TimeSight: Kinetics-aware detection for digital nucleic acid amplification via spatiotemporal convolutional learning

Area of Science:

• Biotechnology
• Molecular Diagnostics
• Bioinformatics

Background:

• Digital nucleic acid amplification offers absolute quantification but current methods analyze single endpoint images, ignoring temporal data.
• Portable point-of-care testing (POCT) instruments face challenges with image quality due to simplified optics and low signal-to-noise ratios.
• Distinguishing true amplification from artifacts is difficult with traditional endpoint analysis, especially in noisy POCT environments.

Purpose of the Study:

• To develop an end-to-end spatiotemporal detection framework, TimeSight, for analyzing time-resolved fluorescence image stacks from digital recombinase polymerase amplification (dRPA) on portable POCT instruments.
• To improve the accuracy and reliability of absolute molecular quantification in resource-limited settings by leveraging temporal amplification kinetics.
• To overcome limitations of single-frame analysis in low-signal and artifact-prone conditions inherent to miniaturized diagnostic devices.

Main Methods:

• Implemented TimeSight, a framework using factorised 3D convolutions with a learned temporal collapse mechanism to jointly analyze spatial and kinetic information.
• Developed a method that bypasses the need for partition segmentation, spatial registration, and manual threshold tuning.
• Utilized time-resolved fluorescence image stacks acquired during dRPA on a portable POCT instrument.

Main Results:

• TimeSight achieved superior performance (F1: 0.968, Precision: 0.977) compared to single-frame and multi-frame baselines on a large test set.
• The framework effectively suppressed temporally invariant artifacts, maintaining performance across varying signal-to-noise ratios and artifact levels.
• Demonstrated accurate matching of reference concentrations across four orders of magnitude and agreement with commercial digital PCR for S. aureus DNA detection.
• Confirmed portable feasibility with rapid analysis (<2% parameter increase) at approximately 113 ms per image stack.

Conclusions:

• Temporal amplification kinetics can compensate for spatial information loss in miniaturized hardware, enabling robust absolute quantification.
• TimeSight provides a novel spatiotemporal detection paradigm for digital nucleic acid quantification, particularly beneficial for resource-limited settings and POCT.
• The framework lowers the barrier for accurate molecular diagnostics at the point-of-care by effectively handling low-quality imaging data.