Kinney Lab Home
Page last modified 09:26, 30 Sep 2015 by jkinney Kinney Lab Home >
Table of contents
Justin B. Kinney
Assistant Professor at Cold Spring Harbor Laboratory
Sequence-function relationships; machine learning; biophysics; transcriptional regulation
My research combines theory, computation, and experiment in an effort to better understand the relationship between sequence and function in molecular biology.
Ultra-high-throughput DNA sequencing is revolutionizing the ability to measure sequence-function relationships in a wide variety of systems. In Kinney et al. (2010), an ultra-high-throughput promoter-bashing assay called ``Sort-Seq’’ was proposed and demonstrated. Sort-Seq uses flow cytometry and deep sequencing to probe the detailed biophysical mechanisms of transcriptional regulation in vivo. My current experimental work uses Sort-Seq in studies of transcriptional regulation, as well as in other areas of molecular biology.
The analysis of Sort-Seq data also presents novel challenges in machine learning, and a substantial fraction of my work is devoted to addressing these theoretical and computational problems. For instance, the need to fit models to Sort-Seq data highlights the general problem of fitting parametric models to data in the absence of a known noise model. In considering this general problem, Kinney and Atwal (2014a) showed that maximizing a quantity from information theory called “mutual information” allows one to essentially solve the maximum likelihood problem without a noise model. This has important implications for the design of Sort-Seq-like experiments in molecular biology, as well as for studies of receptive fields in neuroscience.
Fitting models to Sort-Seq data also requires the ability to estimate probability densities with high precision and explicit uncertainty. This challenge motivated the development of a new field-theoretic approach to density estimation (Kinney, 2013). This nonparametric method of density estimation has essentially no free tunable parameters, and provides an attractive alternative to more standard methods such as kernel density estimation.
I am also proud to be an affiliate member of the the bioRxiv