Designing sensitive viral diagnostics with machine learning.

Area of Science:

• Molecular Biology
• Bioinformatics
• Virology

Background:

• Traditional viral diagnostic design relies on conserved regions, limiting detection of viral variants.
• Optimizing diagnostic sensitivity against the full spectrum of viral genomic variation is a significant challenge.

Purpose of the Study:

• To develop an automated system for designing highly effective nucleic acid-based viral diagnostics.
• To improve sensitivity and specificity in detecting diverse viral strains and lineages.

Main Methods:

• Screened 19,209 diagnostic-target pairs, focusing on CRISPR-based diagnostics.
• Trained a deep neural network to predict diagnostic readout and used combinatorial optimization.
• Introduced the Activity-informed Design with All-inclusive Patrolling of Targets (ADAPT) system for automated design.

Main Results:

• Designed diagnostics for 1,933 vertebrate-infecting viral species within 24 hours.
• ADAPT designs demonstrated lineage-level sensitivity and specificity.
• Achieved lower limits of detection across viral variation compared to standard methods.

Conclusions:

• The ADAPT system enables rapid, automated design of sensitive and specific viral diagnostics.
• This approach can facilitate a proactive resource for pathogen detection and surveillance.
• Optimizing diagnostics using AI models can overcome limitations of traditional heuristic design methods.