Smart pooling: AI-powered COVID-19 informative group testing

Abstract

Massive molecular testing for COVID-19 has been pointed out as fundamental to moderate the spread of the pandemic. Pooling methods can enhance testing efficiency, but they are viable only at low incidences of the disease. We propose Smart Pooling, a machine learning method that uses clinical and sociodemographic data from patients to increase the efficiency of informed Dorfman testing for COVID-19 by arranging samples into all-negative pools. To do this, we ran an automated method to train numerous machine learning models on a retrospective dataset from more than 8000 patients tested for SARS-CoV-2 from April to July 2020 in Bogotá, Colombia. We estimated the efficiency gains of using the predictor to support Dorfman testing by simulating the outcome of tests. We also computed the attainable efficiency gains of non-adaptive pooling schemes mathematically. Moreover, we measured the false-negative error rates in detecting the ORF1ab and N genes of the virus in RT-qPCR dilutions. Finally, we presented the efficiency gains of using our proposed pooling scheme on proof-of-concept pooled tests. We believe Smart Pooling will be efficient for optimizing massive testing of SARS-CoV-2.

Figure 1

Smart Pooling makes pooling methods efficient even at high disease prevalences. (a) In standard Dorfman’s testing methods, samples are pooled randomly. As prevalence increases, the efficiency of pooling without a priori information drops rapidly, making the strategy unviable. (b) Smart Pooling tackles this problem by thresholding samples according to the automated predicted risk probability, based on the sociodemographic characteristics. If the risk probability of a sample surpasses the defined threshold, this sample goes directly to individual testing. We arrange samples into pools with similar risk probability. (c) The Smart Pooling pipeline directly integrates machine learning with laboratory procedures.

Figure 2

Smart Pooling decreases the expected number of tests per specimen. (a) Smart Pooling’s expected number of tests per specimen compared to standard testing methods on Patient Dataset. Efficiency improves by reducing the expected number of tests per specimen. Smart Pooling achieves higher efficiencies than Dorfman’s and individual testing for prevalences of the disease of up to 50%, with a fixed pool size of 10. (b) Smart Pooling’s expected number of tests per specimen trained with the coarse metadata from the Test Center Dataset. Efficiency improves by reducing the expected number of tests per specimen. Despite not having detailed patient metadata, Smart Pooling can produce higher efficiencies than Dorfman’s pooling for the simulated prevalence of the disease of up to 25%, with a fixed pool size of 10. (c) Expected number of tests per specimen for the proof-of-concept Smart Pooling Implementation. Each point on the graph represents the performance of both methods for a day in the proof-of-concept stage. The x-axis depicts the incidence of the corresponding day. Efficiency improves by reducing the expected number of tests per specimen. Smart Pooling has a similar efficiency compared to Dorfman’s testing since the daily incidence is lower than 10%. However, for every day of the implementation, Smart Pooling has a higher efficiency than individual testing.

Figure 3

Comparison between Smart Pooling and Random ordering. Left: samples arranged by Smart Pooling. Right: random ordering of samples. Note that Smart Pooling groups positive samples generating long groups of negative samples, reducing the number of times it is necessary to perform the second stage of Dorfman’s testing, increasing the efficiency.

Figure 4

Efficiency of two-stage Dorfman’s pooling as a function of prevalence p and pool size c.

Figure 5

Optimal pooling strategies given a maximum pool size of (a) 10 and (b) 5.