Glycan-related Drug Property Prediction Service
Your Best Assistant in Glycan-related Drug Property Prediction
A key in Glycan Drug Discovery is the prediction of absorption, distribution, metabolism, excretion, and toxicity (ADMET) from the molecular structure of small molecules. Accurate prediction of glycan-related drug properties is of great importance in drug development and medical research. Computational methods offer the advantage of rapid identification of specifically characterized drug candidates compared to traditional in vitro and in vivo experiments. CD BioGlyco provides our clients with AI-driven molecular property prediction services for glycan-related drugs. Our reliable learning algorithms are used to predict the chemical activity of new and previously unseen molecules.
Predictive modeling
Constructing a predictive model generally requires the selection of a molecular representation, Machine Learning (ML) algorithms, and parameter optimization. Our specialized computer team provides a model for similarity-based prediction of molecular properties through fingerprinting, embedding known molecules into a similarity-based finite environment, calculating the Pareto dominance relationships of known molecules, and converting these dominance relationships into similarity scores and predictions.
Our researchers construct a network of relationships between activity and molecular structure during the training learning phase. The predictive strength of the model is boosted in a context-specific manner and combination with many different molecular structures. The predictive models we provide can be used to predict the chemical activity of new, previously unseen molecules.
Cross-validation of predictive models
The predictive models we provide are trained and evaluated using a five-fold cross-validation strategy. Our lab uses 80% of the data for learning and training and the remaining 20% for evaluation.
Drug property prediction
Our researchers compute physicochemical descriptors and use them to predict glycan-related drug properties, medicinal chemistry friendliness, ADME, and pharmacokinetic properties. Moreover, we provide high-quality assessment services for drug similarity of glycan-related drug molecules.
Qualified drugs have receptor/channel binding properties, availability, gastrointestinal absorption, optimal bioavailability, good absorption/penetration, and brain penetration. Relying on advanced predictive modeling, we provide predictive services for all these characteristics. Ineligible molecules are removed before synthesis.
Publication
Technology: Graph neural networks (GNNs), FRagment-based dual-channEL pre-training (FREL) modeling, Pretraining, Transfer learning, Molecular representation learning
Journal: BMC Bioinformatics
Published: 2023
IF: 3.3
Results: In this work, researchers constructed a new FREL model. It has the advantage of utilizing generative learning and comparative learning to obtain intramolecular and intermolecular consistency, respectively. With resources and tests collected from 10 public databases, the researchers evaluated the FREL model and further analyzed the correlation between drug molecular representations and molecular properties. Visualization and correlation analysis confirmed that molecular fragments have the potential to improve the performance of drug property prediction. Overall, this study confirms the importance and necessity of integrating molecular fragmentation with model design and provides high-quality models for drug property prediction.
Fig.1 Distribution of the predicted property and true label. (Wu, et al., 2023)
Applications of Glycan-related Drug Property Prediction
- Glycan-related drug property prediction advances glycan-related drug-target affinity prediction and molecular synthesis prediction.
- Glycan-related drug property prediction plays an important role in glycan-related drug design including protein structure minimization and lead optimization.
- Glycan-related drug property prediction quickly finds compounds that match the predicted properties among a large number of candidate compounds, speeding up drug screening.
Advantages of Us
- We provide predictive models that meet the needs of our clients and the underlying principles of the predictions are interpretable.
- Our excellent model-building team has extensive industry experience in bioavailability, absorption, and permeation prediction before improving efficiency.
- For different needs, our lab provides different models to predict glycan-related drug properties.
Frequently Asked Questions
- What are the deep learning methods that can be applied in molecular representation and molecular properties prediction?
- Depending on the form of molecules, deep learning can be categorized into, graph-based approach, image-based approach, sequence-based approach, 3D graph-based approach, 3D grid-based approach, mixed data-based approach, migration learning, multi-task learning and meta-learning approach, interpretable deep learning approaches in molecular property prediction.
- What are the basic requirements for a machine-learning approach to predicting the chemical activity of glycan-related drug molecules?
- A molecular representation
- A discriminative supervised learning algorithm
- A library of different reference molecules with known properties to predict
CD BioGlyco has an advanced computer system and professional team to help our clients complete the molecular prediction project of glycan-related drugs quickly. Our high-quality solutions and enthusiastic service concepts have been recognized by our clients. Please feel free to contact us.
Reference
- Wu, Y.; et al. Enhancing drug property prediction with dual-channel transfer learning based on molecular fragment. BMC Bioinformatics. 2023, 24(1): 293.
For research use only. Not intended for any diagnostic use.
Related Services
