Zymergen Inc.
IDENTIFYING ORGANISMS FOR PRODUCTION USING UNSUPERVISED PARAMETER LEARNING FOR OUTLIER DETECTION

Abstract:

Systems, methods and computer-readable media are provided for identifying organisms for production. The identification is based upon determining one or more outlier detection parameters for identifying outliers (e.g., outlier wells, strains, plates holding organisms) from a data set of organism performance metrics. A prediction engine may identify one or more candidate outliers based upon a first set of outlier detection parameters (e.g., outlier detection threshold), and determine probability metrics that represent likelihoods that candidate outliers belong to an outlier class. Based on those metrics, some of the outliers may be excluded from consideration in predicting organism performance for the purpose of selecting organisms for production.