Improving COVID-19 Testing Efficiency using Guided Agglomerative Sampling

Fayyaz Minhas; Dimitris Grammatopoulos; Lawrence Young; Imran Amin; David Snead; Neil Anderson; Asa Ben-Hur; Nasir Rajpoot

In this work, we have developed a community laboratory testing strategy for CoV-2 based on a divide and conquer approach [5] that can reduce the number of tests required for testing a given number of samples. It can optionally be used in conjunction with a belief assignment method such as a machine learning prediction model or with guidance from a human expert to improve testing . CC-BY-NC-ND 4.0 International license author/funder. It is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/2020.04.13.039792 doi: bioRxiv preprint efficiency even further. In terms of machine learning, the proposed scheme can be adapted for use to work together with a machine learning model which generates a ranked list of likely positive samples which can be tested individually followed by agglomerative testing of the remaining samples. Additionally, in the absence of a predictive model or another means of belief assignment, the proposed scheme can use feature-based unsupervised clustering to reduce the number of required tests building on the assumption that the test results of individuals are not independent of each other. We have opted to share the proposed method in the hope that it can be beneficial to large-scale CoV-2 testing and the management of patients with COVID-19. Laboratory trials with the proposed sampling technique are being considered at the University of Warwick to study the impact of the proposed strategy on accuracy of existing testing methods and understand practical limitations.