A research study titled “A deep hybrid GNN based on edge-conditioned and graph isomorphism network convolutions for PC-3 anticancer screening” describes a novel screening method using a modified deep hybrid Graph Neural Network (GNN) architecture for predicting anticancer response in prostate tumours based on the 2-D molecular structures of molecular compounds.
The Department of Science and Technology-Advanced Science and Technology Institute’s (DOST-ASTI) AI Researcher Adrian S. Remigio, Science Research Specialist I of the Research Unit of ASTI-Advanced Labelling Machine (ASTI-ALaM) Project presented this at the 10th International Conference on Signal Processing and Integrated Networks (SPIN 2023) organised by the Department of Electronics and Communications Engineering at Amity University, Noida, India.
This research project focuses on ASTI-ALaM’s pioneering work in enhancing the predictive accuracy and capacity of AI models. When compared to simple GNN models, the modified deep hybrid architecture demonstrated acceptable and enhanced prediction accuracy, which can be used to aid in the search for anticancer treatments in clinical research.
The researcher used practise datasets to test the GNN models during the initial investigation on the implementation of GNNs in Python. The PC3 anticancer screening was one of the practice datasets he used. After reviewing some GNN literature, the researcher opted to test the redesigned architecture on a practice dataset and received encouraging results.
The researcher focused on exploring the application of Graph Neural Networks (GNN) and hybrid Convolutional Neural Networks-Graph Neural Networks (CNN-GNN) for computer vision and devising methods to prevent catastrophic amnesia in these network architectures.
Currently, the ASTI-ALaM Project research focuses on hybrid CNN-GNN models with deep hybrid architecture to develop a novel architecture for land-use mapping: the REsNEt-deep CNN-GNN model. This developed model outperformed the standard CNN model for land-use mapping applications in aerial images, as demonstrated.
The Research Unit of ASTI-ALaM is working to improve the predictive accuracy and capacity of AI models and seeks to publish articles and provide novel solutions that can be incorporated into the repository of AI models. The participation of the ASTI-ALaM Project in this conference represents a significant milestone in the project’s ongoing research efforts, further strengthening DOST-ASTI’s position as the country’s AI research powerhouse.
The IEEE Computational Intelligence Society is a co-sponsor of SPIN 2023, and papers presented at the conference are anticipated to be published in the IEEE Xplore journal, a digital repository for content produced by the Institute of Electrical and Electronics Engineers (IEEE) and its publishing partners.
The purpose of this conference was to bring together researchers, academics, and businesspeople in the fields of integrated networks and signal processing to exchange cutting-edge concepts.
In addition, a Satellite Data Processing Training Session (SDPTS) was organised to further advance the participants’ knowledge and skills on earth observations datasets, resources, and techniques in remote sensing through a series of lectures, practical exercises, and assessments.
The ASTI-ALaM Project researchers also attended the training session and employ conventional remote sensing techniques for mapping and computer vision tasks.
Dr Franz de Leon, director of the DOST-ASTI, expressed his support for the Space Information, Know-How, and Applications Acceleration through Promotion and Training (SIKAP+) Project and cited the importance of conducting capacity-building activities such as training, seminars, and workshops on the peaceful applications of space science and technology in establishing collaborations and partnerships.
He added that it was an opportunity for the participants to network and build relationships with other industry professionals from various Asian regions.
