Machine learning-based detection of adventitious microbes in T-cell therapy cultures using long-read sequencing

Area of Science:

• Microbiology
• Genomics
• Bioinformatics

Background:

• Cell therapy product safety relies on rigorous sterility testing.
• Current compendial sterility tests (USP <71>) are time-consuming, taking 7-14 days.
• Rapid detection of low-abundance microbial contaminants is crucial for cell therapy manufacturing.

Purpose of the Study:

• To develop a rapid, untargeted method for sensitive microbial contaminant detection in cell therapies.
• To utilize long-read sequencing and machine learning for sterility assessment.
• To reduce the time required for microbial testing in cell therapy production.

Main Methods:

• Developed a long-read sequencing workflow using Oxford Nanopore Technologies MinION.
• Employed 16S and 18S amplicon sequencing for microbial detection.
• Utilized metagenomic classification and an XGBoost machine learning model for contaminant identification and sterility determination.

Main Results:

• Achieved sensitive detection of microbial contaminants down to 10 colony-forming units (CFU)/mL.
• The developed pipeline accurately identified microbial species and assessed sample sterility status.
• Demonstrated the capability to detect USP <71> organisms and other microbial species.

Conclusions:

• Coupling long-read sequencing with machine learning provides a rapid and accurate method for cell therapy sterility testing.
• This novel approach significantly shortens detection time compared to traditional methods.
• The method enhances the safety and efficiency of cell therapy manufacturing, benefiting patient outcomes.