Diving deep into the dependencies within data

Analyzing how variables within massive datasets rise and fall in relation to each other can reveal hidden structures in processes like gene expression. The methods for studying such dependencies, however, often focus on linear relationships or don't scale to millions of samples and tens of thousands of variables.

In response, Adit Radhakrishnan, Yajit Jain, Caroline Uhler, and Eric Lander have developed the InterDependence Score (IDS), a computationally light algorithm that uncovers relationships in large datasets that elude other correlation measures, such as complex expression patterns underlying cellular programs and states. It also, they found, provides fundamental insights into neural networks' predictive capabilities.

Radhakrishnan is a former Eric and Wendy Schmidt Center postdoctoral fellow and current assistant professor at MIT; Jain is a former Schmidt Center postdoctoral fellow and current Senior ML Scientist at the Lander Lab; Uhler is the director of the Schmidt Center and the Andrew and Erna Viterbi Professor of Engineering at MIT; and Lander is the founding director and a core institute member of the Broad Institute.

Learn more about their work in PNAS.

Adapted from an update written by Broad Communications.