MLDAAPP: Machine Learning Data Acquisition for Assessing Population Phenotypes

Abstract

AbstractCollecting phenotypic data from many individuals is critical to numerous biological disciplines. Yet, organismal phenotypic or trait data are still often collected manually, limiting the scale of data collection, precluding reproducible workflows, and creating the potential for human bias. Computer vision could largely ameliorate these issues, but currently available packages only operate with specific inputs and hence are not scalable or accessible for many biologists. We present Machine Learning Data Acquisition for Assessing Population Phenotypes (MLDAAPP), a package of tools for collecting phenotypic data from groups of individuals. We demonstrate that MLDAAPP is both accurate and uniquely effective at measuring phenotypes in challenging conditions - particularly images and videos of varying quality derived from both lab and field environments. Employing MLDAAPP solves key issues of reproducibility, increases both the scale and scope of data generation, and reduces the potential for human bias.