Ambrose J. Carr

Single Cell Data Modeling and Engineering

mail@ambrosejcarr.com

At The Chan Zuckerberg Initiative, The Broad Institute, and The Human Cell Atlas

ABOUT

As a Computational Biologist with the Chan Zuckerberg Initiative, my goal is to foster community-driven software projects that enable rapid development of promising biological technologies. As a result, my work sits at the intersection of computational biology, data science, and software engineering. Recently, I find myself building data-driven models to quantify and correct technical biases and variances introduced by single cell approaches. These models are woven into modular and cloud-scalable frameworks that produce high quality, comparable data atlases that we use to reason about variation in human development and in diseases like cancer.

In my spare time, I am an avid sport climber and cyclist.

EDUCATION

Columbia University
Biological Sciences and Systems Biology
Doctor of Philosophy - JAN 2018
Master of Philosophy - JUNE 2014
Master of Arts - OCTOBER 2013

Princeton University
Molecular Biology
Bachelor of Arts - JUNE 2009
Certificate in Neuroscience - JUNE 2009



RESEARCH


Computational Biologist
The Chan Zuckerberg Initiative

Spatial transcriptomics approaches are a group of emerging technologies that enable the expression of hundreds to thousands of genes to be measured in individual cells while retaining information about their local environment and cell-cell interactions. These data provide the clearest window into the effects of the tissue microenvironment by allowing researchers to directly observe the relative abundances of each cell population, enabling them to infer the effect that each population of cells has on each other.

However, the rapid evolution of spatial technology has caused each approach to generate different data formats and processing tools. This has enabled agile technology development, but has fragmented the data processing ecosystem, making data and tool sharing difficult and slowing its adoption by biologists and computational researchers.

With the Chan Zuckerberg Initiative, I contribute to the starfish project, which is building community consensus on standard data types for spatial transcriptomics technologies. In addition, it provides a common image processing toolkit to enable construction of cloud-scalable data processing pipelines for any spatial single-cell transcriptomics technology. Combined with the SpaceTx consortium, which brings together the community of technology developers driving this technology forward and the Human Cell Atlas Data Coordination Platform, which will enable large-scale open sharing of these data, Starfish is well poised to faciliate widespread use and adoption of spatial transcriptomics approaches.

APR 2018 - PRESENT

Visiting Scientist
The Data Sciences Platform
The Broad Institute

With the Broad Institute and the Human Cell Atlas (HCA) Mint team, I am building the data processing framework for the HCA. We are constructing cloud-scalable workflows to process single cell data by combining workflow design language (WDL), which enable code execution on arbitrary cloud architectures (AWS, Google Cloud Platform, Microsoft Azure) with data driven benchmarking based on cutting-edge metric frameworks.

Together, these approaches are enabling us to construct efficient, portable analysis tools that are expected to be used to process data at petabyte scales, which will allow new categories of biological questions to be asked and answered.

AUG 2017 - PRESENT

Graduate Student
Pe'er Laboratory of Computational Systems Biology
Columbia University

In the Pe'er lab, I built a proof-of-concept breast cancer immune cell atlas. To enable this work, we needed to account for technical errors and biases in 3' sequencing platforms. Unchecked, these errors allow ambient RNA contamination to masquerade as cells, allow molecules to switch between cells of different types, and allow molecules to multiply through introduction of barcode errors. We wrote methods to account for the most significant error types, as well as to resolve multiple alignments, which were combined into SEQC, a modular and cloud scalable package capable of processing data from each 3' sequencing platform.

Because data from each patient is influenced by a different set of biological and technical biases, identification of shared cell states required a method that corrects for technical effects but retains biological diversity. To separate these effects, we wrote BISCUIT, an algorithm that iteratively normalizes and clusters. BISCUIT enables cell-type specific normalization that adjusts for technical effects by fitting clusters using both mean and covariance parameters.

By applying BISCUIT and SEQC to InDrops 3' RNA sequencing data, we were able to construct an immune atlas that identified 83 technically robust and biologically explicable cell states. The identified states describe immune cells experiencing a diverse panel of environmental signal, with expression patterns that predict divergent responses to checkpoint blockade immunotherapies. The success of this project has prompted us to assay larger cohorts with more targeted questions about treatment outcomes.

AUGUST 2011 - JAN 2018

Research Assistant
Leibowitz Lab of Behavioral Neurobiology
Rockefeller University

In the Leibowitz lab, I endeavored to answer two questions. First, how do low doses of nicotine snowball into full-blown addiction? To answer this question, I created a model of nicotine self-administration. Rats trained on this model were tested for modified neurotransmitter production using qRT-PCR and immunofluorescence, which identified changes in hypothalamic and arcuate neurotransmitter production as the level of addiction grew stronger.

Second, dietary fat is well known to influence neurological cognates of hunger and food seeking behaviors, but no one had asked how different types of fats exert these effects. By both creating modified diets and directly injecting different fatty acids into the brain, I discovered that long, saturated fatty acids decreased the relative production of orexin in the hypothalamus, a response associated with satiety. This suggests that longer-chain fats may be more filling, relative to their caloric value.

JUNE 2009 - AUGUST 2011

Undergraduate Research Assistant
Hoebel Lab of Behavioral Neurobiology
Princeton University

In the Hoebel lab I focused on characterizing the neurological systems required to initiate and perpetuate alcoholism. To accomplish this, I investigated the roles of several neuronal systems known to act in other addictive disorders. I first identified that opioid-expressing neurons in the hypothalamus stimulate ethanol intake. Second, I determined that they do so through a positive feedback loop with dopamine neurons in the striatum. Finally, I determined that the baseline expression of these neuropeptides can be used to predict an animal's likelihood to develop alcohol addiction, and that the levels of these peptides correlate with the average amount of alcohol consumption. Together, these discoveries illuminate a neurological system that facilitates alcohol addiction, which might be useful in the future to identify a person's susceptibility to alcoholism.

JUNE 2007 - JUNE 2009



WORK

Vice President, Science
InSITE
Through half-year partnerships, InSITE matches teams of graduate students with early stage start-up companies to assist with marketing and brand development, product development, consumer research, and pricing strategy, among many other services.

AUGUST 2012 - SEPTEMBER 2015

Consultant
Valamont
I leveraged broad experience in biology and technology development to carry out a market analyses of the DNA sequencing space for a large, multinational diagnostics company

JUNE 2014 - JUNE 2015



AWARDS

HHMI International Pre-doctoral Fellow
Howard Hughes Medical Institute
Awarded to pursue computational approaches based on single-cell sequencing to improve cancer diagnosis and treatment

AUGUST 2014 - AUGUST 2016

Columbia Faculty Fellow
Columbia University Biological Sciences
Highest entrance award for Columbia Ph.D. students, confers full funding for 5 years of graduate study

AUGUST 2011 - AUGUST 2016

Dataminr Prize for Best Poster
New York Academy of Sciences 10th Annual Machine Learning Symposium
For "Dirichlet Process Mixture Model for Correcting Technical Variation in Single-Cell Gene Expression Data"

MARCH 2016

Thomas Hunt Morgan Prize for Best Poster
Columbia University Biological Sciences Biannual Symposium
For "A highly scalable platform for single-cell RNA-seq"

OCTOBER 2014

Best Poster
Columbia MAGNet symposium
For "A highly scalable platform for single-cell RNA-seq"

JULY 2014

HHMI Princeton Undergraduate Research Grant Recipient
Howard Hughes Medical Institute
Grant to pursue neurobiological research on alchol addiction

JULY 2008

Mellon Foundation Summer Research Grant Recipient
Princeton University & The Mellon Foundation
Grant to pursue neurobiological research on audio-visual stimulus responses

JULY 2006



PUBLICATIONS

Single-Cell Immune Map of Breast Carcinoma Reveals Diverse Phenotypic States Driven by the Tumor Microenvironment
Azizi E*, Carr AJ*, Plitas G*, Cornish AE*, Konopacki C, Prabhakaran S, Nainys J, Wu K, Kiseliovas V, Setty M, Choi K, Dao P, Mazutis L, Rudensky AY, Pe'er D
Biorxiv, 2017

MAGIC: A diffusion-based imputation method reveals gene-gene interactions in single-cell RNA-sequencing data
van Dijk D, Nainys J, Sharma R, Kathail P, Carr AJ, Moon KR, Mazutis L, Wolf G, Krishnaswamy S, Pe'er D
Biorxiv, 2017.

Measuring Signaling and RNA-Seq in the Same Cell Links Gene Expression to Dynamic Patterns of NF-κB Activation
Lane K, Van Valen D, DeFelice MM, Macklin DN, Kudo T, Jaimovich A, Carr AJ, Meyer T, Pe'er D, Boutet SC, Covert MW
Cell Syst. 26;4(4):458-469

Tensor Decomposition for Single-cell RNA-seq Data
Choi, K*, Carr AJ*, Prabhakaran, S, Pe'er D
30th Conference on Neural Information Processing Systems (NIPS), 2016, Workshop on Practical Bayesian Nonparametrics

Dirichlet Process Mixture Model for Correcting Technical Variation in Single-Cell Gene Expression Data
Prabhakaran S*, Azizi E*, Carr AJ, Pe'er D
Journal of Machine Learning Research, W&CP (ICML). 48, 2016

Scalable microfluidics for single-cell RNA printing and sequencing.
Bose S, Wan Z, Carr AJ, Rizvi AH, Vieira G, Pe'er D, Sims PA.
Genome Biol. 2015 Jun 6;16:120

Predictors of ethanol consumption in adult Sprague-Dawley rats: relation to hypothalamic peptides that stimulate ethanol intake.
Karatayev O, Barson JR, Carr AJ, Baylan J, Chen YW, Leibowitz SF
ALCOHOL. 2010 JUN;44(4):323-34

Opioids in the hypothalamic paraventricular nucleus stimulate ethanol intake.
Barson JR, Carr AJ, Soun JE, Sobhani NC, Rada P, Leibowitz SF, Hoebel BG
ALCOHOL CLIN EXP RES. 2010 Feb;34(2):214-22

Opioids in the nucleus accumbens stimulate ethanol intake.
Barson JR, Carr AJ, Soun JE, Sobhani NC, Leibowitz SF, Hoebel BG
PHYSIOL BEHAV. 2009 Oct 19;98(4):453-9


SKILLS

Machine Learning


Python & R


Photoshop & Illustrator


Databases


Amazon Web Services


Google Cloud Platform


HTML, CSS & Javascript


Workflow Design Language (WDL)



CONTACT

Copyright © Ambrose J. Carr 2018