Data-driven discovery of innate immunomodulators via machine learning-guided high throughput screening.

Area of Science:

• Immunology
• Computational Biology
• Drug Discovery

Background:

• The innate immune response is critical for vaccine and immunotherapy efficacy.
• Controlling innate immune signaling can mitigate adverse inflammation and boost therapeutic immune stimulation.

Purpose of the Study:

• To develop a machine learning-enabled active learning pipeline for discovering small molecule immunomodulators.
• To guide in vitro experimental screening of molecules that alter innate immune signaling pathways.

Main Methods:

• Utilized in vitro high-throughput screening (HTS) to measure modulation of NF-κB and IRF pathways.
• Employed deep representational learning, Gaussian process regression, and Bayesian optimization for predictive modeling.
• Interleaved model training with HTS for active learning-guided library traversal.

Main Results:

• Discovered novel immunomodulators by screening ~2% of a 139,998-molecule library.
• Identified molecules suppressing NF-κB activity (up to 15-fold) and elevating NF-κB (up to 5-fold) and IRF (up to 6-fold) activity.
• Validated immunomodulatory effects, with one molecule enhancing IFN-β production by 3-fold with a STING agonist.

Conclusions:

• The ML-enabled screening pipeline efficiently discovers small molecule immunomodulators.
• Novel molecules were identified with potent capacity to enhance or suppress innate immune signaling.
• This approach can significantly improve prophylactic vaccination and immunotherapy development.