AISMPred: A Machine Learning Approach for Predicting Anti-Inflammatory Small Molecules.

Area of Science:

• Computational chemistry and cheminformatics
• Pharmacology and drug discovery
• Machine learning in medicinal chemistry

Background:

• Inflammation is a critical biological response, but chronic inflammation drives disease.
• Peptide therapeutics offer specificity but face development challenges.
• Small molecules are promising for anti-inflammatory drug development due to stability and bioavailability.

Purpose of the Study:

• To develop and validate a computational method, AISMPred, for classifying anti-inflammatory small molecules (AISMs) and non-AISMs.
• To enhance the efficiency and reduce the cost of identifying potential anti-inflammatory drug candidates.

Main Methods:

• A dataset of 1750 AISMs and non-AISMs was curated with IC50 values from PubChem.
• Molecular descriptors were computed using PaDEL and Mordred, then combined into a hybrid feature set.
• Support Vector Classifier with L1 regularization (SVC-L1) was used for feature selection, and five ML classifiers (RF, ET, KNN, LR, Ensemble) were trained.

Main Results:

• 15 Machine Learning (ML) models were developed using 2D, fingerprint (FP), and hybrid feature sets.
• The Extra Trees (ET) model utilizing hybrid features achieved the highest performance.
• The top-performing ET model demonstrated 92% accuracy and an AUC of 0.97 on an independent test set.

Conclusions:

• AISMPred offers an effective computational strategy for screening anti-inflammatory small molecules.
• This method has the potential to significantly impact and accelerate drug discovery and design pipelines.
• The study highlights the utility of ML in identifying novel therapeutic agents for inflammatory conditions.