Embedded Techniques for High Content Analysis Feature Selection
Abstract
High content analysis (HCA) is a useful technique for extracting unprejudiced explanations for phenotypic responses. However, the massive number of features generated -- often exceeding the number of samples -- necessitates an intermediate feature selection step as part of an overall analytic pipeline. While typical feature selection techniques in the literature focus on more modest feature sizes p < 100, we found that our large feature sets diminished the feasibility of direct wrapper-based approaches, whereas filter-based approaches produced limited feature complementarity.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 23, 2020
- Accession Number
- AD1123159
Entities
People
- Daniel Cowan
- Guanpeng A Xu
- Heather A. Pangburn
- Patrick M. Mclendon
Organizations
- Massachusetts Institute of Technology