Disease-related compound identification based on deeping learning method.

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence

Background:

• Acute lung injury (ALI) is a severe respiratory condition linked to COVID-19 pathogenesis.
• Traditional Chinese Medicine (TCM) shows promise in ALI intervention.
• Network pharmacology is crucial for understanding disease-gene-target-drug interactions.

Purpose of the Study:

• To develop a deep learning model for accurate construction of "disease-gene-target-drug" networks.
• To identify active compounds in TCM for ALI treatment.
• To evaluate the performance of a novel deep learning algorithm against existing methods.

Main Methods:

• Utilized deep learning algorithm, specifically the Forest graph embedded deep feed forward network (forgeNet).
• Characterized compounds using molecular descriptors and fingerprints.
• Trained forgeNet on active and inactive compounds related to ALI target genes (REAL and SATA3).

Main Results:

• ForgeNet demonstrated superior performance compared to SVM, RF, LR, NB, XGBoost, LightGBM, and gcForest.
• The model accurately identified 19 compounds in Erhuang decoction (EhD) and Dexamethasone (DXMS).

Conclusions:

• ForgeNet offers a more accurate approach for identifying therapeutic compounds in TCM for ALI.
• This deep learning model advances network pharmacology applications in respiratory disease research.
• The findings support the potential of TCM in managing ALI and related conditions.