DeepHDAC3i: Leveraging an Interpretable Deep Learning-Based Framework for the Accelerated Discovery of HDAC3...

Area of Science:

• Epigenetics and computational drug discovery.

Background:

• Epigenetic modifications regulate gene expression without altering DNA sequence.
• Histone deacetylase (HDAC) inhibitors are used in cancer therapy, but lack specificity.
• Selective HDAC inhibitors are needed to improve cancer treatment outcomes.

Purpose of the Study:

• To develop a novel, interpretable deep learning framework, DeepHDAC3i, for accurate in silico identification of HDAC3 inhibitors (HDAC3i).
• To utilize only SMILES notation for inhibitor identification, bypassing the need for 3D ligand structures.

Main Methods:

• Employed five molecular encoding methods (CDKExt, KR, KRC, Pubchem, RDKit) to extract multi-view features from SMILES notation.
• Used elastic net for optimal feature selection and a 1D convolutional neural network (1D-CNN) for model construction.
• Leveraged Shapley Additive exPlanation for feature interpretability.

Main Results:

• DeepHDAC3i achieved high performance on an independent test set with accuracy of 0.965, MCC of 0.930, and AUC of 0.985.
• Demonstrated superior performance compared to conventional machine learning and deep learning models, with significant improvements in accuracy, F1, AUC, and MCC.
• The framework provides interpretable insights into key features driving HDAC3i identification.

Conclusions:

• DeepHDAC3i is a highly accurate and interpretable deep learning tool for identifying HDAC3 inhibitors.
• The framework offers a cost-efficient and rapid approach for drug discovery in cancer therapy.
• DeepHDAC3i outperforms existing methods, presenting a valuable tool for precise HDAC3i identification.