Functional microRNA-targeting drug discovery by graph-based deep learning.

Area of Science:

• Oncology
• Bioinformatics
• Molecular Biology

Background:

• MicroRNAs (miRNAs) are crucial in cancer development, but targeting them with small molecules is difficult.
• Specific miRNAs, like miR-21, are implicated as drivers in various cancers, including breast cancer.

Purpose of the Study:

• To introduce RiboStrike, a novel deep-learning framework for identifying small molecules against specific microRNAs.
• To validate RiboStrike's efficacy by targeting miR-21 in breast cancer models.

Main Methods:

• Screened nine million molecules using RiboStrike, incorporating counter-screens for selectivity against miR-122 and DICER.
• Utilized auxiliary models for toxicity assessment and candidate ranking.
• Validated anti-miR-21 activity and target selectivity through reporter assays, RNA sequencing, and microRNA profiling.

Main Results:

• Identified eight candidate molecules, with three demonstrating significant anti-miR-21 activity.
• The top candidate compound effectively reduced lung metastases in a preclinical breast cancer xenograft mouse model.
• Confirmed target selectivity of the identified compounds.

Conclusions:

• RiboStrike is a powerful deep-learning framework capable of nominating effective small molecules targeting microRNA activity in cancer.
• The identified compounds show promise for therapeutic intervention in miR-21-driven cancers, particularly breast cancer metastasis.