

Neptun code

HYKT8D

Előadó munkahelye

Semmelweis University, Department of Pharmacology and Pharmacotherapy

Előadó telefonszáma

+36203685079

Előadó e-mail címe

balogh.oliver@med.semmelweis-univ.hu

Az előadás címe

Generative Machine Learning Framework for the Prediction of Links in PPI Networks

Szerző(k) neve és munkahelye

Olivér M. Balogh1,2, Bettina Benczik1,3, András Horváth2, Mátyás Pétervári1, Péter Csermely4, Péter Ferdinandy1,3, Bence Ágg1,3

1 Cardiometabolic and MTA-SE System Pharmacology Research Group, Department of Pharmacology and Pharmacotherapy, Semmelweis University, Nagyvárad tér 4, Budapest 1089, Hungary
2 Faculty of Information Technology and Bionics, Pázmány Péter Catholic University, Budapest, Hungary
3 Pharmahungary Group, Szeged, Hungary. 4 Department of Molecular Biology, Semmelweis University, Budapest, Hungary

Bemutatás módja

Poszter

Szekció

Pharmaceutical Sciences - Posters E

Language of the presentation

English

Preferred session

Pharmaceutical Sciences

Összefoglaló szövege

Introduction
Large scale wet-lab evaluation of potential protein-protein interactions (PPIs) is a costly procedure, thus there is a demand for efficient in-silico methods that are able to predict the most probable interactions and narrow down the candidates. Reformulating the problem with a systems biology perspective, PPIs can be represented in a network form, enabling us to use alternative approaches, such as link prediction tools for the prediction of PPIs.

Aims
We aimed to develop a machine learning model that operates on input data derived from the topology of PPI networks (interactomes), and learns to predict potential new links, corresponding to previously unknown PPIs.

Method
Our machine learning model used for link prediction was inspired by a conditional generative adversarial network (cGAN) architecture. We adapted the cGAN model for the purpose of translating “images” of extracted subgraphs into their denser forms, where the new edges represent the predicted PPIs. We tested our method on the STRING database (cutoff = 0.95), using the interactomes of five different species.

Results
The model yielded consistent and good results across the different species, demonstrating the applicability and efficiency of our method. As a particular example, the 10-fold evaluation on the human interactome resulted in an average of AUROC = 0.913, AUPRC = 0.169, NDCG = 0.761 with 1310 seconds computing time per fold, which includes the training of the model.

Conclusion
We developed a cGAN-based method for the purpose of link prediction using only raw topological information of the network. We provided the first demonstration that such an approach is applicable for the task of predicting unknown PPIs using only the interactomes without external molecular attributes, and serves as a robust base for further improvements in PPI prediction.

Funding
The research was financed by the Thematic Excellence Programme (2020-4.1.1.-TKP2020) of the Ministry for Innovation and Technology in Hungary, within the framework of the Therapeutic Development and Bioimaging thematic programmes of the Semmelweis University. Project no. RRF-2.3.1-21-2022-00003 has been implemented with the support provided by the European Union. OB was supported by EFOP-3.6.3-VEKOP-16-2017-00009 “Semmelweis 250+ Kiválósági PhD Ösztöndíj” grant.

University and Doctoral School

Semmelweis University, Doctoral School of Pharmaceutical Sciences

Supervisor

Bence Ágg, MD, PhD

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

7403

Start

11:48

End

11:53

Authors (legacy)

Olivér M. Balogh1,2, Bettina Benczik1,3, András Horváth2, Mátyás Pétervári1, Péter Csermely4, Péter Ferdinandy1,3, Bence Ágg1,3

1 Cardiometabolic and MTA-SE System Pharmacology Research Group, Department of Pharmacology and Pharmacotherapy, Semmelweis University, Nagyvárad tér 4, Budapest 1089, Hungary
2 Faculty of Information Technology and Bionics, Pázmány Péter Catholic University, Budapest, Hungary
3 Pharmahungary Group, Szeged, Hungary. 4 Department of Molecular Biology, Semmelweis University, Budapest, Hungary