Predicting drug-induced QT prolongation effects using multi-view learning.

Area of Science:

• Pharmacology
• Computational Biology
• Machine Learning

Background:

• Drug-induced QT prolongation is a life-threatening adverse drug effect.
• Predicting this effect early in drug development is crucial but challenging due to limited and noisy data.
• Multi-view learning (MVL) is effective for complex data with limited labels.

Purpose of the Study:

• To develop a novel multi-view learning algorithm for predicting drug-induced QT prolongation.
• To improve computational efficiency and interpretability in QT prolongation prediction.
• To address the limitations of existing l2-norm co-regularized MVL methods.

Main Methods:

• Proposed an l1-norm co-regularized MVL algorithm for QT prolongation prediction.
• Reformulated the objective function for analytic gradient derivation.
• Optimized mapping functions on all views simultaneously for enhanced computational efficiency.
• Enforced sparsity in learned mapping functions for improved interpretability.

Main Results:

• The proposed l1-norm co-regularized MVL method achieved 3-4 times higher computational efficiency.
• The new algorithm significantly outperformed previous MVL and single-view learning methods.
• Experimental comparisons demonstrated superior performance and efficiency.

Conclusions:

• The l1-norm co-regularized MVL algorithm is an effective and efficient method for predicting drug-induced QT prolongation.
• The sparsity-inducing nature of l1-norm leads to more interpretable results.
• This approach offers a promising advancement for early-stage drug safety assessment.