Structure-Based Ligand Activity Prediction With Binding Mode Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing structure-based activity prediction methods for ligands suffer from dataset bias and misrepresentation of ligand-protein interactions due to randomly distributed 'incorrect' poses, leading to inaccurate activity predictions.
Innovation Solution
A system and method that incorporates a binding mode prediction model with transfer learning techniques to improve activity prediction by selecting reliable binding modes using a binding mode selector, enhancing the performance of activity prediction models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If docking programs are used to sample binding modes, then binding mode predictions can be obtained, but incorrect poses are randomly distributed leading to misrepresentation of ligand-protein contacts
Solution Approach 1:
The patent introduces an intermediate step between docking and activity prediction: a binding mode prediction model that evaluates and selects the most accurate binding modes from docking outputs. This intermediary filters out incorrect poses before they can corrupt the activity prediction training data, thereby resolving the contradiction between obtaining binding mode predictions and ensuring reliable ligand-protein contact representation.
Solution Approach 2:
The binding mode prediction model performs preliminary evaluation and selection of binding modes before the activity prediction model processes the data. By pre-filtering and pre-ranking binding modes based on their accuracy, the system ensures that only high-quality binding mode data is used for training activity predictions, thus preventing the propagation of errors from randomly distributed incorrect poses.
2Measurement precision
If deep learning neural networks are combined with structural data, then activity prediction performance can be improved, but dataset bias occurs making protein-related features irrelevant
Solution Approach 1:
The patent applies local quality by focusing the neural network's attention on specific, relevant local features within the binding mode data. Rather than using all structural data equally, the system identifies and weights important local interactions (such as key ligand-protein contacts) more heavily, making the model adapt to the specific task of activity prediction while filtering out irrelevant protein features through the binding mode selection mechanism.
3Quantity of substance
If all generated docked structures are used for training, then more training data is available, but incorrect poses lead to misrepresented contacts and inaccurate predictions
Solution Approach 1:
The patent applies partial action by selectively using only a subset of the generated docked structures for training - specifically, only those binding modes that are predicted to be accurate by the binding mode prediction model. Rather than using all available docked structures (excessive action), the system filters to use only the necessary high-quality portion, thereby maintaining adequate training data volume while ensuring accuracy.
Data Source
AI summary
A system and method for structure-based, small molecule activity prediction using binding mode prediction information. Binding scores between ligands and target molecules, (e.g. proteins, RNA, DNA, lipids, sugars) are first generated using molecular docking. A first machine learned deep neural network (DNN) model is developed using data representing the molecular ligand-target pair 3D structures and docking features to predict binding modes. Using transfer learning, weights of layers learned in the first machine learned model are used as weights in layers of a second machine learned DNN model used to more accurately improve the performance of activity prediction of the second machine learned model. For a target newly paired ligand-target complex, the method further implements a binding mode selector for selecting one or more particular binding poses for input to the activity prediction model for use in activity mode prediction of an activity of the target paired ligand-protein complex.