Exploring Anti-osteoporosis Medicinal Herbs using Cheminformatics and Deep Learning Approaches

Osteoporosis is a prevalent disease for the aged population. Chinese herbderived natural compounds have anti-osteoporosis effects. Due to the complexity of chemical ingredients and natural products, it is necessary to develop a high-throughput approach with the integration of cheminformatics and deep-learning methods to explore their mechanistic action, especially herb/drug-gene interaction networks.

Ten medicinal herbs for clinical osteoporosis treatment were selected. Chemical ingredients of the top 10 herbs were retrieved from the TCMIO database, and their predicted targets were obtained from the SEA server. Anti-osteoporosis clinical drugs and targets were collected from multidatabases. Chemical space, fingerprint similarity, and scaffold comparison of the compounds between herbs and clinical drugs were analyzed by RDKit and SKlearn. A network of herb-ingredient-target was constructed via Gephi, and GO and KEGG enrichment analyses were performed using clusterProfiler. Additionally, the bioactivity of compounds and targets was predicted by DeepScreening. Molecular docking of YYH flavonoids to HSD17B2 was accomplished by AutoDockTools.

Cheminformatics result depicts a pharmacological network consisting of 89 active components and 30 potential genes. The chemical structures of plant steroids, flavonoids, and alkaloids are key components for anti-osteoporosis effects. Moreover, bioinformatics result demonstrates that the active components of herbs mainly participate in steroid hormone biosynthesis and the TNF signaling pathway. Finally, deep-learning-based regression models were constructed to evaluate 22 anti-osteoporosis-related protein targets and predict the activity of 1350 chemical ingredients of the 10 herbs.

The combination of cheminformatics and deep-learning approaches sheds light on the exploration of medicinal herbs mechanisms, and the identification of novel and active compounds from medical herbs in complex molecular systems.