Postdoc opportunity! to work with me here at Columbia! on Bayesian workflow! for contamination models! With some wonderful collaborators!!

Laboratory assays are central to much of biomedical research. My colleagues and I recently received a research grant to do better assays using Bayesian inference. Beyond the usual challenges of fitting nonlinear hierarchical models to real data that can sometimes be above or below detection limits, the project involves several special challenges:

– We estimate concentrations of a compound using calibration assays. But when samples are contaminated, which can happen all the time in the real world (for example, you gather dust samples from the kitchen, and they are contaminated with sugar), the calibration curve can change.

– More generally, we want to simultaneously classify a sample as contaminated or not, and estimate its concentration under some model of possible contamination, while recognizing that our contamination model could be way off.

– Experimental design. This includes designing protocols for spiking some dilutions with contaminants to estimate the contaminated calibration curves, and also, once everything is working, designing assays to get more information out of each plate. We believe that current designs are inefficient, devoting too much of the assay space to estimating the calibration curve and not enough for each unknown sample.

– We want to build a robust implementation so that practitioners can use our new approach (whatever exactly it is) with confidence, in place of what is currently given out by the standard software that comes with the assay machines.

In short, workflow. We’re not just analyzing a dataset, we’re proposing to create the new default, and then assess where it makes a difference in real-world studies.

And to do this we’re hiring a postdoc. Someone who wants to build models in Stan, and who is also interested in the super-important but often underemphasized area of experimental design, and who is also interested in building a reliable tool that includes diagnostics and warnings when it’s not working well. We’re aiming for an end-to-end statistics project here, with immediate applications to public health, and my colleagues and I are really stoked to do it.

You, the postdoc, will be at the center of this. In addition to working with us on this project, you’ll be part of the active research communities in Columbia’s statistics and biostatistics departments, with lots of opportunity for collaborations on theory and applications alike. Really a wonderful opportunity.

This exciting project is a collaboration between me, Prof. Qixuan Chen in the biostatistics department, and Prof. Matt Perzanowski in the department of environmental health.

All of us have collaborated before, and we each bring strengths to the project:

– I have experience in Bayesian modeling and computing and am particularly interested in methods for diagnosing and expanding models that do not fit the data.

– Qixuan has worked on a wide variety of problems involving Bayesian latent-parameter modeling in the biomedical sciences.

– Matt operates a lab performing bioassays for public health studies, which allows us to develop and evaluate our research on an ongoing data stream and have real-world impact.

The official job announcement is here, and some background is in these papers from 2004 and from 2007.

Interested candidates should submit a detailed CV, a cover letter outlining research interests and career goals, and contact information for three references to Dr. Qixuan Chen at qc2138@cumc.columbia.edu.