[Pictured: Protein surface with ligand]
Mr Joseph Azzopardi, an M.Sc. student in Artificial Intelligence, together with his dissertation supervisor, Dr Jean Paul Ebejer from the Centre for Molecular Medicine and Biobanking, have recently presented the findings resulting from Joseph's dissertation, at the .
This work, titled ‘LigityScore: Convolutional Neural Network for Binding-affinity Predictions’, was later published in the conference proceedings. This work focuses on the use of AI to predict binding of small molecules with target proteins. It outlines significant contributions in the areas of bioinformatics and the use of AI for drug discovery.
Convolution Neural Networks (CNN) are specialized types of deep neural network that typically process data with a grid-like input type, which uses the convolution mathematical operation at the centre of its learning process.
CNNs were applied to develop a scoring function within structure-based drug design to estimate the binding strengths between ligands and target proteins. Seeking a scoring function that can accurately predict the binding affinity is key for successful virtual screening methods.
In this study, the authors presented the LigityScore family of methods, which are rotationally-invariant scoring functions based on CNNs. LigityScore descriptors are extracted directly from the structural and interacting properties of the protein-ligand complex to create a data representation which is input to a CNN for automatic feature extraction and binding strength prediction. This data representation component and the CNN architecture together, constitute the LigityScore scoring function.
The main contribution for this study was to present a novel protein-ligand representation for use as a CNN based scoring function for binding affinity prediction and obtain a good prediction performance. LigityScore models were evaluated for scoring power on the latest two Comparative Assessment of Scoring Functions (CASF) benchmarks and ranked 5th place for the CASF-2013 benchmark, and 8th for CASF-2016, with an average R-score correlations performance of 0.713 and 0.725 respectively between predicted and actual binding affinity values.
The conference provided a great opportunity for the authors to interact with other researchers but was restricted to an online event due to the COVID-19 situation.
Although the experience was less engaging than in previous years it was still a great experience and honour to present and share their work with other experts in the field of bioinformatics. The study is available online through ScitePress digital library and the source code for LigityScore is available .
For further information about this research kindly contact Dr Jean Paul Ebejer.
