THANK YOU FOR SUBSCRIBING
Pharma Tech Outlook | Tuesday, January 31, 2023
The application of DL to the discovery of new drugs requires a rigorous examination and explanation of its respective applications.
FREMONT, CA: Computer scientists and medicinal chemists collaborate to develop drug discovery and development tools, predictive models, and algorithms utilizing deep learning.
Stay ahead of the industry with exclusive feature stories on the top companies, expert insights and the latest news delivered straight to your inbox. Subscribe today.
Computational strategies have shown efficiency in producing accurate DTI predictions, whereas traditional methods have faced monetary and technical constraints. Several computational approaches are currently in use in DTI prediction, including ligand-based approaches, docking simulations, chemo-genomic approaches, text mining methods, ML/DL-based, and network-based approaches.
DL-based tools aid the drug discovery and development process.
DTI-CNN: A convolutional neural network (CNN) uses NNs to analyze visual imagery. The DTI-CNN tool is a simple, DL-based tool for predicting drug–target interactions that outperforms existing state-of-the-art methods through intelligent interaction between three components: heterogeneous network-based feature extraction, denoising autoencoding-based feature selection, and CNN-based interaction prediction. This model uses a random walk with restart (RWR) and denoising auto encoder (DAE) to incorporate low-dimensional feature vectors and noisy incomplete and high-dimensional features from various heterogeneous sources, including drugs, proteins, and illnesses.
DeepCPI: A new generation of DL algorithms has shown superior performance in predicting DTBA to conventional ML algorithms. This algorithm uses a compact text format to represent molecular structure known as simplified molecular input line entry (SMILES). Each chemical entity is represented by a single ASCII string between 20 and 90 characters. A combination of these three inputs or input features includes ligand maximum common substructures (LMCS), extended connectivity fingerprints (ECFP), or any combination of these three. NNs have different strengths, making them suitable for various applications since they use different types of neural networks (NNs).
DeepDTA: DTBA-predicting DL-based approaches are currently dominated by DeepDTA, which uses SMILES for drug input data. It is similar to the way SMILES encodes amino acids and protein sequences. In DeepDTA, CNN uses three 1D convolutional layers that follow max-pooling functions. These 3D convolutional layers are applied to the drug embedding to learn latent features. Developers construct and apply a similar CNN block to the protein embedding. The validation of DeepDTA involves tuning many hyper-parameters, such as the number of filters, filter length of drug and protein, batch size, optimizers, and learning rates. Prediction and real DBTA values should be as similar as possible during the training period. It is possible to overcome the limitations of this model due to the use of CNN by using more appropriate architectures that can learn from longer protein sequences, such as long short-term memory
More in News