Deep Learning-Based Drug–Protein Binding Affinity Prediction Using Sequence Information

Drug-protein interaction plays a vital
role in drug discovery, as the interaction between drug
and protein, i.e., binding affinity, plays a key role in
drug efficacy and safety. The binding affinity between
drug and target protein can be represented as a
dissociation constant (Kd). Assessing binding affinity
often requires considerable time, money, and other
resources, as experimental studies are required to
compute binding affinity between drug-target proteins.
In that context, this paper discusses a comprehensive
approach to utilize a deep drug-target interaction
(DTI) approach that can be utilized to predict drug
protein binding affinity, utilizing a deep drug-target
interaction approach, considering that drug molecules
can be represented as a sequence of SMILES
annotation, and similarly, target proteins are
represented as a sequence of FASTA annotation.
Experiments are conducted utilizing a Davis dataset,
considering
suitable
preprocessing
steps,
i.e.,
tokenization, normalization, and division of dataset into
train-test sets. The efficacy of the proposed model is
assessed in accordance with various regression-based
performance metrics such as Mean Square Error
(MSE), Mean Absolute Error (MAE), Concordance
Index (CI), Pearson correlation coefficient, and R²
score. The experiments show that the model performs
well in generating accurate binding affinity predictions
and has substantial applications in computer-aided
drug screening.