Industrial Consultancy & Sponsored Research (IC&SR) , IIT Madras
A Method And System For Predicting Mutation-induced Binding Affinity Changes In Membrane Protein Complexes
Technology Category/Market
Category: Artificial Intelligence (AI) and Machine Learning/ Drugs and Pharmaceutical Engineering
Industry Classification:
Pharmaceutical and Drug Development; Biotechnology and Genetic Engineering
Applications:
Identifying the effects of mutations on membrane proteins; Critical to drug-target interactions; Precision Medicine and Personalized Healthcare; Study of disease pathogenesis; Structural Biology; Functional Analysis of Membrane Proteins; High-Throughput Screening and Computational Biology
Market report:
The global protein engineering market was valued at USD 4.35 billion in 2024 and is projected to grow to USD 20.86 billion by 2034 with a CAGR of 16.97%
Problem Statement
- Membrane proteins are key targets in drug design, influencing therapeutic interventions for diseases like cancer and cardiovascular conditions.
- Conventionally used experimental methods such as Surface Plasmon Resonance (SPR) and Isothermal Titration Calorimetry (ITC) are resource-intensive and time-consuming.
- Further, computational methods, while efficient, lack specificity for membrane proteins. Current machine learning and deep learning methods fail to specifically predict mutation-induced binding affinity changes in membrane protein complexes.
- There is a need for an improved method to integrate structural and sequence-based features with advanced models (e.g., Gradient Boosting Regressor (GBR)) for accurate, efficient binding affinity predictions.
Technology
- The technology predicts mutation-induced binding affinity changes (ΔΔG) in membrane protein (MP) complexes, aiding in drug design, therapeutic interventions, and understanding mutation impacts on MP stability and interactions.
- Utilizes structure-based (e.g., total energy, solvent accessibility) and sequence-based (e.g., physicochemical properties, conservation score) features. Employs forward feature selection for optimization, ensuring minimal multicollinearity and maximal relevance in predictions.
- Achieves high prediction accuracy with Pearson correlation r=0.75, mean absolute error (MAE) =0.73 kcal/mol. Trained on a dataset of 770 MP mutations using Gradient Boosting Regressor and advanced bioinformatics tools.
- Effective across MP functional classes (enzymes, receptors, transporters) and mutation types. Outperforms conventional methods (e.g., mCSM, SAAMBE) with reduced MAE and higher correlation, ensuring robust predictive performance.
- Integrates a modular setup with processors, memory, and a database. Features extraction, feature selection, and prediction are automated for precise analysis, offering scalability for bioinformatics research and pharmaceutical applications.
Key Features/Value Proposition
- Enhanced Feature Integration: Combines structure-based features (e.g., total energy, inter-residue contacts) and sequence-based features (e.g., PSSM profiles, conservation score), unlike conventional methods that focus on limited aspects, ensuring a comprehensive analysis of mutation impacts.
- Superior Prediction Accuracy: Achieves Pearson correlation r=0.75 and MAE=0.73 kcal/mol, outperforming conventional methods (e.g., mCSM, SAAMBE) that show lower correlations and higher MAE (>1 kcal/mol).
- Optimized Feature Selection: Employs forward feature selection (FFS) to minimize multicollinearity and select the most relevant features, resulting in a more precise and computationally efficient prediction compared to exhaustive or manual feature selection methods.
- Versatility Across Functional Classes: Demonstrates robust performance across various MP functional classes (enzymes, receptors, transporters), with MAE ranging from 0.63 to 0.87, ensuring applicability to diverse biological systems.
- Dataset and Methodological Superiority: Trains on a high-quality dataset of single mutations with experimentally validated binding affinity data, leveraging Gradient Boosting Regressor (GBR) for capturing complex feature relationships, outperforming simpler regression models used in competing technologies.
Questions about this Technology?
Contact for Licensing
Research Lab
Prof. Michael Gromiha M
Department of Biotechnology
Intellectual Property
- IITM IDF Ref 2979
- IN 202441050394 Patent Application
Technology Readiness Level
TRL 4
Technology Validated in Lab