Abstract

Highly Multiplexed Diagnostics with Droplet Microfluidics Enhanced by Compressed Sensing

Pavan Kota, PhD Student, Rice University

Microfluidics can generate thousands of droplets to capture individual analytes, but usually only a few measurements can be acquired from each droplet at high throughput. Multiplexing efficiently with limited sensors is possible through compressed sensing if samples are sparse; most possible analytes must be absent from any particular sample. The authors recently developed a new compressed sensing algorithm called Sparse Poisson Recovery (SPoRe) that further exploits the Poisson statistics of microfluidic capture. Given an application-driven measurement model, SPoRe efficiently solves a maximum likelihood estimation problem to recover total analyte abundances. This work presents the first in vitro demonstration of SPoRe towards bacterial infection diagnostics with 16S droplet digital PCR (ddPCR). Five nonspecific probes assign binary barcodes to nine bacterial genera. Each droplet measurement is modeled as an OR operation among the present 16S genes’ barcodes. Although a single droplet’s contents may be ambiguous under this model, SPoRe solves for bacterial abundances by considering all droplets simultaneously. Moreover, SPoRe can pool data from multiple reactions with different subsets of probes to address limitations caused by probe cross-reactivity and fluorescence spectral overlap. SPoRe raises new possibilities in ddPCR-based diagnostics, and its modularity for nearly any measurement model enables applications beyond digital sensing.