Chemistry-Informed Machine Learning Enables Discovery of DNA-Stabilized Silver Nanoclusters with Near-Infrared...

Area of Science:

• Materials Science
• Nanotechnology
• Biochemistry

Background:

• DNA-stabilized silver nanoclusters (Ag-DNAs) exhibit tunable fluorescence properties based on their nucleobase sequence.
• Developing Ag-DNAs with specific optical properties, especially near-infrared (NIR) luminescence (>800 nm), is challenging due to the vast DNA sequence space.
• NIR-emissive Ag-DNAs are crucial for deep-tissue bioimaging applications within the tissue transparency windows.

Purpose of the Study:

• To develop a predictive design method for creating near-infrared emissive Ag-DNAs.
• To identify key DNA sequence features that dictate Ag-DNA optical properties across the entire emission spectrum.
• To enhance the discovery rate of Ag-DNAs with bright NIR luminescence.

Main Methods:

• Combined high-throughput experimentation with machine learning models.
• Integrated fundamental information from Ag-DNA crystal structures into the design process.
• Utilized a machine learning approach trained on DNA sequence motifs predictive of Ag-DNA color.

Main Results:

• Distilled salient DNA sequence features that predict Ag-DNA color.
• Achieved a 12.3-fold increase in design success for NIR-emissive Ag-DNAs compared to existing data.
• Nearly doubled the number of known Ag-DNAs exhibiting bright NIR luminescence above 800 nm.

Conclusions:

• Incorporating structure-property relationships into machine learning significantly enhances materials design.
• The developed method effectively addresses challenges posed by sparse and imbalanced training data in materials discovery.
• This work advances the design and application of Ag-DNAs, particularly for bioimaging.