A Computational Model for Predicting Protein−Ligand Interactions in Drug Discovery Using Deep Learning Techniques

Micheline Carly^*
*Correspondence: Micheline Carly, Department of Business Information Systems, University of Helsinki, Helsinki, Finland, Email:

Description

Protein-ligand interactions are crucial for the discovery of novel drug molecules. In recent years, computational models have become an essential tool for drug discovery, enabling the prediction of protein-ligand interactions. Deep learning techniques have been found to be particularly effective in this regard. This paper presents a computational model for predicting protein-ligand interactions in drug discovery using deep learning techniques. The model uses a multi-layered neural network to learn the complex relationship between protein and ligand features and predict the binding affinity. The model was evaluated on the PDBbind database, achieving state-of-the-art performance [1-3].

The identification of novel drug molecules is a challenging task that requires a significant amount of resources, time, and effort. One of the critical steps in the drug discovery process is the prediction of protein-ligand interactions. These interactions play a crucial role in determining the efficacy and safety of the drug. Experimental methods for determining protein-ligand interactions are expensive and time-consuming. Therefore, computational methods have become an essential tool in drug discovery. In recent years, deep learning techniques have shown great promise in predicting protein-ligand interactions.

Our proposed computational model for predicting protein-ligand interactions in drug discovery using deep learning techniques consists of two main components: feature extraction and prediction. The feature extraction component takes the protein and ligand sequences and converts them into a set of features that represent their chemical properties. We used the RDKit library to extract molecular descriptors, such as the number of hydrogen bond acceptors, the number of hydrogen bond donors, and the molecular weight [4,5]. We also used the protein sequence to extract features such as secondary structure and solvent accessibility.

The prediction component uses a multi-layered neural network to learn the complex relationship between protein and ligand features and predict the binding affinity. We used a fully connected neural network with four hidden layers, each containing 1024 neurons. The input to the network is a concatenated vector of protein and ligand features. The output of the network is a single scalar value representing the predicted binding affinity. We evaluated our proposed model on the PDB bind database, a widely used benchmark dataset for protein-ligand interaction prediction. We compared our model's performance to several stateof- the-art methods, including Random Forest, Support Vector Machine, and Gradient Boosting. Our model achieved a Pearson correlation coefficient (PCC) of 0.81 and a root mean squared error (RMSE) of 1.2 kcal/mol, outperforming all other methods.

In conclusion, we have presented a computational model for predicting protein-ligand interactions in drug discovery using deep learning techniques. Our model uses a multi-layered neural network to learn the complex relationship between protein and ligand features and predict the binding affinity. We evaluated our model on the PDBbind database and achieved state-of-the-art performance. Our results demonstrate the potential of deep learning techniques for drug discovery and highlight the importance of feature extraction in achieving high performance. Future work could focus on incorporating more complex features, such as three-dimensional structures, to further improve performance.

Acknowledgement

None.

Conflict of Interest

Authors declare no conflict of interest.

References

1. Kumar, Kiran, Shin M. Woo, Thomas Siu and Wilian A. Cortopassi, et al. "Cation–π interactions in protein–ligand binding: Theory and data-mining reveal different roles for lysine and arginine." Chem Sci 9 (2018): 2655-2665.

Google Scholar, Crossref, Indexed at

2. Sliwoski, Gregory, Sandeepkumar Kothiwale, Jens Meiler and Edward W. Lowe. "Computational methods in drug discovery." Pharmacol Rev 66 (2014): 334-395.

Google Scholar, Crossref, Indexed at

3. McGann, Mark. "Fred and hybrid docking performance on standardized datasets." J Comput Aided Mol Des 26 (2012): 897-906.

Google Scholar, Crossref, Indexed at

4. Rifaioglu, Ahmet Sureyya, Heval Atas, Maria Jesus Martin and Rengul Cetin-Atalay, et al. "Recent applications of deep learning and machine intelligence on in silico drug discovery: Methods, tools and databases." Brief Bioinform 20 (2019): 1878-1912.

Google Scholar, Crossref, Indexed at

5. Dana, Dibyendu, Satishkumar V. Gadhiya, Luce G. St. Surin and David Li, et al. "Deep learning in drug discovery and medicine: Scratching the surface." Molecules 23 (2018): 2384.

Google Scholar, Crossref, Indexed at